VictoriaMetrics/lib/protoparser/clusternative/streamparser.go
Aliaksandr Valialkin 8752cce157
app/vminsert: reduce the max packet size, which vminsert can send to vmstorage
This reduces the max memory usage for vminsert and vmstorage under heavy ingestion rate
by up to 50% on production workload
2022-04-05 15:39:58 +03:00

163 lines
5.3 KiB
Go

package clusternative
import (
"fmt"
"io"
"sync"
"time"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/consts"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/encoding"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/handshake"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/protoparser/common"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/storage"
"github.com/VictoriaMetrics/metrics"
)
// ParseStream parses data sent from vminsert to bc and calls callback for parsed rows.
// Optional function isReadOnly must return true if the storage cannot accept new data.
// In thic case the data read from bc isn't accepted and the readonly status is sent back bc.
//
// The callback can be called concurrently multiple times for streamed data from req.
//
// callback shouldn't hold block after returning.
func ParseStream(bc *handshake.BufferedConn, callback func(rows []storage.MetricRow) error, isReadOnly func() bool) error {
var wg sync.WaitGroup
var (
callbackErrLock sync.Mutex
callbackErr error
)
for {
// Do not use unmarshalWork pool, since every unmarshalWork structure usually occupies
// big amounts of memory (more than consts.MaxInsertPacketSizeForVMStorage bytes).
// The pool would result in increased memory usage.
uw := &unmarshalWork{}
uw.callback = func(rows []storage.MetricRow) {
if err := callback(rows); err != nil {
processErrors.Inc()
callbackErrLock.Lock()
if callbackErr == nil {
callbackErr = fmt.Errorf("error when processing native block: %w", err)
}
callbackErrLock.Unlock()
}
}
uw.wg = &wg
var err error
uw.reqBuf, err = readBlock(uw.reqBuf[:0], bc, isReadOnly)
if err != nil {
wg.Wait()
if err == io.EOF {
// Remote end gracefully closed the connection.
return nil
}
return err
}
blocksRead.Inc()
wg.Add(1)
common.ScheduleUnmarshalWork(uw)
}
}
// readBlock reads the next data block from vminsert-initiated bc, appends it to dst and returns the result.
func readBlock(dst []byte, bc *handshake.BufferedConn, isReadOnly func() bool) ([]byte, error) {
sizeBuf := auxBufPool.Get()
defer auxBufPool.Put(sizeBuf)
sizeBuf.B = bytesutil.ResizeNoCopyMayOverallocate(sizeBuf.B, 8)
if _, err := io.ReadFull(bc, sizeBuf.B); err != nil {
if err != io.EOF {
readErrors.Inc()
err = fmt.Errorf("cannot read packet size: %w", err)
}
return dst, err
}
packetSize := encoding.UnmarshalUint64(sizeBuf.B)
if packetSize > consts.MaxInsertPacketSizeForVMStorage {
parseErrors.Inc()
return dst, fmt.Errorf("too big packet size: %d; shouldn't exceed %d", packetSize, consts.MaxInsertPacketSizeForVMStorage)
}
dstLen := len(dst)
dst = bytesutil.ResizeWithCopyMayOverallocate(dst, dstLen+int(packetSize))
if n, err := io.ReadFull(bc, dst[dstLen:]); err != nil {
readErrors.Inc()
return dst, fmt.Errorf("cannot read packet with size %d bytes: %w; read only %d bytes", packetSize, err, n)
}
if isReadOnly != nil && isReadOnly() {
// The vmstorage is in readonly mode, so drop the read block of data
// and send `read only` status to vminsert.
dst = dst[:dstLen]
if err := sendAck(bc, 2); err != nil {
writeErrors.Inc()
return dst, fmt.Errorf("cannot send readonly status to vminsert: %w", err)
}
return dst, nil
}
// Send `ack` to vminsert that the packet has been received.
if err := sendAck(bc, 1); err != nil {
writeErrors.Inc()
return dst, fmt.Errorf("cannot send `ack` to vminsert: %w", err)
}
return dst, nil
}
func sendAck(bc *handshake.BufferedConn, status byte) error {
deadline := time.Now().Add(5 * time.Second)
if err := bc.SetWriteDeadline(deadline); err != nil {
return fmt.Errorf("cannot set write deadline: %w", err)
}
b := auxBufPool.Get()
defer auxBufPool.Put(b)
b.B = append(b.B[:0], status)
if _, err := bc.Write(b.B); err != nil {
return err
}
if err := bc.Flush(); err != nil {
return err
}
return nil
}
var auxBufPool bytesutil.ByteBufferPool
var (
readErrors = metrics.NewCounter(`vm_protoparser_read_errors_total{type="clusternative"}`)
writeErrors = metrics.NewCounter(`vm_protoparser_write_errors_total{type="clusternative"}`)
rowsRead = metrics.NewCounter(`vm_protoparser_rows_read_total{type="clusternative"}`)
blocksRead = metrics.NewCounter(`vm_protoparser_blocks_read_total{type="clusternative"}`)
parseErrors = metrics.NewCounter(`vm_protoparser_parse_errors_total{type="clusternative"}`)
processErrors = metrics.NewCounter(`vm_protoparser_process_errors_total{type="clusternative"}`)
)
type unmarshalWork struct {
wg *sync.WaitGroup
callback func(rows []storage.MetricRow)
reqBuf []byte
mrs []storage.MetricRow
}
// Unmarshal implements common.UnmarshalWork
func (uw *unmarshalWork) Unmarshal() {
reqBuf := uw.reqBuf
for len(reqBuf) > 0 {
// Limit the number of rows passed to callback in order to reduce memory usage
// when processing big packets of rows.
mrs, tail, err := storage.UnmarshalMetricRows(uw.mrs[:0], reqBuf, maxRowsPerCallback)
uw.mrs = mrs
if err != nil {
parseErrors.Inc()
logger.Errorf("cannot unmarshal MetricRow from clusternative block with size %d (remaining %d bytes): %s", len(reqBuf), len(tail), err)
break
}
rowsRead.Add(len(mrs))
uw.callback(mrs)
reqBuf = tail
}
wg := uw.wg
wg.Done()
}
const maxRowsPerCallback = 10000