VictoriaMetrics/lib/promauth/config.go

1004 lines
30 KiB
Go
Raw Normal View History

package promauth
import (
lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk (#5725) * lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk Added a custom `http.RoundTripper` implementation which checks for root CA content changes and updates `tls.Config` used by `http.RoundTripper` after detecting CA change. Client certificate changes are not tracked by this implementation since `tls.Config` already supports passing certificate dynamically by overriding `tls.Config.GetClientCertificate`. This change implements dynamic reload of root CA only for streaming client used for scraping. Blocking client (`fasthttp.HostClient`) does not support using custom transport so can't use this implementation. See: https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5526 Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: update NewRoundTripper API Update API to allow user to update only parameters required for transport. Add warning log when reloading Root CA failed. Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: fix mutex acquire logic Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: replace RWMutex with regular mutex to simplify the code - remove additional mutex used for getRootCABytes - require callee to use mutex - replace RWMutex with regular mutex Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: refactor - hold the mutex lock to avoid round tripper being re-created twice - move recreation logic into separate func to simplify the code Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> --------- Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> Co-authored-by: Nikolay <nik@victoriametrics.com>
2024-04-03 10:01:43 +02:00
"bytes"
"context"
"crypto/tls"
"crypto/x509"
"encoding/base64"
"fmt"
"net/http"
"net/url"
"strings"
"sync"
"github.com/cespare/xxhash/v2"
"golang.org/x/oauth2"
"golang.org/x/oauth2/clientcredentials"
lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk (#5725) * lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk Added a custom `http.RoundTripper` implementation which checks for root CA content changes and updates `tls.Config` used by `http.RoundTripper` after detecting CA change. Client certificate changes are not tracked by this implementation since `tls.Config` already supports passing certificate dynamically by overriding `tls.Config.GetClientCertificate`. This change implements dynamic reload of root CA only for streaming client used for scraping. Blocking client (`fasthttp.HostClient`) does not support using custom transport so can't use this implementation. See: https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5526 Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: update NewRoundTripper API Update API to allow user to update only parameters required for transport. Add warning log when reloading Root CA failed. Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: fix mutex acquire logic Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: replace RWMutex with regular mutex to simplify the code - remove additional mutex used for getRootCABytes - require callee to use mutex - replace RWMutex with regular mutex Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: refactor - hold the mutex lock to avoid round tripper being re-created twice - move recreation logic into separate func to simplify the code Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> --------- Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> Co-authored-by: Nikolay <nik@victoriametrics.com>
2024-04-03 10:01:43 +02:00
"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/fs/fscore"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/netutil"
)
// Secret represents a string secret such as password or auth token.
//
// It is marshaled to "<secret>" string in yaml.
//
// This is needed for hiding secret strings in /config page output.
// See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/1764
type Secret struct {
S string
}
// NewSecret returns new secret for s.
func NewSecret(s string) *Secret {
if s == "" {
return nil
}
return &Secret{
S: s,
}
}
// MarshalYAML implements yaml.Marshaler interface.
//
// It substitutes the secret with "<secret>" string.
func (s *Secret) MarshalYAML() (interface{}, error) {
return "<secret>", nil
}
// UnmarshalYAML implements yaml.Unmarshaler interface.
func (s *Secret) UnmarshalYAML(f func(interface{}) error) error {
var secret string
if err := f(&secret); err != nil {
return fmt.Errorf("cannot parse secret: %w", err)
}
s.S = secret
return nil
}
// String returns the secret in plaintext.
func (s *Secret) String() string {
if s == nil {
return ""
}
return s.S
}
// TLSConfig represents TLS config.
//
// See https://prometheus.io/docs/prometheus/latest/configuration/configuration/#tls_config
type TLSConfig struct {
CA string `yaml:"ca,omitempty"`
CAFile string `yaml:"ca_file,omitempty"`
Cert string `yaml:"cert,omitempty"`
CertFile string `yaml:"cert_file,omitempty"`
Key string `yaml:"key,omitempty"`
KeyFile string `yaml:"key_file,omitempty"`
ServerName string `yaml:"server_name,omitempty"`
InsecureSkipVerify bool `yaml:"insecure_skip_verify,omitempty"`
MinVersion string `yaml:"min_version,omitempty"`
// Do not define MaxVersion field (max_version), since this has no sense from security PoV.
// This can only result in lower security level if improperly set.
}
// Authorization represents generic authorization config.
//
// See https://prometheus.io/docs/prometheus/latest/configuration/configuration/
type Authorization struct {
Type string `yaml:"type,omitempty"`
Credentials *Secret `yaml:"credentials,omitempty"`
CredentialsFile string `yaml:"credentials_file,omitempty"`
}
// BasicAuthConfig represents basic auth config.
type BasicAuthConfig struct {
Username string `yaml:"username,omitempty"`
UsernameFile string `yaml:"username_file,omitempty"`
Password *Secret `yaml:"password,omitempty"`
PasswordFile string `yaml:"password_file,omitempty"`
}
// HTTPClientConfig represents http client config.
type HTTPClientConfig struct {
Authorization *Authorization `yaml:"authorization,omitempty"`
BasicAuth *BasicAuthConfig `yaml:"basic_auth,omitempty"`
BearerToken *Secret `yaml:"bearer_token,omitempty"`
BearerTokenFile string `yaml:"bearer_token_file,omitempty"`
OAuth2 *OAuth2Config `yaml:"oauth2,omitempty"`
TLSConfig *TLSConfig `yaml:"tls_config,omitempty"`
// Headers contains optional HTTP headers, which must be sent in the request to the server
Headers []string `yaml:"headers,omitempty"`
// FollowRedirects specifies whether the client should follow HTTP 3xx redirects.
FollowRedirects *bool `yaml:"follow_redirects,omitempty"`
// Do not support enable_http2 option because of the following reasons:
//
// - http2 is used very rarely comparing to http for Prometheus metrics exposition and service discovery
// - http2 is much harder to debug than http
// - http2 has very bad security record because of its complexity - see https://portswigger.net/research/http2
//
// VictoriaMetrics components are compiled with nethttpomithttp2 tag because of these issues.
//
// EnableHTTP2 bool
}
// ProxyClientConfig represents proxy client config.
