prometheus/storage/metric/processor.go
Julius Volz 740d448983 Use custom timestamp type for sample timestamps and related code.
So far we've been using Go's native time.Time for anything related to sample
timestamps. Since the range of time.Time is much bigger than what we need, this
has created two problems:

- there could be time.Time values which were out of the range/precision of the
  time type that we persist to disk, therefore causing incorrectly ordered keys.
  One bug caused by this was:

  https://github.com/prometheus/prometheus/issues/367

  It would be good to use a timestamp type that's more closely aligned with
  what the underlying storage supports.

- sizeof(time.Time) is 192, while Prometheus should be ok with a single 64-bit
  Unix timestamp (possibly even a 32-bit one). Since we store samples in large
  numbers, this seriously affects memory usage. Furthermore, copying/working
  with the data will be faster if it's smaller.

*MEMORY USAGE RESULTS*
Initial memory usage comparisons for a running Prometheus with 1 timeseries and
100,000 samples show roughly a 13% decrease in total (VIRT) memory usage. In my
tests, this advantage for some reason decreased a bit the more samples the
timeseries had (to 5-7% for millions of samples). This I can't fully explain,
but perhaps garbage collection issues were involved.

*WHEN TO USE THE NEW TIMESTAMP TYPE*
The new clientmodel.Timestamp type should be used whenever time
calculations are either directly or indirectly related to sample
timestamps.

For example:
- the timestamp of a sample itself
- all kinds of watermarks
- anything that may become or is compared to a sample timestamp (like the timestamp
  passed into Target.Scrape()).

When to still use time.Time:
- for measuring durations/times not related to sample timestamps, like duration
  telemetry exporting, timers that indicate how frequently to execute some
  action, etc.

*NOTE ON OPERATOR OPTIMIZATION TESTS*
We don't use operator optimization code anymore, but it still lives in
the code as dead code. It still has tests, but I couldn't get all of them to
pass with the new timestamp format. I commented out the failing cases for now,
but we should probably remove the dead code soon. I just didn't want to do that
in the same change as this.

