2020-02-23 12:35:47 +01:00
|
|
|
package vmimport
|
|
|
|
|
|
|
|
import (
|
2020-09-26 03:20:47 +02:00
|
|
|
"bufio"
|
2020-02-23 12:35:47 +01:00
|
|
|
"fmt"
|
|
|
|
"io"
|
|
|
|
"net/http"
|
|
|
|
"runtime"
|
|
|
|
"sync"
|
|
|
|
|
|
|
|
"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
|
2020-08-16 16:05:52 +02:00
|
|
|
"github.com/VictoriaMetrics/VictoriaMetrics/lib/flagutil"
|
2020-09-26 03:20:47 +02:00
|
|
|
"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
|
2020-02-23 12:35:47 +01:00
|
|
|
"github.com/VictoriaMetrics/VictoriaMetrics/lib/protoparser/common"
|
|
|
|
"github.com/VictoriaMetrics/metrics"
|
|
|
|
)
|
|
|
|
|
2020-09-26 03:20:47 +02:00
|
|
|
var maxLineLen = flagutil.NewBytes("import.maxLineLen", 100*1024*1024, "The maximum length in bytes of a single line accepted by /api/v1/import; "+
|
|
|
|
"the line length can be limited with `max_rows_per_line` query arg passed to /api/v1/export")
|
2020-02-23 12:35:47 +01:00
|
|
|
|
|
|
|
// ParseStream parses /api/v1/import lines from req and calls callback for the parsed rows.
|
|
|
|
//
|
|
|
|
// The callback can be called multiple times for streamed data from req.
|
|
|
|
//
|
|
|
|
// callback shouldn't hold rows after returning.
|
2020-09-26 03:20:47 +02:00
|
|
|
// callback is called from multiple concurrent goroutines.
|
2020-02-23 12:35:47 +01:00
|
|
|
func ParseStream(req *http.Request, callback func(rows []Row) error) error {
|
|
|
|
r := req.Body
|
|
|
|
if req.Header.Get("Content-Encoding") == "gzip" {
|
|
|
|
zr, err := common.GetGzipReader(r)
|
|
|
|
if err != nil {
|
2020-06-30 21:58:18 +02:00
|
|
|
return fmt.Errorf("cannot read gzipped vmimport data: %w", err)
|
2020-02-23 12:35:47 +01:00
|
|
|
}
|
|
|
|
defer common.PutGzipReader(zr)
|
|
|
|
r = zr
|
|
|
|
}
|
2020-09-26 03:20:47 +02:00
|
|
|
// By default req.Body uses 4Kb buffer. This size is too small for typical request to /api/v1/import,
|
|
|
|
// so use slightly bigger buffer in order to reduce read syscall overhead.
|
|
|
|
br := bufio.NewReaderSize(r, 1024*1024)
|
|
|
|
|
|
|
|
// Start gomaxprocs workers for processing the parsed data in parallel.
|
|
|
|
gomaxprocs := runtime.GOMAXPROCS(-1)
|
|
|
|
workCh := make(chan *unmarshalWork, 8*gomaxprocs)
|
|
|
|
var wg sync.WaitGroup
|
|
|
|
defer func() {
|
|
|
|
close(workCh)
|
|
|
|
wg.Wait()
|
|
|
|
}()
|
|
|
|
wg.Add(gomaxprocs)
|
|
|
|
for i := 0; i < gomaxprocs; i++ {
|
|
|
|
go func() {
|
|
|
|
defer wg.Done()
|
|
|
|
for uw := range workCh {
|
|
|
|
uw.rows.Unmarshal(bytesutil.ToUnsafeString(uw.reqBuf))
|
|
|
|
rows := uw.rows.Rows
|
|
|
|
for i := range rows {
|
|
|
|
row := &rows[i]
|
|
|
|
rowsRead.Add(len(row.Timestamps))
|
|
|
|
}
|
|
|
|
if err := callback(rows); err != nil {
|
|
|
|
logger.Errorf("error when processing imported data: %s", err)
|
|
|
|
putUnmarshalWork(uw)
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
putUnmarshalWork(uw)
|
|
|
|
}
|
|
|
|
}()
|
|
|
|
}
|
2020-02-23 12:35:47 +01:00
|
|
|
|
|
|
|
ctx := getStreamContext()
|
|
|
|
defer putStreamContext(ctx)
|
2020-09-26 03:20:47 +02:00
|
|
|
for ctx.Read(br) {
|
|
|
|
uw := getUnmarshalWork()
|
|
|
|
uw.reqBuf = append(uw.reqBuf[:0], ctx.reqBuf...)