type ProxyClientConfig struct {
Authorization *Authorization `yaml:"proxy_authorization,omitempty"`
BasicAuth *BasicAuthConfig `yaml:"proxy_basic_auth,omitempty"`
BearerToken *Secret `yaml:"proxy_bearer_token,omitempty"`
BearerTokenFile string `yaml:"proxy_bearer_token_file,omitempty"`
OAuth2 *OAuth2Config `yaml:"proxy_oauth2,omitempty"`
TLSConfig *TLSConfig `yaml:"proxy_tls_config,omitempty"`
// Headers contains optional HTTP headers, which must be sent in the request to the proxy
Headers []string `yaml:"proxy_headers,omitempty"`
}
// OAuth2Config represent OAuth2 configuration
type OAuth2Config struct {
ClientID string `yaml:"client_id"`
ClientSecret *Secret `yaml:"client_secret,omitempty"`
ClientSecretFile string `yaml:"client_secret_file,omitempty"`
Scopes []string `yaml:"scopes,omitempty"`
TokenURL string `yaml:"token_url"`
EndpointParams map[string]string `yaml:"endpoint_params,omitempty"`
TLSConfig *TLSConfig `yaml:"tls_config,omitempty"`
ProxyURL string `yaml:"proxy_url,omitempty"`
}
func (o *OAuth2Config) validate() error {
if o.ClientID == "" {
return fmt.Errorf("client_id cannot be empty")
}
if o.ClientSecret == nil && o.ClientSecretFile == "" {
return fmt.Errorf("ClientSecret or ClientSecretFile must be set")
}
if o.ClientSecret != nil && o.ClientSecretFile != "" {
return fmt.Errorf("ClientSecret and ClientSecretFile cannot be set simultaneously")
}
if o.TokenURL == "" {
return fmt.Errorf("token_url cannot be empty")
}
return nil
}
type oauth2ConfigInternal struct {
mu sync.Mutex
cfg *clientcredentials.Config
clientSecretFile string
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
// ac contains auth config needed for initializing tls config
ac *Config
proxyURL string
proxyURLFunc func(*http.Request) (*url.URL, error)
ctx context.Context
tokenSource oauth2.TokenSource
}
func (oi *oauth2ConfigInternal) String() string {
return fmt.Sprintf("clientID=%q, clientSecret=%q, clientSecretFile=%q, scopes=%q, endpointParams=%q, tokenURL=%q, proxyURL=%q, tlsConfig={%s}",
oi.cfg.ClientID, oi.cfg.ClientSecret, oi.clientSecretFile, oi.cfg.Scopes, oi.cfg.EndpointParams, oi.cfg.TokenURL, oi.proxyURL, oi.ac.String())
}
func newOAuth2ConfigInternal(baseDir string, o *OAuth2Config) (*oauth2ConfigInternal, error) {
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
if err := o.validate(); err != nil {
return nil, err
}
oi := &oauth2ConfigInternal{
cfg: &clientcredentials.Config{
ClientID: o.ClientID,
ClientSecret: o.ClientSecret.String(),
TokenURL: o.TokenURL,
Scopes: o.Scopes,
EndpointParams: urlValuesFromMap(o.EndpointParams),
},
}
if o.ClientSecretFile != "" {
oi.clientSecretFile = fscore.GetFilepath(baseDir, o.ClientSecretFile)
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
// There is no need in reading oi.clientSecretFile now, since it may be missing right now.
// It is read later before performing oauth2 request to server.
}
opts := &Options{
BaseDir: baseDir,
TLSConfig: o.TLSConfig,
}
ac, err := opts.NewConfig()
if err != nil {
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
return nil, fmt.Errorf("cannot parse TLS config for OAuth2: %w", err)
}
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
oi.ac = ac
if o.ProxyURL != "" {
u, err := url.Parse(o.ProxyURL)
if err != nil {
return nil, fmt.Errorf("cannot parse proxy_url=%q: %w", o.ProxyURL, err)
}
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
oi.proxyURL = o.ProxyURL
oi.proxyURLFunc = http.ProxyURL(u)
}
return oi, nil
}
func urlValuesFromMap(m map[string]string) url.Values {
result := make(url.Values, len(m))
for k, v := range m {
result[k] = []string{v}
}
return result
}
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
func (oi *oauth2ConfigInternal) initTokenSource() error {
tlsCfg, err := oi.ac.NewTLSConfig()
if err != nil {
return fmt.Errorf("cannot initialize TLS config for OAuth2: %w", err)
}
c := &http.Client{
Transport: &http.Transport{
TLSClientConfig: tlsCfg,
Proxy: oi.proxyURLFunc,
},
}
oi.ctx = context.WithValue(context.Background(), oauth2.HTTPClient, c)
oi.tokenSource = oi.cfg.TokenSource(oi.ctx)
return nil
}
func (oi *oauth2ConfigInternal) getTokenSource() (oauth2.TokenSource, error) {
oi.mu.Lock()
defer oi.mu.Unlock()
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
if oi.tokenSource == nil {
if err := oi.initTokenSource(); err != nil {
return nil, err
}
}
if oi.clientSecretFile == "" {
return oi.tokenSource, nil
}
newSecret, err := fscore.ReadPasswordFromFileOrHTTP(oi.clientSecretFile)
if err != nil {
return nil, fmt.Errorf("cannot read OAuth2 secret from %q: %w", oi.clientSecretFile, err)
}
if newSecret == oi.cfg.ClientSecret {
return oi.tokenSource, nil
}
oi.cfg.ClientSecret = newSecret
oi.tokenSource = oi.cfg.TokenSource(oi.ctx)
return oi.tokenSource, nil
}
// Config is auth config.
type Config struct {
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
tlsServerName string
tlsInsecureSkipVerify bool
tlsMinVersion uint16
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
getTLSRootCACached getTLSRootCAFunc
tlsRootCADigest string
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
getTLSCertCached getTLSCertFunc
tlsCertDigest string
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
getAuthHeaderCached getAuthHeaderFunc
authHeaderDigest string
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
headers []keyValue
headersDigest string
lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk (#5725) * lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk Added a custom `http.RoundTripper` implementation which checks for root CA content changes and updates `tls.Config` used by `http.RoundTripper` after detecting CA change. Client certificate changes are not tracked by this implementation since `tls.Config` already supports passing certificate dynamically by overriding `tls.Config.GetClientCertificate`. This change implements dynamic reload of root CA only for streaming client used for scraping. Blocking client (`fasthttp.HostClient`) does not support using custom transport so can't use this implementation. See: https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5526 Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: update NewRoundTripper API Update API to allow user to update only parameters required for transport. Add warning log when reloading Root CA failed. Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: fix mutex acquire logic Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: replace RWMutex with regular mutex to simplify the code - remove additional mutex used for getRootCABytes - require callee to use mutex - replace RWMutex with regular mutex Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: refactor - hold the mutex lock to avoid round tripper being re-created twice - move recreation logic into separate func to simplify the code Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> --------- Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> Co-authored-by: Nikolay <nik@victoriametrics.com>
2024-04-03 10:01:43 +02:00
tctx *tlsContext
}
type keyValue struct {
key string
value string
}
func parseHeaders(headers []string) ([]keyValue, error) {
if len(headers) == 0 {
return nil, nil
}
kvs := make([]keyValue, len(headers))
for i, h := range headers {
n := strings.IndexByte(h, ':')
if n < 0 {
return nil, fmt.Errorf(`missing ':' in header %q; expecting "key: value" format`, h)
}
kv := &kvs[i]
kv.key = http.CanonicalHeaderKey(strings.TrimSpace(h[:n]))
kv.value = strings.TrimSpace(h[n+1:])
}
return kvs, nil
}
// HeadersNoAuthString returns string representation of ac headers
func (ac *Config) HeadersNoAuthString() string {
if len(ac.headers) == 0 {
return ""
}
a := make([]string, len(ac.headers))
for i, h := range ac.headers {
a[i] = h.key + ": " + h.value + "\r\n"
}
return strings.Join(a, "")
}
// SetHeaders sets the configured ac headers to req.
func (ac *Config) SetHeaders(req *http.Request, setAuthHeader bool) error {
if ac.tlsServerName != "" {
// It tlsServerName is set, then it is likely the request is performed via IP address instead of hostname.
// In this case users expect that the specified tlsServerName is used as a Host header in the request to https server.