Change-Id: I821787414b0debe85c9fffaeb57abd453727af0f
2013-12-03 09:11:28 +01:00

442 lines
13 KiB
Go

// Copyright 2013 Prometheus Team
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package metric
import (
"fmt"
"code.google.com/p/goprotobuf/proto"
clientmodel "github.com/prometheus/client_golang/model"
"github.com/prometheus/prometheus/storage/raw"
"github.com/prometheus/prometheus/storage/raw/leveldb"
dto "github.com/prometheus/prometheus/model/generated"
)
// processor models a post-processing agent that performs work given a sample
// corpus.
type Processor interface {
// Name emits the name of this processor's signature encoder. It must be
// fully-qualified in the sense that it could be used via a Protocol Buffer
// registry to extract the descriptor to reassemble this message.
Name() string
// Signature emits a byte signature for this process for the purpose of
// remarking how far along it has been applied to the database.
Signature() []byte
// Apply runs this processor against the sample set. sampleIterator expects
// to be pre-seeked to the initial starting position. The processor will
// run until up until stopAt has been reached. It is imperative that the
// provided stopAt is within the interval of the series frontier.
//
// Upon completion or error, the last time at which the processor finished
// shall be emitted in addition to any errors.
Apply(sampleIterator leveldb.Iterator, samplesPersistence raw.Persistence, stopAt clientmodel.Timestamp, fingerprint *clientmodel.Fingerprint) (lastCurated clientmodel.Timestamp, err error)
}
// CompactionProcessor combines sparse values in the database together such
// that at least MinimumGroupSize-sized chunks are grouped together.
type CompactionProcessor struct {
maximumMutationPoolBatch int
minimumGroupSize int
// signature is the byte representation of the CompactionProcessor's settings,
// used for purely memoization purposes across an instance.
signature []byte
dtoSampleKeys *dtoSampleKeyList
sampleKeys *sampleKeyList
}
func (p *CompactionProcessor) Name() string {
return "io.prometheus.CompactionProcessorDefinition"
}
func (p *CompactionProcessor) Signature() []byte {
if len(p.signature) == 0 {
out, err := proto.Marshal(&dto.CompactionProcessorDefinition{
MinimumGroupSize: proto.Uint32(uint32(p.minimumGroupSize)),
})
if err != nil {
panic(err)
}
p.signature = out
}
return p.signature
}
func (p *CompactionProcessor) String() string {
return fmt.Sprintf("compactionProcessor for minimum group size %d", p.minimumGroupSize)
}
func (p *CompactionProcessor) Apply(sampleIterator leveldb.Iterator, samplesPersistence raw.Persistence, stopAt clientmodel.Timestamp, fingerprint *clientmodel.Fingerprint) (lastCurated clientmodel.Timestamp, err error) {
var pendingBatch raw.Batch = nil
defer func() {
if pendingBatch != nil {
pendingBatch.Close()
}
}()
var pendingMutations = 0
var pendingSamples Values
var unactedSamples Values
var lastTouchedTime clientmodel.Timestamp
var keyDropped bool
sampleKey, _ := p.sampleKeys.Get()
defer p.sampleKeys.Give(sampleKey)
sampleKeyDto, _ := p.dtoSampleKeys.Get()
defer p.dtoSampleKeys.Give(sampleKeyDto)
if err = sampleIterator.Key(sampleKeyDto); err != nil {
return
}
sampleKey.Load(sampleKeyDto)
unactedSamples, err = extractSampleValues(sampleIterator)
if err != nil {
return
}
for lastCurated.Before(stopAt) && lastTouchedTime.Before(stopAt) && sampleKey.Fingerprint.Equal(fingerprint) {
switch {
// Furnish a new pending batch operation if none is available.
case pendingBatch == nil:
pendingBatch = leveldb.NewBatch()
// If there are no sample values to extract from the datastore, let's
// continue extracting more values to use. We know that the time.Before()
// block would prevent us from going into unsafe territory.
case len(unactedSamples) == 0:
if !sampleIterator.Next() {
return lastCurated, fmt.Errorf("Illegal Condition: Invalid Iterator on Continuation")
