mirror of
https://github.com/VictoriaMetrics/VictoriaMetrics.git
synced 2024-11-21 14:44:00 +00:00
26242f526e
Tests showed that importing a single line with 70MB size takes 5.3GiB RSS memory for VictoriaMetrics single-node. In the scenario when user exports and imports data from one VM to another, it could possibly lead to OOM exception for destination VM. Importing a single line with 16MB size taks 1.3GiB RSS memory. Hence, the limit for `import.maxLineLen` was decreased from 100MB to 10MB to improve reliability of VictoriaMetrics during imports. Signed-off-by: hagen1778 <roman@victoriametrics.com> Co-authored-by: Aliaksandr Valialkin <valyala@victoriametrics.com>
192 lines
4.6 KiB
Go
192 lines
4.6 KiB
Go
package stream
|
|
|
|
import (
|
|
"bufio"
|
|
"fmt"
|
|
"io"
|
|
"sync"
|
|
|
|
"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
|
|
"github.com/VictoriaMetrics/VictoriaMetrics/lib/cgroup"
|
|
"github.com/VictoriaMetrics/VictoriaMetrics/lib/flagutil"
|
|
"github.com/VictoriaMetrics/VictoriaMetrics/lib/protoparser/common"
|
|
"github.com/VictoriaMetrics/VictoriaMetrics/lib/protoparser/vmimport"
|
|
"github.com/VictoriaMetrics/VictoriaMetrics/lib/writeconcurrencylimiter"
|
|
"github.com/VictoriaMetrics/metrics"
|
|
)
|
|
|
|
var maxLineLen = flagutil.NewBytes("import.maxLineLen", 10*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")
|
|
|
|
// Parse parses /api/v1/import lines from req and calls callback for the parsed rows.
|
|
//
|
|
// The callback can be called concurrently multiple times for streamed data from reader.
|
|
//
|
|
// callback shouldn't hold rows after returning.
|
|
func Parse(r io.Reader, isGzipped bool, callback func(rows []vmimport.Row) error) error {
|
|
wcr := writeconcurrencylimiter.GetReader(r)
|
|
defer writeconcurrencylimiter.PutReader(wcr)
|
|
r = wcr
|
|
|
|
if isGzipped {
|
|
zr, err := common.GetGzipReader(r)
|
|
if err != nil {
|
|
return fmt.Errorf("cannot read gzipped vmimport data: %w", err)
|
|
}
|
|
defer common.PutGzipReader(zr)
|
|
r = zr
|
|
}
|
|
ctx := getStreamContext(r)
|
|
defer putStreamContext(ctx)
|
|
for ctx.Read() {
|
|
uw := getUnmarshalWork()
|
|
uw.ctx = ctx
|
|
uw.callback = callback
|
|
uw.reqBuf, ctx.reqBuf = ctx.reqBuf, uw.reqBuf
|
|
ctx.wg.Add(1)
|
|
common.ScheduleUnmarshalWork(uw)
|
|
wcr.DecConcurrency()
|
|
}
|
|
ctx.wg.Wait()
|
|
if err := ctx.Error(); err != nil {
|
|
return err
|
|
}
|
|
return ctx.callbackErr
|
|
}
|
|
|
|
func (ctx *streamContext) Read() bool {
|
|
readCalls.Inc()
|
|
if ctx.err != nil || ctx.hasCallbackError() {
|
|
return false
|
|
}
|
|
ctx.reqBuf, ctx.tailBuf, ctx.err = common.ReadLinesBlockExt(ctx.br, ctx.reqBuf, ctx.tailBuf, maxLineLen.IntN())
|
|
if ctx.err != nil {
|
|
if ctx.err != io.EOF {
|
|
readErrors.Inc()
|
|
ctx.err = fmt.Errorf("cannot read vmimport data: %w", ctx.err)
|
|
}
|
|
return false
|
|
}
|
|
return true
|
|
}
|
|
|
|
var (
|
|
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"}`)
|
|
)
|
|
|
|
type streamContext struct {
|
|
br *bufio.Reader
|
|
reqBuf []byte
|
|
tailBuf []byte
|
|
err error
|
|
|
|
wg sync.WaitGroup
|
|
callbackErrLock sync.Mutex
|
|
callbackErr error
|
|
}
|
|
|
|
func (ctx *streamContext) Error() error {
|
|
if ctx.err == io.EOF {
|
|
return nil
|
|
}
|
|
return ctx.err
|
|
}
|
|
|
|
func (ctx *streamContext) hasCallbackError() bool {
|
|
ctx.callbackErrLock.Lock()
|
|
ok := ctx.callbackErr != nil
|
|
ctx.callbackErrLock.Unlock()
|
|
return ok
|
|
}
|
|
|
|
func (ctx *streamContext) reset() {
|
|
ctx.br.Reset(nil)
|
|
ctx.reqBuf = ctx.reqBuf[:0]
|
|
ctx.tailBuf = ctx.tailBuf[:0]
|
|
ctx.err = nil
|
|
ctx.callbackErr = nil
|
|
}
|
|
|
|
func getStreamContext(r io.Reader) *streamContext {
|
|
select {
|
|
case ctx := <-streamContextPoolCh:
|
|
ctx.br.Reset(r)
|
|
return ctx
|
|
default:
|
|
if v := streamContextPool.Get(); v != nil {
|
|
ctx := v.(*streamContext)
|
|
ctx.br.Reset(r)
|
|
return ctx
|
|
}
|
|
return &streamContext{
|
|
br: bufio.NewReaderSize(r, 64*1024),
|
|
}
|
|
}
|
|
}
|
|
|
|
func putStreamContext(ctx *streamContext) {
|
|
ctx.reset()
|
|
select {
|
|
case streamContextPoolCh <- ctx:
|
|
default:
|
|
streamContextPool.Put(ctx)
|
|
}
|
|
}
|
|
|
|
var streamContextPool sync.Pool
|
|
var streamContextPoolCh = make(chan *streamContext, cgroup.AvailableCPUs())
|
|
|
|
type unmarshalWork struct {
|
|
rows vmimport.Rows
|
|
ctx *streamContext
|
|
callback func(rows []vmimport.Row) error
|
|
reqBuf []byte
|
|
}
|
|
|
|
func (uw *unmarshalWork) reset() {
|
|
uw.rows.Reset()
|
|
uw.ctx = nil
|
|
uw.callback = nil
|
|
uw.reqBuf = uw.reqBuf[:0]
|
|
}
|
|
|
|
func (uw *unmarshalWork) runCallback(rows []vmimport.Row) {
|
|
ctx := uw.ctx
|
|
if err := uw.callback(rows); err != nil {
|
|
ctx.callbackErrLock.Lock()
|
|
if ctx.callbackErr == nil {
|
|
ctx.callbackErr = fmt.Errorf("error when processing imported data: %w", err)
|
|
}
|
|
ctx.callbackErrLock.Unlock()
|
|
}
|
|
ctx.wg.Done()
|
|
}
|
|
|
|
// Unmarshal implements common.UnmarshalWork
|
|
func (uw *unmarshalWork) Unmarshal() {
|
|
uw.rows.Unmarshal(bytesutil.ToUnsafeString(uw.reqBuf))
|
|
rows := uw.rows.Rows
|
|
for i := range rows {
|
|
row := &rows[i]
|
|
rowsRead.Add(len(row.Timestamps))
|
|
}
|
|
uw.runCallback(rows)
|
|
putUnmarshalWork(uw)
|
|
}
|
|
|
|
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
|