// See https://github.com/VictoriaMetrics/VictoriaMetrics/pull/5802
req.Host = ac.tlsServerName
}
reqHeaders := req.Header
for _, h := range ac.headers {
if h.key == "Host" {
// Host header must be set via req.Host - see https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5969
req.Host = h.value
} else {
reqHeaders.Set(h.key, h.value)
}
}
if setAuthHeader {
ah, err := ac.GetAuthHeader()
if err != nil {
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
return fmt.Errorf("failed to obtain Authorization request header: %w", err)
}
if ah != "" {
reqHeaders.Set("Authorization", ah)
}
}
return nil
}
// GetAuthHeader returns optional `Authorization: ...` http header.
func (ac *Config) GetAuthHeader() (string, error) {
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
if f := ac.getAuthHeaderCached; f != nil {
return f()
}
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
return "", nil
}
// String returns human-readable representation for ac.
//
// It is also used for comparing Config objects for equality. If two Config
// objects have the same string representation, then they are considered equal.
func (ac *Config) String() string {
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
return fmt.Sprintf("AuthHeader=%s, Headers=%s, TLSRootCA=%s, TLSCert=%s, TLSServerName=%s, TLSInsecureSkipVerify=%v, TLSMinVersion=%d",
ac.authHeaderDigest, ac.headersDigest, ac.tlsRootCADigest, ac.tlsCertDigest, ac.tlsServerName, ac.tlsInsecureSkipVerify, ac.tlsMinVersion)
}
lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk (#5725) * lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk Added a custom `http.RoundTripper` implementation which checks for root CA content changes and updates `tls.Config` used by `http.RoundTripper` after detecting CA change. Client certificate changes are not tracked by this implementation since `tls.Config` already supports passing certificate dynamically by overriding `tls.Config.GetClientCertificate`. This change implements dynamic reload of root CA only for streaming client used for scraping. Blocking client (`fasthttp.HostClient`) does not support using custom transport so can't use this implementation. See: https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5526 Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: update NewRoundTripper API Update API to allow user to update only parameters required for transport. Add warning log when reloading Root CA failed. Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: fix mutex acquire logic Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: replace RWMutex with regular mutex to simplify the code - remove additional mutex used for getRootCABytes - require callee to use mutex - replace RWMutex with regular mutex Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: refactor - hold the mutex lock to avoid round tripper being re-created twice - move recreation logic into separate func to simplify the code Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> --------- Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> Co-authored-by: Nikolay <nik@victoriametrics.com>
2024-04-03 10:01:43 +02:00
const tlsCertsCacheSeconds = 1
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
// getAuthHeaderFunc must return <value> for 'Authorization: <value>' http request header
type getAuthHeaderFunc func() (string, error)
func newGetAuthHeaderCached(getAuthHeader getAuthHeaderFunc) getAuthHeaderFunc {
if getAuthHeader == nil {
return nil
}
var mu sync.Mutex
var deadline uint64
var ah string
var err error
return func() (string, error) {
// Cahe the auth header and the error for up to a second in order to save CPU time
// on reading and parsing auth headers from files.
// This also reduces load on OAuth2 server when oauth2 config is enabled.
mu.Lock()
defer mu.Unlock()
if fasttime.UnixTimestamp() > deadline {
ah, err = getAuthHeader()
lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk (#5725) * lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk Added a custom `http.RoundTripper` implementation which checks for root CA content changes and updates `tls.Config` used by `http.RoundTripper` after detecting CA change. Client certificate changes are not tracked by this implementation since `tls.Config` already supports passing certificate dynamically by overriding `tls.Config.GetClientCertificate`. This change implements dynamic reload of root CA only for streaming client used for scraping. Blocking client (`fasthttp.HostClient`) does not support using custom transport so can't use this implementation. See: https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5526 Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: update NewRoundTripper API Update API to allow user to update only parameters required for transport. Add warning log when reloading Root CA failed. Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: fix mutex acquire logic Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: replace RWMutex with regular mutex to simplify the code - remove additional mutex used for getRootCABytes - require callee to use mutex - replace RWMutex with regular mutex Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: refactor - hold the mutex lock to avoid round tripper being re-created twice - move recreation logic into separate func to simplify the code Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> --------- Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> Co-authored-by: Nikolay <nik@victoriametrics.com>
2024-04-03 10:01:43 +02:00
deadline = fasttime.UnixTimestamp() + tlsCertsCacheSeconds
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
}
return ah, err
}
}
type getTLSRootCAFunc func() (*x509.CertPool, error)
func newGetTLSRootCACached(getTLSRootCA getTLSRootCAFunc) getTLSRootCAFunc {
if getTLSRootCA == nil {
return nil
}
var mu sync.Mutex
var deadline uint64
var rootCA *x509.CertPool
var err error
return func() (*x509.CertPool, error) {
// Cache the root CA and the error for up to a second in order to save CPU time
// on reading and parsing the root CA from files.
mu.Lock()
defer mu.Unlock()
if fasttime.UnixTimestamp() > deadline {
rootCA, err = getTLSRootCA()
lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk (#5725) * lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk Added a custom `http.RoundTripper` implementation which checks for root CA content changes and updates `tls.Config` used by `http.RoundTripper` after detecting CA change. Client certificate changes are not tracked by this implementation since `tls.Config` already supports passing certificate dynamically by overriding `tls.Config.GetClientCertificate`. This change implements dynamic reload of root CA only for streaming client used for scraping. Blocking client (`fasthttp.HostClient`) does not support using custom transport so can't use this implementation. See: https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5526 Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: update NewRoundTripper API Update API to allow user to update only parameters required for transport. Add warning log when reloading Root CA failed. Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: fix mutex acquire logic Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: replace RWMutex with regular mutex to simplify the code - remove additional mutex used for getRootCABytes - require callee to use mutex - replace RWMutex with regular mutex Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: refactor - hold the mutex lock to avoid round tripper being re-created twice - move recreation logic into separate func to simplify the code Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> --------- Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> Co-authored-by: Nikolay <nik@victoriametrics.com>
2024-04-03 10:01:43 +02:00
deadline = fasttime.UnixTimestamp() + tlsCertsCacheSeconds
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
}
return rootCA, err
}
}
type getTLSCertFunc func(cri *tls.CertificateRequestInfo) (*tls.Certificate, error)
func newGetTLSCertCached(getTLSCert getTLSCertFunc) getTLSCertFunc {
if getTLSCert == nil {
return nil
}
var mu sync.Mutex
var deadline uint64
var cert *tls.Certificate
var err error
return func(cri *tls.CertificateRequestInfo) (*tls.Certificate, error) {
// Cache the certificate and the error for up to a second in order to save CPU time
// on certificate parsing when TLS connections are frequently re-established.
mu.Lock()
defer mu.Unlock()
if fasttime.UnixTimestamp() > deadline {
cert, err = getTLSCert(cri)
lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk (#5725) * lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk Added a custom `http.RoundTripper` implementation which checks for root CA content changes and updates `tls.Config` used by `http.RoundTripper` after detecting CA change. Client certificate changes are not tracked by this implementation since `tls.Config` already supports passing certificate dynamically by overriding `tls.Config.GetClientCertificate`. This change implements dynamic reload of root CA only for streaming client used for scraping. Blocking client (`fasthttp.HostClient`) does not support using custom transport so can't use this implementation. See: https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5526 Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: update NewRoundTripper API Update API to allow user to update only parameters required for transport. Add warning log when reloading Root CA failed. Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: fix mutex acquire logic Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: replace RWMutex with regular mutex to simplify the code - remove additional mutex used for getRootCABytes - require callee to use mutex - replace RWMutex with regular mutex Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: refactor - hold the mutex lock to avoid round tripper being re-created twice - move recreation logic into separate func to simplify the code Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> --------- Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> Co-authored-by: Nikolay <nik@victoriametrics.com>
2024-04-03 10:01:43 +02:00
deadline = fasttime.UnixTimestamp() + tlsCertsCacheSeconds
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
}
return cert, err
}
}
// NewTLSConfig returns new TLS config for the given ac.