}
keyDropped = false
if err = sampleIterator.Key(sampleKeyDto); err != nil {
return
}
sampleKey.Load(sampleKeyDto)
if !sampleKey.Fingerprint.Equal(fingerprint) {
break
}
unactedSamples, err = extractSampleValues(sampleIterator)
if err != nil {
return
}
// If the number of pending mutations exceeds the allowed batch amount,
// commit to disk and delete the batch. A new one will be recreated if
// necessary.
case pendingMutations >= p.maximumMutationPoolBatch:
err = samplesPersistence.Commit(pendingBatch)
if err != nil {
return
}
pendingMutations = 0
pendingBatch.Close()
pendingBatch = nil
case len(pendingSamples) == 0 && len(unactedSamples) >= p.minimumGroupSize:
lastTouchedTime = unactedSamples[len(unactedSamples)-1].Timestamp
unactedSamples = Values{}
case len(pendingSamples)+len(unactedSamples) < p.minimumGroupSize:
if !keyDropped {
k := new(dto.SampleKey)
sampleKey.Dump(k)
pendingBatch.Drop(k)
keyDropped = true
}
pendingSamples = append(pendingSamples, unactedSamples...)
lastTouchedTime = unactedSamples[len(unactedSamples)-1].Timestamp
unactedSamples = Values{}
pendingMutations++
// If the number of pending writes equals the target group size
case len(pendingSamples) == p.minimumGroupSize:
k := new(dto.SampleKey)
newSampleKey := pendingSamples.ToSampleKey(fingerprint)
newSampleKey.Dump(k)
b := new(dto.SampleValueSeries)
pendingSamples.dump(b)
pendingBatch.Put(k, b)
pendingMutations++
lastCurated = newSampleKey.FirstTimestamp
if len(unactedSamples) > 0 {
if !keyDropped {
sampleKey.Dump(k)
pendingBatch.Drop(k)
keyDropped = true
}
if len(unactedSamples) > p.minimumGroupSize {
pendingSamples = unactedSamples[:p.minimumGroupSize]
unactedSamples = unactedSamples[p.minimumGroupSize:]
lastTouchedTime = unactedSamples[len(unactedSamples)-1].Timestamp
} else {
pendingSamples = unactedSamples
lastTouchedTime = pendingSamples[len(pendingSamples)-1].Timestamp
unactedSamples = Values{}
}
}
case len(pendingSamples)+len(unactedSamples) >= p.minimumGroupSize:
if !keyDropped {
k := new(dto.SampleKey)
sampleKey.Dump(k)
pendingBatch.Drop(k)
keyDropped = true
}
remainder := p.minimumGroupSize - len(pendingSamples)
pendingSamples = append(pendingSamples, unactedSamples[:remainder]...)
unactedSamples = unactedSamples[remainder:]
if len(unactedSamples) == 0 {
lastTouchedTime = pendingSamples[len(pendingSamples)-1].Timestamp
} else {
lastTouchedTime = unactedSamples[len(unactedSamples)-1].Timestamp
}
pendingMutations++
default:
err = fmt.Errorf("Unhandled processing case.")
}
}
if len(unactedSamples) > 0 || len(pendingSamples) > 0 {
pendingSamples = append(pendingSamples, unactedSamples...)
k := new(dto.SampleKey)
newSampleKey := pendingSamples.ToSampleKey(fingerprint)
newSampleKey.Dump(k)
b := new(dto.SampleValueSeries)
pendingSamples.dump(b)
pendingBatch.Put(k, b)
pendingSamples = Values{}
pendingMutations++
lastCurated = newSampleKey.FirstTimestamp
}
// This is not deferred due to the off-chance that a pre-existing commit
// failed.
if pendingBatch != nil && pendingMutations > 0 {
err = samplesPersistence.Commit(pendingBatch)
if err != nil {
return
}
}
return
}
func (p *CompactionProcessor) Close() {
p.dtoSampleKeys.Close()
p.sampleKeys.Close()
}
type CompactionProcessorOptions struct {
// MaximumMutationPoolBatch represents approximately the largest pending
// batch of mutation operations for the database before pausing to
// commit before resumption.
//
// A reasonable value would be (MinimumGroupSize * 2) + 1.
MaximumMutationPoolBatch int
// MinimumGroupSize represents the smallest allowed sample chunk size in the
// database.
MinimumGroupSize int
}
func NewCompactionProcessor(o *CompactionProcessorOptions) *CompactionProcessor {
return &CompactionProcessor{
maximumMutationPoolBatch: o.MaximumMutationPoolBatch,
minimumGroupSize: o.MinimumGroupSize,
dtoSampleKeys: newDtoSampleKeyList(10),
sampleKeys: newSampleKeyList(10),
}
}