|
|
|
|
workCh <- uw
|
2020-02-23 12:35:47 +01:00
|
|
|
}
|
|
|
|
return ctx.Error()
|
|
|
|
}
|
|
|
|
|
|
|
|
func (ctx *streamContext) Read(r io.Reader) bool {
|
2020-07-10 10:11:17 +02:00
|
|
|
readCalls.Inc()
|
2020-02-23 12:35:47 +01:00
|
|
|
if ctx.err != nil {
|
|
|
|
return false
|
|
|
|
}
|
2020-08-16 16:05:52 +02:00
|
|
|
ctx.reqBuf, ctx.tailBuf, ctx.err = common.ReadLinesBlockExt(r, ctx.reqBuf, ctx.tailBuf, maxLineLen.N)
|
2020-02-23 12:35:47 +01:00
|
|
|
if ctx.err != nil {
|
|
|
|
if ctx.err != io.EOF {
|
|
|
|
readErrors.Inc()
|
2020-06-30 21:58:18 +02:00
|
|
|
ctx.err = fmt.Errorf("cannot read vmimport data: %w", ctx.err)
|
2020-02-23 12:35:47 +01:00
|
|
|
}
|
|
|
|
return false
|
|
|
|
}
|
|
|
|
return true
|
|
|
|
}
|
|
|
|
|
|
|
|
var (
|
2020-02-28 19:19:35 +01:00
|
|
|
readCalls = metrics.NewCounter(`vm_protoparser_read_calls_total{type="vmimport"}`)
|
|
|
|
readErrors = metrics.NewCounter(`vm_protoparser_read_errors_total{type="vmimport"}`)
|
|
|
|
rowsRead = metrics.NewCounter(`vm_protoparser_rows_read_total{type="vmimport"}`)
|
2020-02-23 12:35:47 +01:00
|
|
|
)
|
|
|
|
|
|
|
|
type streamContext struct {
|
|
|
|
reqBuf []byte
|
|
|
|
tailBuf []byte
|
|
|
|
err error
|
|
|
|
}
|
|
|
|
|
|
|
|
func (ctx *streamContext) Error() error {
|
|
|
|
if ctx.err == io.EOF {
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
return ctx.err
|
|
|
|
}
|
|
|
|
|
|
|
|
func (ctx *streamContext) reset() {
|
|
|
|
ctx.reqBuf = ctx.reqBuf[:0]
|
|
|
|
ctx.tailBuf = ctx.tailBuf[:0]
|
|
|
|
ctx.err = nil
|
|
|
|
}
|
|
|
|
|
|
|
|
func getStreamContext() *streamContext {
|
|
|
|
select {
|
|
|
|
case ctx := <-streamContextPoolCh:
|
|
|
|
return ctx
|
|
|
|
default:
|
|
|
|
if v := streamContextPool.Get(); v != nil {
|
|
|
|
return v.(*streamContext)
|
|
|
|
}
|
|
|
|
return &streamContext{}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
func putStreamContext(ctx *streamContext) {
|
|
|
|
ctx.reset()
|
|
|
|
select {
|
|
|
|
case streamContextPoolCh <- ctx:
|
|
|
|
default:
|
|
|
|
streamContextPool.Put(ctx)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
var streamContextPool sync.Pool
|
|
|
|
var streamContextPoolCh = make(chan *streamContext, runtime.GOMAXPROCS(-1))
|
2020-09-26 03:20:47 +02:00
|
|
|
|
|
|
|
type unmarshalWork struct {
|
|
|
|
rows Rows
|
|
|
|
reqBuf []byte
|
|
|
|
}
|
|
|
|
|
|
|
|
func (uw *unmarshalWork) reset() {
|
|
|
|
uw.rows.Reset()
|
|
|
|
uw.reqBuf = uw.reqBuf[:0]
|
|
|
|
}
|
|
|
|
|
|
|
|
func getUnmarshalWork() *unmarshalWork {
|
|
|
|
v := unmarshalWorkPool.Get()
|
|
|
|
if v == nil {
|
|
|
|
return &unmarshalWork{}
|
|
|
|
}
|
|
|
|
return v.(*unmarshalWork)
|
|
|
|
}
|
|
|
|
|
|
|
|
func putUnmarshalWork(uw *unmarshalWork) {
|
|
|
|
uw.reset()
|
|
|
|
unmarshalWorkPool.Put(uw)
|
|
|
|
}
|
|
|
|
|
|
|
|
var unmarshalWorkPool sync.Pool
|