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
func (ac *Config) NewTLSConfig() (*tls.Config, error) {
tlsCfg := &tls.Config{
ClientSessionCache: tls.NewLRUClientSessionCache(0),
}
if ac == nil {
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
return tlsCfg, nil
}
tlsCfg.GetClientCertificate = ac.getTLSCertCached
if f := ac.getTLSRootCACached; f != nil {
rootCA, err := f()
if err != nil {
return nil, fmt.Errorf("cannot load root CAs: %w", err)
}
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
tlsCfg.RootCAs = rootCA
}
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
tlsCfg.ServerName = ac.tlsServerName
tlsCfg.InsecureSkipVerify = ac.tlsInsecureSkipVerify
tlsCfg.MinVersion = ac.tlsMinVersion
// Do not set tlsCfg.MaxVersion, since this has no sense from security PoV.
// This can only result in lower security level if improperly set.
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
return tlsCfg, nil
}
lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk (#5725) * lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk Added a custom `http.RoundTripper` implementation which checks for root CA content changes and updates `tls.Config` used by `http.RoundTripper` after detecting CA change. Client certificate changes are not tracked by this implementation since `tls.Config` already supports passing certificate dynamically by overriding `tls.Config.GetClientCertificate`. This change implements dynamic reload of root CA only for streaming client used for scraping. Blocking client (`fasthttp.HostClient`) does not support using custom transport so can't use this implementation. See: https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5526 Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: update NewRoundTripper API Update API to allow user to update only parameters required for transport. Add warning log when reloading Root CA failed. Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: fix mutex acquire logic Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: replace RWMutex with regular mutex to simplify the code - remove additional mutex used for getRootCABytes - require callee to use mutex - replace RWMutex with regular mutex Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: refactor - hold the mutex lock to avoid round tripper being re-created twice - move recreation logic into separate func to simplify the code Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> --------- Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> Co-authored-by: Nikolay <nik@victoriametrics.com>
2024-04-03 10:01:43 +02:00
// NewRoundTripper returns new http.RoundTripper for the given ac.
func (ac *Config) NewRoundTripper(builder func(*http.Transport)) (http.RoundTripper, error) {
cfg, err := ac.NewTLSConfig()
if err != nil {
return nil, fmt.Errorf("failed to initialize TLS config: %w", err)
}
if ac == nil {
tr := &http.Transport{
TLSClientConfig: cfg,
}
builder(tr)
return tr, nil
}
return ac.tctx.NewTLSRoundTripper(cfg, builder)
}
// NewConfig creates auth config for the given hcc.
func (hcc *HTTPClientConfig) NewConfig(baseDir string) (*Config, error) {
opts := &Options{
BaseDir: baseDir,
Authorization: hcc.Authorization,
BasicAuth: hcc.BasicAuth,
BearerToken: hcc.BearerToken.String(),
BearerTokenFile: hcc.BearerTokenFile,
OAuth2: hcc.OAuth2,
TLSConfig: hcc.TLSConfig,
Headers: hcc.Headers,
}
return opts.NewConfig()
}
// NewConfig creates auth config for the given pcc.
func (pcc *ProxyClientConfig) NewConfig(baseDir string) (*Config, error) {
opts := &Options{
BaseDir: baseDir,
Authorization: pcc.Authorization,
BasicAuth: pcc.BasicAuth,
BearerToken: pcc.BearerToken.String(),
BearerTokenFile: pcc.BearerTokenFile,
OAuth2: pcc.OAuth2,
TLSConfig: pcc.TLSConfig,
Headers: pcc.Headers,
}
return opts.NewConfig()
}
// NewConfig creates auth config for the given ba.
func (ba *BasicAuthConfig) NewConfig(baseDir string) (*Config, error) {
opts := &Options{
BaseDir: baseDir,
BasicAuth: ba,
}
return opts.NewConfig()
}
// Options contain options, which must be passed to NewConfig.
type Options struct {
// BaseDir is an optional path to a base directory for resolving
// relative filepaths in various config options.
//
// It is set to the current directory by default.
BaseDir string
// Authorization contains optional Authorization.
Authorization *Authorization
// BasicAuth contains optional BasicAuthConfig.
BasicAuth *BasicAuthConfig
// BearerToken contains optional bearer token.
BearerToken string
// BearerTokenFile contains optional path to a file with bearer token.
BearerTokenFile string
// OAuth2 contains optional OAuth2Config.
OAuth2 *OAuth2Config
// TLSconfig contains optional TLSConfig.
TLSConfig *TLSConfig
// Headers contains optional http request headers in the form 'Foo: bar'.
Headers []string
}
// NewConfig creates auth config from the given opts.
func (opts *Options) NewConfig() (*Config, error) {
baseDir := opts.BaseDir
if baseDir == "" {
baseDir = "."