// DeletionProcessor deletes sample blocks older than a defined value.
type DeletionProcessor struct {
maximumMutationPoolBatch int
// signature is the byte representation of the DeletionProcessor's settings,
// used for purely memoization purposes across an instance.
signature []byte
dtoSampleKeys *dtoSampleKeyList
sampleKeys *sampleKeyList
}
func (p *DeletionProcessor) Name() string {
return "io.prometheus.DeletionProcessorDefinition"
}
func (p *DeletionProcessor) Signature() []byte {
if len(p.signature) == 0 {
out, err := proto.Marshal(&dto.DeletionProcessorDefinition{})
if err != nil {
panic(err)
}
p.signature = out
}
return p.signature
}
func (p *DeletionProcessor) String() string {
return "deletionProcessor"
}
func (p *DeletionProcessor) Apply(sampleIterator leveldb.Iterator, samplesPersistence raw.Persistence, stopAt clientmodel.Timestamp, fingerprint *clientmodel.Fingerprint) (lastCurated clientmodel.Timestamp, err error) {
var pendingBatch raw.Batch = nil
defer func() {
if pendingBatch != nil {
pendingBatch.Close()
}
}()
sampleKeyDto, _ := p.dtoSampleKeys.Get()
defer p.dtoSampleKeys.Give(sampleKeyDto)
sampleKey, _ := p.sampleKeys.Get()
defer p.sampleKeys.Give(sampleKey)
if err = sampleIterator.Key(sampleKeyDto); err != nil {
return
}
sampleKey.Load(sampleKeyDto)
sampleValues, err := extractSampleValues(sampleIterator)
if err != nil {
return
}
pendingMutations := 0
for lastCurated.Before(stopAt) && sampleKey.Fingerprint.Equal(fingerprint) {
switch {
// Furnish a new pending batch operation if none is available.
case pendingBatch == nil:
pendingBatch = leveldb.NewBatch()
// If there are no sample values to extract from the datastore, let's
// continue extracting more values to use. We know that the time.Before()
// block would prevent us from going into unsafe territory.
case len(sampleValues) == 0:
if !sampleIterator.Next() {
return lastCurated, fmt.Errorf("Illegal Condition: Invalid Iterator on Continuation")
}
if err = sampleIterator.Key(sampleKeyDto); err != nil {
return
}
sampleKey.Load(sampleKeyDto)
sampleValues, err = extractSampleValues(sampleIterator)
if err != nil {
return
}
// If the number of pending mutations exceeds the allowed batch amount,
// commit to disk and delete the batch. A new one will be recreated if
// necessary.
case pendingMutations >= p.maximumMutationPoolBatch:
err = samplesPersistence.Commit(pendingBatch)
if err != nil {
return
}
pendingMutations = 0
pendingBatch.Close()
pendingBatch = nil
case !sampleKey.MayContain(stopAt):
k := &dto.SampleKey{}
sampleKey.Dump(k)
pendingBatch.Drop(k)
lastCurated = sampleKey.LastTimestamp
sampleValues = Values{}
pendingMutations++
case sampleKey.MayContain(stopAt):
k := &dto.SampleKey{}
sampleKey.Dump(k)
pendingBatch.Drop(k)
pendingMutations++
sampleValues = sampleValues.TruncateBefore(stopAt)
if len(sampleValues) > 0 {
k := &dto.SampleKey{}
sampleKey = sampleValues.ToSampleKey(fingerprint)
sampleKey.Dump(k)
v := &dto.SampleValueSeries{}
sampleValues.dump(v)
lastCurated = sampleKey.FirstTimestamp
pendingBatch.Put(k, v)
pendingMutations++
} else {
lastCurated = sampleKey.LastTimestamp
}
default:
err = fmt.Errorf("Unhandled processing case.")
}
}
// This is not deferred due to the off-chance that a pre-existing commit
// failed.
if pendingBatch != nil && pendingMutations > 0 {
err = samplesPersistence.Commit(pendingBatch)
if err != nil {
return
}
}
return
}
func (p *DeletionProcessor) Close() {
p.dtoSampleKeys.Close()
p.sampleKeys.Close()
}
type DeletionProcessorOptions struct {
// MaximumMutationPoolBatch represents approximately the largest pending
// batch of mutation operations for the database before pausing to
// commit before resumption.
MaximumMutationPoolBatch int
}
func NewDeletionProcessor(o *DeletionProcessorOptions) *DeletionProcessor {
return &DeletionProcessor{
maximumMutationPoolBatch: o.MaximumMutationPoolBatch,
dtoSampleKeys: newDtoSampleKeyList(10),
sampleKeys: newSampleKeyList(10),
}
}