mudhorn/prometheus
1
0
Fork 0
mirror of https://github.com/prometheus/prometheus.git synced 2024-09-19 23:37:31 -07:00
Code Issues Actions 7 Packages Projects Releases Wiki Activity

Fixed iterator regression: Avoid using heap for each sample when iterating.

Browse source 
Fixes: https://github.com/prometheus/prometheus/issues/7873

Signed-off-by: Bartlomiej Plotka <bwplotka@gmail.com>
This commit is contained in:
Julien Pivotto 2020-08-31 14:00:46 +02:00 committed by Bartlomiej Plotka
parent a5366cb236
commit 2c8b2c5915
1 changed files with 26 additions and 25 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

49
storage/merge.go
View file

@ -16,6 +16,7 @@ package storage
import ( import (
"bytes" "bytes"
"container/heap" "container/heap"
"fmt"
"sort" "sort"
"strings" "strings"
"sync" "sync"
@ -418,8 +419,7 @@ func (h *genericSeriesSetHeap) Pop() interface{} {
// with "almost" the same data, e.g. from 2 Prometheus HA replicas. This is fine, since from the Prometheus perspective // with "almost" the same data, e.g. from 2 Prometheus HA replicas. This is fine, since from the Prometheus perspective
// this never happens. // this never happens.
// //
// NOTE: Use this merge function only when you see potentially overlapping series, as this introduces a small overhead // It's optimized for non-overlap cases as well.
// to handle overlaps between series.
func ChainedSeriesMerge(series ...Series) Series { func ChainedSeriesMerge(series ...Series) Series {
if len(series) == 0 { if len(series) == 0 {
return nil return nil
@ -438,10 +438,12 @@ func ChainedSeriesMerge(series ...Series) Series {
// chainSampleIterator is responsible to iterate over samples from different iterators of the same time series in timestamps // chainSampleIterator is responsible to iterate over samples from different iterators of the same time series in timestamps
// order. If one or more samples overlap, one sample from random overlapped ones is kept and all others with the same // order. If one or more samples overlap, one sample from random overlapped ones is kept and all others with the same
// timestamp are dropped. // timestamp are dropped. It's optimized for non-overlap cases as well.
type chainSampleIterator struct { type chainSampleIterator struct {
iterators []chunkenc.Iterator iterators []chunkenc.Iterator
h samplesIteratorHeap h samplesIteratorHeap
curr chunkenc.Iterator
} }
func newChainSampleIterator(iterators []chunkenc.Iterator) chunkenc.Iterator { func newChainSampleIterator(iterators []chunkenc.Iterator) chunkenc.Iterator {
@ -462,45 +464,44 @@ func (c *chainSampleIterator) Seek(t int64) bool {
} }
func (c *chainSampleIterator) At() (t int64, v float64) { func (c *chainSampleIterator) At() (t int64, v float64) {
if len(c.h) == 0 { if c.h == nil {
panic("chainSampleIterator.At() called after .Next() returned false.") panic("chainSampleIterator.At() called after .Next() returned false.")
} }
return c.curr.At()
return c.h[0].At()
} }
func (c *chainSampleIterator) Next() bool { func (c *chainSampleIterator) Next() bool {
if c.h == nil { if c.h == nil {
for _, iter := range c.iterators { c.curr = c.iterators[0]
for _, iter := range c.iterators[1:] {
if iter.Next() { if iter.Next() {
heap.Push(&c.h, iter) heap.Push(&c.h, iter)
} }
} }
} else if len(c.h) == 0 {
return len(c.h) > 0 return c.curr != nil && c.curr.Next()
} }
if len(c.h) == 0 { for {
return false if c.curr.Next() {
} currt, _ := c.curr.At()
currt, _ := c.At()
for len(c.h) > 0 {
nextt, _ := c.h[0].At() nextt, _ := c.h[0].At()
// All but one of the overlapping samples will be dropped. if currt < nextt {
if nextt != currt { return true
break }
if currt == nextt {
fmt.Println("same ts", currt, nextt)
// Ignoring sample.
continue
}
heap.Push(&c.h, c.curr)
} }
iter := heap.Pop(&c.h).(chunkenc.Iterator) c.curr = heap.Pop(&c.h).(chunkenc.Iterator)
if iter.Next() { return true
heap.Push(&c.h, iter)
} }
} }
return len(c.h) > 0
}
func (c *chainSampleIterator) Err() error { func (c *chainSampleIterator) Err() error {
var errs tsdb_errors.MultiError var errs tsdb_errors.MultiError
for _, iter := range c.iterators { for _, iter := range c.iterators {