}
var actx authContext
if opts.Authorization != nil {
if err := actx.initFromAuthorization(baseDir, opts.Authorization); err != nil {
return nil, err
}
}
if opts.BasicAuth != nil {
if actx.getAuthHeader != nil {
return nil, fmt.Errorf("cannot use both `authorization` and `basic_auth`")
}
if err := actx.initFromBasicAuthConfig(baseDir, opts.BasicAuth); err != nil {
return nil, err
}
}
if opts.BearerTokenFile != "" {
if actx.getAuthHeader != nil {
return nil, fmt.Errorf("cannot simultaneously use `authorization`, `basic_auth` and `bearer_token_file`")
}
if opts.BearerToken != "" {
return nil, fmt.Errorf("both `bearer_token`=%q and `bearer_token_file`=%q are set", opts.BearerToken, opts.BearerTokenFile)
}
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
actx.mustInitFromBearerTokenFile(baseDir, opts.BearerTokenFile)
}
if opts.BearerToken != "" {
if actx.getAuthHeader != nil {
return nil, fmt.Errorf("cannot simultaneously use `authorization`, `basic_auth` and `bearer_token`")
}
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
actx.mustInitFromBearerToken(opts.BearerToken)
}
if opts.OAuth2 != nil {
if actx.getAuthHeader != nil {
return nil, fmt.Errorf("cannot simultaneously use `authorization`, `basic_auth, `bearer_token` and `ouath2`")
}
if err := actx.initFromOAuth2Config(baseDir, opts.OAuth2); err != nil {
return nil, fmt.Errorf("cannot initialize oauth2: %w", err)
}
}
var tctx tlsContext
if opts.TLSConfig != nil {
if err := tctx.initFromTLSConfig(baseDir, opts.TLSConfig); err != nil {
return nil, fmt.Errorf("cannot initialize tls: %w", err)
}
}
headers, err := parseHeaders(opts.Headers)
if err != nil {
return nil, fmt.Errorf("cannot parse headers: %w", err)
}
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
hd := xxhash.New()
for _, kv := range headers {
hd.Sum([]byte(kv.key))
hd.Sum([]byte("="))
hd.Sum([]byte(kv.value))
hd.Sum([]byte(","))
}
headersDigest := fmt.Sprintf("digest(headers)=%d", hd.Sum64())
ac := &Config{
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
tlsServerName: tctx.serverName,
tlsInsecureSkipVerify: tctx.insecureSkipVerify,
tlsMinVersion: tctx.minVersion,
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
getTLSRootCACached: newGetTLSRootCACached(tctx.getTLSRootCA),
tlsRootCADigest: tctx.tlsRootCADigest,
getTLSCertCached: newGetTLSCertCached(tctx.getTLSCert),
tlsCertDigest: tctx.tlsCertDigest,
getAuthHeaderCached: newGetAuthHeaderCached(actx.getAuthHeader),
authHeaderDigest: actx.authHeaderDigest,
headers: headers,
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
headersDigest: headersDigest,
lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk (#5725) * lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk Added a custom `http.RoundTripper` implementation which checks for root CA content changes and updates `tls.Config` used by `http.RoundTripper` after detecting CA change. Client certificate changes are not tracked by this implementation since `tls.Config` already supports passing certificate dynamically by overriding `tls.Config.GetClientCertificate`. This change implements dynamic reload of root CA only for streaming client used for scraping. Blocking client (`fasthttp.HostClient`) does not support using custom transport so can't use this implementation. See: https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5526 Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: update NewRoundTripper API Update API to allow user to update only parameters required for transport. Add warning log when reloading Root CA failed. Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: fix mutex acquire logic Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: replace RWMutex with regular mutex to simplify the code - remove additional mutex used for getRootCABytes - require callee to use mutex - replace RWMutex with regular mutex Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: refactor - hold the mutex lock to avoid round tripper being re-created twice - move recreation logic into separate func to simplify the code Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> --------- Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> Co-authored-by: Nikolay <nik@victoriametrics.com>
2024-04-03 10:01:43 +02:00
tctx: &tctx,
}
return ac, nil
}
type authContext struct {
// getAuthHeader must return <value> for 'Authorization: <value>' http request header
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
getAuthHeader getAuthHeaderFunc
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
// authHeaderDigest must contain the digest for the used authorization
// The digest must be changed whenever the original config changes.
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
authHeaderDigest string
}
func (actx *authContext) initFromAuthorization(baseDir string, az *Authorization) error {
azType := "Bearer"
if az.Type != "" {
azType = az.Type
}
if az.CredentialsFile == "" {
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
ah := azType + " " + az.Credentials.String()
actx.getAuthHeader = func() (string, error) {
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
return ah, nil
}
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
actx.authHeaderDigest = fmt.Sprintf("custom(type=%q, creds=%q)", az.Type, az.Credentials)
return nil
}
if az.Credentials != nil {
return fmt.Errorf("both `credentials`=%q and `credentials_file`=%q are set", az.Credentials, az.CredentialsFile)
}
filePath := fscore.GetFilepath(baseDir, az.CredentialsFile)
actx.getAuthHeader = func() (string, error) {
token, err := fscore.ReadPasswordFromFileOrHTTP(filePath)
if err != nil {
return "", fmt.Errorf("cannot read credentials from `credentials_file`=%q: %w", az.CredentialsFile, err)
}
return azType + " " + token, nil
}
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
actx.authHeaderDigest = fmt.Sprintf("custom(type=%q, credsFile=%q)", az.Type, filePath)
return nil
}
func (actx *authContext) initFromBasicAuthConfig(baseDir string, ba *BasicAuthConfig) error {
username := ba.Username
usernameFile := ba.UsernameFile
password := ""
if ba.Password != nil {
password = ba.Password.S
}
passwordFile := ba.PasswordFile
if username == "" && usernameFile == "" {
return fmt.Errorf("missing `username` and `username_file` in `basic_auth` section; please specify one; " +
"see https://docs.victoriametrics.com/sd_configs.html#http-api-client-options")
}
if username != "" && usernameFile != "" {
return fmt.Errorf("both `username` and `username_file` are set in `basic_auth` section; please specify only one; " +
"see https://docs.victoriametrics.com/sd_configs.html#http-api-client-options")
}
if password != "" && passwordFile != "" {
return fmt.Errorf("both `password` and `password_file` are set in `basic_auth` section; please specify only one; " +
"see https://docs.victoriametrics.com/sd_configs.html#http-api-client-options")
}
if usernameFile != "" {
usernameFile = fscore.GetFilepath(baseDir, usernameFile)
}
if passwordFile != "" {
passwordFile = fscore.GetFilepath(baseDir, passwordFile)
}
actx.getAuthHeader = func() (string, error) {
usernameLocal := username
if usernameFile != "" {
s, err := fscore.ReadPasswordFromFileOrHTTP(usernameFile)
if err != nil {
return "", fmt.Errorf("cannot read username from `username_file`=%q: %w", usernameFile, err)
}
usernameLocal = s
}
passwordLocal := password
if passwordFile != "" {
s, err := fscore.ReadPasswordFromFileOrHTTP(passwordFile)
if err != nil {
return "", fmt.Errorf("cannot read password from `password_file`=%q: %w", passwordFile, err)
}
passwordLocal = s
}
// See https://en.wikipedia.org/wiki/Basic_access_authentication
token := usernameLocal + ":" + passwordLocal
token64 := base64.StdEncoding.EncodeToString([]byte(token))
return "Basic " + token64, nil
}
actx.authHeaderDigest = fmt.Sprintf("basic(username=%q, usernameFile=%q, password=%q, passwordFile=%q)", username, usernameFile, password, passwordFile)
return nil
}
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
func (actx *authContext) mustInitFromBearerTokenFile(baseDir string, bearerTokenFile string) {
filePath := fscore.GetFilepath(baseDir, bearerTokenFile)
actx.getAuthHeader = func() (string, error) {
token, err := fscore.ReadPasswordFromFileOrHTTP(filePath)
if err != nil {
return "", fmt.Errorf("cannot read bearer token from `bearer_token_file`=%q: %w", bearerTokenFile, err)
}
return "Bearer " + token, nil
}
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
actx.authHeaderDigest = fmt.Sprintf("bearer(tokenFile=%q)", filePath)
}
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
func (actx *authContext) mustInitFromBearerToken(bearerToken string) {
ah := "Bearer " + bearerToken
actx.getAuthHeader = func() (string, error) {
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
return ah, nil
}
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
actx.authHeaderDigest = fmt.Sprintf("bearer(token=%q)", bearerToken)
}
func (actx *authContext) initFromOAuth2Config(baseDir string, o *OAuth2Config) error {
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
oi, err := newOAuth2ConfigInternal(baseDir, o)
if err != nil {
return err
}
actx.getAuthHeader = func() (string, error) {
ts, err := oi.getTokenSource()
if err != nil {
return "", fmt.Errorf("cannot get OAuth2 tokenSource: %w", err)
}
t, err := ts.Token()
if err != nil {
return "", fmt.Errorf("cannot get OAuth2 token: %w", err)
}
return t.Type() + " " + t.AccessToken, nil
}
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
actx.authHeaderDigest = fmt.Sprintf("oauth2(%s)", oi.String())
return nil
}
type tlsContext struct {
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
getTLSCert getTLSCertFunc
tlsCertDigest string
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
getTLSRootCA getTLSRootCAFunc
lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk (#5725) * lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk Added a custom `http.RoundTripper` implementation which checks for root CA content changes and updates `tls.Config` used by `http.RoundTripper` after detecting CA change. Client certificate changes are not tracked by this implementation since `tls.Config` already supports passing certificate dynamically by overriding `tls.Config.GetClientCertificate`. This change implements dynamic reload of root CA only for streaming client used for scraping. Blocking client (`fasthttp.HostClient`) does not support using custom transport so can't use this implementation. See: https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5526 Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: update NewRoundTripper API Update API to allow user to update only parameters required for transport. Add warning log when reloading Root CA failed. Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: fix mutex acquire logic Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: replace RWMutex with regular mutex to simplify the code - remove additional mutex used for getRootCABytes - require callee to use mutex - replace RWMutex with regular mutex Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: refactor - hold the mutex lock to avoid round tripper being re-created twice - move recreation logic into separate func to simplify the code Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> --------- Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> Co-authored-by: Nikolay <nik@victoriametrics.com>
2024-04-03 10:01:43 +02:00
getRootCAPEM func() ([]byte, error)
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
tlsRootCADigest string
serverName string
insecureSkipVerify bool
minVersion uint16
}
func (tctx *tlsContext) initFromTLSConfig(baseDir string, tc *TLSConfig) error {
tctx.serverName = tc.ServerName
tctx.insecureSkipVerify = tc.InsecureSkipVerify
if len(tc.Key) != 0 || len(tc.Cert) != 0 {
cert, err := tls.X509KeyPair([]byte(tc.Cert), []byte(tc.Key))
if err != nil {
return fmt.Errorf("cannot load TLS certificate from the provided `cert` and `key` values: %w", err)
}
tctx.getTLSCert = func(*tls.CertificateRequestInfo) (*tls.Certificate, error) {
return &cert, nil
}
h := xxhash.Sum64([]byte(tc.Key)) ^ xxhash.Sum64([]byte(tc.Cert))
tctx.tlsCertDigest = fmt.Sprintf("digest(key+cert)=%d", h)
} else if tc.CertFile != "" || tc.KeyFile != "" {
certPath := fscore.GetFilepath(baseDir, tc.CertFile)
keyPath := fscore.GetFilepath(baseDir, tc.KeyFile)
lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk (#5725) * lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk Added a custom `http.RoundTripper` implementation which checks for root CA content changes and updates `tls.Config` used by `http.RoundTripper` after detecting CA change. Client certificate changes are not tracked by this implementation since `tls.Config` already supports passing certificate dynamically by overriding `tls.Config.GetClientCertificate`. This change implements dynamic reload of root CA only for streaming client used for scraping. Blocking client (`fasthttp.HostClient`) does not support using custom transport so can't use this implementation. See: https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5526 Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: update NewRoundTripper API Update API to allow user to update only parameters required for transport. Add warning log when reloading Root CA failed. Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: fix mutex acquire logic Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: replace RWMutex with regular mutex to simplify the code - remove additional mutex used for getRootCABytes - require callee to use mutex - replace RWMutex with regular mutex Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: refactor - hold the mutex lock to avoid round tripper being re-created twice - move recreation logic into separate func to simplify the code Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> --------- Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> Co-authored-by: Nikolay <nik@victoriametrics.com>
2024-04-03 10:01:43 +02:00
getCertsPEM := func() ([]byte, []byte, error) {
// Re-read TLS certificate from disk. This is needed for https://github.com/VictoriaMetrics/VictoriaMetrics/issues/1420
certData, err := fscore.ReadFileOrHTTP(certPath)
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
if err != nil {
lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk (#5725) * lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk Added a custom `http.RoundTripper` implementation which checks for root CA content changes and updates `tls.Config` used by `http.RoundTripper` after detecting CA change. Client certificate changes are not tracked by this implementation since `tls.Config` already supports passing certificate dynamically by overriding `tls.Config.GetClientCertificate`. This change implements dynamic reload of root CA only for streaming client used for scraping. Blocking client (`fasthttp.HostClient`) does not support using custom transport so can't use this implementation. See: https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5526 Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: update NewRoundTripper API Update API to allow user to update only parameters required for transport. Add warning log when reloading Root CA failed. Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: fix mutex acquire logic Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: replace RWMutex with regular mutex to simplify the code - remove additional mutex used for getRootCABytes - require callee to use mutex - replace RWMutex with regular mutex Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: refactor - hold the mutex lock to avoid round tripper being re-created twice - move recreation logic into separate func to simplify the code Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> --------- Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> Co-authored-by: Nikolay <nik@victoriametrics.com>
2024-04-03 10:01:43 +02:00
return nil, nil, fmt.Errorf("cannot read TLS certificate from %q: %w", certPath, err)
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
}
keyData, err := fscore.ReadFileOrHTTP(keyPath)
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
if err != nil {
lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk (#5725) * lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk Added a custom `http.RoundTripper` implementation which checks for root CA content changes and updates `tls.Config` used by `http.RoundTripper` after detecting CA change. Client certificate changes are not tracked by this implementation since `tls.Config` already supports passing certificate dynamically by overriding `tls.Config.GetClientCertificate`. This change implements dynamic reload of root CA only for streaming client used for scraping. Blocking client (`fasthttp.HostClient`) does not support using custom transport so can't use this implementation. See: https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5526 Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: update NewRoundTripper API Update API to allow user to update only parameters required for transport. Add warning log when reloading Root CA failed. Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: fix mutex acquire logic Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: replace RWMutex with regular mutex to simplify the code - remove additional mutex used for getRootCABytes - require callee to use mutex - replace RWMutex with regular mutex Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: refactor - hold the mutex lock to avoid round tripper being re-created twice - move recreation logic into separate func to simplify the code Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> --------- Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> Co-authored-by: Nikolay <nik@victoriametrics.com>
2024-04-03 10:01:43 +02:00
return nil, nil, fmt.Errorf("cannot read TLS key from %q: %w", keyPath, err)
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
}
lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk (#5725) * lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk Added a custom `http.RoundTripper` implementation which checks for root CA content changes and updates `tls.Config` used by `http.RoundTripper` after detecting CA change. Client certificate changes are not tracked by this implementation since `tls.Config` already supports passing certificate dynamically by overriding `tls.Config.GetClientCertificate`. This change implements dynamic reload of root CA only for streaming client used for scraping. Blocking client (`fasthttp.HostClient`) does not support using custom transport so can't use this implementation. See: https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5526 Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: update NewRoundTripper API Update API to allow user to update only parameters required for transport. Add warning log when reloading Root CA failed. Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: fix mutex acquire logic Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: replace RWMutex with regular mutex to simplify the code - remove additional mutex used for getRootCABytes - require callee to use mutex - replace RWMutex with regular mutex Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: refactor - hold the mutex lock to avoid round tripper being re-created twice - move recreation logic into separate func to simplify the code Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> --------- Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> Co-authored-by: Nikolay <nik@victoriametrics.com>
2024-04-03 10:01:43 +02:00
return certData, keyData, nil
}
tctx.getTLSCert = func(*tls.CertificateRequestInfo) (*tls.Certificate, error) {
certData, keyData, err := getCertsPEM()
if err != nil {
return nil, err
}
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
cert, err := tls.X509KeyPair(certData, keyData)
if err != nil {
return nil, fmt.Errorf("cannot load TLS certificate from `cert_file`=%q, `key_file`=%q: %w", tc.CertFile, tc.KeyFile, err)
}
return &cert, nil
}
tctx.tlsCertDigest = fmt.Sprintf("certFile=%q, keyFile=%q", tc.CertFile, tc.KeyFile)
}
if len(tc.CA) != 0 {
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
rootCA := x509.NewCertPool()
if !rootCA.AppendCertsFromPEM([]byte(tc.CA)) {
return fmt.Errorf("cannot parse data from `ca` value")
}
lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk (#5725) * lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk Added a custom `http.RoundTripper` implementation which checks for root CA content changes and updates `tls.Config` used by `http.RoundTripper` after detecting CA change. Client certificate changes are not tracked by this implementation since `tls.Config` already supports passing certificate dynamically by overriding `tls.Config.GetClientCertificate`. This change implements dynamic reload of root CA only for streaming client used for scraping. Blocking client (`fasthttp.HostClient`) does not support using custom transport so can't use this implementation. See: https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5526 Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: update NewRoundTripper API Update API to allow user to update only parameters required for transport. Add warning log when reloading Root CA failed. Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: fix mutex acquire logic Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: replace RWMutex with regular mutex to simplify the code - remove additional mutex used for getRootCABytes - require callee to use mutex - replace RWMutex with regular mutex Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: refactor - hold the mutex lock to avoid round tripper being re-created twice - move recreation logic into separate func to simplify the code Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> --------- Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> Co-authored-by: Nikolay <nik@victoriametrics.com>
2024-04-03 10:01:43 +02:00
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
tctx.getTLSRootCA = func() (*x509.CertPool, error) {
return rootCA, nil
}
lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk (#5725) * lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk Added a custom `http.RoundTripper` implementation which checks for root CA content changes and updates `tls.Config` used by `http.RoundTripper` after detecting CA change. Client certificate changes are not tracked by this implementation since `tls.Config` already supports passing certificate dynamically by overriding `tls.Config.GetClientCertificate`. This change implements dynamic reload of root CA only for streaming client used for scraping. Blocking client (`fasthttp.HostClient`) does not support using custom transport so can't use this implementation. See: https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5526 Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: update NewRoundTripper API Update API to allow user to update only parameters required for transport. Add warning log when reloading Root CA failed. Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: fix mutex acquire logic Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: replace RWMutex with regular mutex to simplify the code - remove additional mutex used for getRootCABytes - require callee to use mutex - replace RWMutex with regular mutex Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: refactor - hold the mutex lock to avoid round tripper being re-created twice - move recreation logic into separate func to simplify the code Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> --------- Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> Co-authored-by: Nikolay <nik@victoriametrics.com>
2024-04-03 10:01:43 +02:00
tctx.getRootCAPEM = func() ([]byte, error) {
return []byte(tc.CA), nil
}
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
h := xxhash.Sum64([]byte(tc.CA))
tctx.tlsRootCADigest = fmt.Sprintf("digest(CA)=%d", h)
} else if tc.CAFile != "" {
path := fscore.GetFilepath(baseDir, tc.CAFile)
lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk (#5725) * lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk Added a custom `http.RoundTripper` implementation which checks for root CA content changes and updates `tls.Config` used by `http.RoundTripper` after detecting CA change. Client certificate changes are not tracked by this implementation since `tls.Config` already supports passing certificate dynamically by overriding `tls.Config.GetClientCertificate`. This change implements dynamic reload of root CA only for streaming client used for scraping. Blocking client (`fasthttp.HostClient`) does not support using custom transport so can't use this implementation. See: https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5526 Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: update NewRoundTripper API Update API to allow user to update only parameters required for transport. Add warning log when reloading Root CA failed. Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: fix mutex acquire logic Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: replace RWMutex with regular mutex to simplify the code - remove additional mutex used for getRootCABytes - require callee to use mutex - replace RWMutex with regular mutex Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: refactor - hold the mutex lock to avoid round tripper being re-created twice - move recreation logic into separate func to simplify the code Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> --------- Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> Co-authored-by: Nikolay <nik@victoriametrics.com>
2024-04-03 10:01:43 +02:00
getRootCAPEM := func() ([]byte, error) {
data, err := fscore.ReadFileOrHTTP(path)
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
if err != nil {
return nil, fmt.Errorf("cannot read `ca_file`: %w", err)
}
lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk (#5725) * lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk Added a custom `http.RoundTripper` implementation which checks for root CA content changes and updates `tls.Config` used by `http.RoundTripper` after detecting CA change. Client certificate changes are not tracked by this implementation since `tls.Config` already supports passing certificate dynamically by overriding `tls.Config.GetClientCertificate`. This change implements dynamic reload of root CA only for streaming client used for scraping. Blocking client (`fasthttp.HostClient`) does not support using custom transport so can't use this implementation. See: https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5526 Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: update NewRoundTripper API Update API to allow user to update only parameters required for transport. Add warning log when reloading Root CA failed. Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: fix mutex acquire logic Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: replace RWMutex with regular mutex to simplify the code - remove additional mutex used for getRootCABytes - require callee to use mutex - replace RWMutex with regular mutex Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: refactor - hold the mutex lock to avoid round tripper being re-created twice - move recreation logic into separate func to simplify the code Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> --------- Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> Co-authored-by: Nikolay <nik@victoriametrics.com>
2024-04-03 10:01:43 +02:00
return data, nil
}
tctx.getTLSRootCA = func() (*x509.CertPool, error) {
data, err := getRootCAPEM()
if err != nil {
return nil, err
}
lib/promauth: follow-up for e16d3f5639d67ff970975d342aaa276e339e9b0c - Make sure that invalid/missing TLS CA file or TLS client certificate files at vmagent startup don't prevent from processing the corresponding scrape targets after the file becomes correct, without the need to restart vmagent. Previously scrape targets with invalid TLS CA file or TLS client certificate files were permanently dropped after the first attempt to initialize them, and they didn't appear until the next vmagent reload or the next change in other places of the loaded scrape configs. - Make sure that TLS CA is properly re-loaded from file after it changes without the need to restart vmagent. Previously the old TLS CA was used until vmagent restart. - Properly handle errors during http request creation for the second attempt to send data to remote system at vmagent and vmalert. Previously failed request creation could result in nil pointer dereferencing, since the returned request is nil on error. - Add more context to the logged error during AWS sigv4 request signing before sending the data to -remoteWrite.url at vmagent. Previously it could miss details on the source of the request. - Do not create a new HTTP client per second when generating OAuth2 token needed to put in Authorization header of every http request issued by vmagent during service discovery or target scraping. Re-use the HTTP client instead until the corresponding scrape config changes. - Cache error at lib/promauth.Config.GetAuthHeader() in the same way as the auth header is cached, e.g. the error is cached for a second now. This should reduce load on CPU and OAuth2 server when auth header cannot be obtained because of temporary error. - Share tls.Config.GetClientCertificate function among multiple scrape targets with the same tls_config. Cache the loaded certificate and the error for one second. This should significantly reduce CPU load when scraping big number of targets with the same tls_config. - Allow loading TLS certificates from HTTP and HTTPs urls by specifying these urls at `tls_config->cert_file` and `tls_config->key_file`. - Improve test coverage at lib/promauth - Skip unreachable or invalid files specified at `scrape_config_files` during vmagent startup, since these files may become valid later. Previously vmagent was exitting in this case. Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4959
2023-10-25 23:19:33 +02:00
rootCA := x509.NewCertPool()
if !rootCA.AppendCertsFromPEM(data) {
return nil, fmt.Errorf("cannot parse data read from `ca_file` %q", tc.CAFile)
}
return rootCA, nil
}
lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk (#5725) * lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk Added a custom `http.RoundTripper` implementation which checks for root CA content changes and updates `tls.Config` used by `http.RoundTripper` after detecting CA change. Client certificate changes are not tracked by this implementation since `tls.Config` already supports passing certificate dynamically by overriding `tls.Config.GetClientCertificate`. This change implements dynamic reload of root CA only for streaming client used for scraping. Blocking client (`fasthttp.HostClient`) does not support using custom transport so can't use this implementation. See: https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5526 Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: update NewRoundTripper API Update API to allow user to update only parameters required for transport. Add warning log when reloading Root CA failed. Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: fix mutex acquire logic Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: replace RWMutex with regular mutex to simplify the code - remove additional mutex used for getRootCABytes - require callee to use mutex - replace RWMutex with regular mutex Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: refactor - hold the mutex lock to avoid round tripper being re-created twice - move recreation logic into separate func to simplify the code Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> --------- Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> Co-authored-by: Nikolay <nik@victoriametrics.com>
2024-04-03 10:01:43 +02:00
tctx.getRootCAPEM = func() ([]byte, error) {
data, err := getRootCAPEM()
if err != nil {
return nil, err
}
return data, nil
}
lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk (#5725) * lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk Added a custom `http.RoundTripper` implementation which checks for root CA content changes and updates `tls.Config` used by `http.RoundTripper` after detecting CA change. Client certificate changes are not tracked by this implementation since `tls.Config` already supports passing certificate dynamically by overriding `tls.Config.GetClientCertificate`. This change implements dynamic reload of root CA only for streaming client used for scraping. Blocking client (`fasthttp.HostClient`) does not support using custom transport so can't use this implementation. See: https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5526 Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: update NewRoundTripper API Update API to allow user to update only parameters required for transport. Add warning log when reloading Root CA failed. Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: fix mutex acquire logic Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: replace RWMutex with regular mutex to simplify the code - remove additional mutex used for getRootCABytes - require callee to use mutex - replace RWMutex with regular mutex Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: refactor - hold the mutex lock to avoid round tripper being re-created twice - move recreation logic into separate func to simplify the code Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> --------- Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> Co-authored-by: Nikolay <nik@victoriametrics.com>
2024-04-03 10:01:43 +02:00
// Does not hash file contents, since they may change at any time.
// TLSRoundTripper must be used to automatically update the root CA.
tctx.tlsRootCADigest = fmt.Sprintf("caFile=%q", tc.CAFile)
}
v, err := netutil.ParseTLSVersion(tc.MinVersion)
if err != nil {
return fmt.Errorf("cannot parse `min_version`: %w", err)
}
tctx.minVersion = v
return nil
}
lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk (#5725) * lib/{promauth,promscrape}: automatically refresh root CA certificates after changes on disk Added a custom `http.RoundTripper` implementation which checks for root CA content changes and updates `tls.Config` used by `http.RoundTripper` after detecting CA change. Client certificate changes are not tracked by this implementation since `tls.Config` already supports passing certificate dynamically by overriding `tls.Config.GetClientCertificate`. This change implements dynamic reload of root CA only for streaming client used for scraping. Blocking client (`fasthttp.HostClient`) does not support using custom transport so can't use this implementation. See: https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5526 Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: update NewRoundTripper API Update API to allow user to update only parameters required for transport. Add warning log when reloading Root CA failed. Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: fix mutex acquire logic Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: replace RWMutex with regular mutex to simplify the code - remove additional mutex used for getRootCABytes - require callee to use mutex - replace RWMutex with regular mutex Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> * lib/promauth/config: refactor - hold the mutex lock to avoid round tripper being re-created twice - move recreation logic into separate func to simplify the code Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> --------- Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com> Co-authored-by: Nikolay <nik@victoriametrics.com>
2024-04-03 10:01:43 +02:00
// NewTLSRoundTripper returns new http.RoundTripper which automatically updates
// RootCA in tls.Config whenever it changes.
func (tctx *tlsContext) NewTLSRoundTripper(cfg *tls.Config, builder func(transport *http.Transport)) (http.RoundTripper, error) {
// TLS context is not initialized so use the provided RoundTripper without wrapper
if tctx == nil || tctx.getRootCAPEM == nil {
tr := &http.Transport{
TLSClientConfig: cfg,
}
builder(tr)
return tr, nil
}
var deadline uint64
var rootCA []byte
var err error
getTLSDigestsLocked := func() []byte {
if fasttime.UnixTimestamp() > deadline {
rootCA, err = tctx.getRootCAPEM()
if err != nil {
logger.Warnf("cannot load root CA: %s", err)
}
deadline = fasttime.UnixTimestamp() + tlsCertsCacheSeconds
}
return rootCA
}
tr := &http.Transport{
TLSClientConfig: cfg,
}
builder(tr)
return &TLSRoundTripper{
builder: builder,
getRootCABytesLocked: getTLSDigestsLocked,
getTLSRootCA: newGetTLSRootCACached(tctx.getTLSRootCA),
config: cfg,
tr: tr,
}, nil
}
var _ http.RoundTripper = &TLSRoundTripper{}
// TLSRoundTripper is an implementation of http.RoundTripper which automatically
// updates RootCA in tls.Config whenever it changes.
type TLSRoundTripper struct {
builder func(*http.Transport)
getRootCABytesLocked func() []byte
getTLSRootCA func() (*x509.CertPool, error)
config *tls.Config
tr http.RoundTripper
m sync.Mutex
rootCABytes []byte
}
// RoundTrip implements http.RoundTripper.RoundTrip.
// It automatically updates RootCA in tls.Config whenever it changes.
func (t *TLSRoundTripper) RoundTrip(request *http.Request) (*http.Response, error) {
t.m.Lock()
rootCABytes := t.getRootCABytesLocked()
equal := bytes.Equal(rootCABytes, t.rootCABytes)
if equal {
t.m.Unlock()
return t.tr.RoundTrip(request)
}
err := t.recreateRoundTripperLocked(rootCABytes)
t.m.Unlock()
if err != nil {
return nil, fmt.Errorf("cannot recreate RoundTripper: %w", err)
}
return t.tr.RoundTrip(request)
}
func (t *TLSRoundTripper) recreateRoundTripperLocked(rootCABytes []byte) error {
newRootCaPool, err := t.getTLSRootCA()
if err != nil {
return fmt.Errorf("cannot load root CAs: %w", err)
}
newTLSConfig := t.config.Clone()
newTLSConfig.RootCAs = newRootCaPool
// Reset ClientSessionCache, since it may contain sessions for the old root CA.
newTLSConfig.ClientSessionCache = tls.NewLRUClientSessionCache(0)
tr := &http.Transport{
TLSClientConfig: newTLSConfig,
}
t.builder(tr)
oldRt := t.tr
t.tr = tr
t.rootCABytes = rootCABytes
t.config = newTLSConfig
if oldRt != nil {
closeIdleConnections(oldRt)
}
return nil
}
type closeIdler interface {
CloseIdleConnections()
}
func closeIdleConnections(t http.RoundTripper) {
if ci, ok := t.(closeIdler); ok {
ci.CloseIdleConnections()
}
}