prometheus/storage/metric/view.go

140 lines
5.3 KiB
Go
Raw Normal View History

2013-02-08 09:03:26 -08:00
// 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 (
"container/heap"
2013-02-08 09:03:26 -08:00
"time"
clientmodel "github.com/prometheus/client_golang/model"
2013-02-08 09:03:26 -08:00
)
var (
// firstSupertime is the smallest valid supertime that may be seeked to.
firstSupertime = []byte{0, 0, 0, 0, 0, 0, 0, 0}
// lastSupertime is the largest valid supertime that may be seeked to.
lastSupertime = []byte{127, 255, 255, 255, 255, 255, 255, 255}
)
// ViewRequestBuilder represents the summation of all datastore queries that
// shall be performed to extract values. Call the Get... methods to record the
// queries. Once done, use HasOp and PopOp to retrieve the resulting
// operations. The operations are sorted by their fingerprint (and, for equal
// fingerprints, by the StartsAt timestamp of their operation).
2013-02-08 09:03:26 -08:00
type ViewRequestBuilder interface {
// GetMetricAtTime records a query to get, for the given Fingerprint,
// either the value at that time if there is a match or the one or two
// values adjacent thereto.
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
2013-10-28 06:35:02 -07:00
GetMetricAtTime(fingerprint *clientmodel.Fingerprint, time clientmodel.Timestamp)
// GetMetricAtInterval records a query to get, for the given
// Fingerprint, either the value at that interval from From through
// Through if there is a match or the one or two values adjacent for
// each point.
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
2013-10-28 06:35:02 -07:00
GetMetricAtInterval(fingerprint *clientmodel.Fingerprint, from, through clientmodel.Timestamp, interval time.Duration)
// GetMetricRange records a query to get, for the given Fingerprint, the
// values that occur inclusively from From through Through.
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
2013-10-28 06:35:02 -07:00
GetMetricRange(fingerprint *clientmodel.Fingerprint, from, through clientmodel.Timestamp)
// GetMetricRangeAtInterval records a query to get value ranges at
// intervals for the given Fingerprint:
//
// |----| |----| |----| |----|
// ^ ^ ^ ^ ^ ^
// | \------------/ \----/ |
// from interval rangeDuration through
GetMetricRangeAtInterval(fp *clientmodel.Fingerprint, from, through clientmodel.Timestamp, interval, rangeDuration time.Duration)
// PopOp emits the next operation in the queue (sorted by
// fingerprint). If called while HasOps returns false, the
// behavior is undefined.
PopOp() op
// HasOp returns true if there is at least one more operation in the
// queue.
HasOp() bool
2013-02-08 09:03:26 -08:00
}
// viewRequestBuilder contains the various requests for data.
2013-02-08 09:03:26 -08:00
type viewRequestBuilder struct {
operations ops
2013-02-08 09:03:26 -08:00
}
// NewViewRequestBuilder furnishes a ViewRequestBuilder for remarking what types
// of queries to perform.
func NewViewRequestBuilder() *viewRequestBuilder {
return &viewRequestBuilder{}
2013-02-08 09:03:26 -08:00
}
var getValuesAtTimes = newValueAtTimeList(10 * 1024)
// GetMetricAtTime implements ViewRequestBuilder.
func (v *viewRequestBuilder) GetMetricAtTime(fp *clientmodel.Fingerprint, time clientmodel.Timestamp) {
heap.Push(&v.operations, getValuesAtTimes.Get(fp, time))
2013-02-08 09:03:26 -08:00
}
var getValuesAtIntervals = newValueAtIntervalList(10 * 1024)
// GetMetricAtInterval implements ViewRequestBuilder.
func (v *viewRequestBuilder) GetMetricAtInterval(fp *clientmodel.Fingerprint, from, through clientmodel.Timestamp, interval time.Duration) {
heap.Push(&v.operations, getValuesAtIntervals.Get(fp, from, through, interval))
2013-02-08 09:03:26 -08:00
}
var getValuesAlongRanges = newValueAlongRangeList(10 * 1024)
// GetMetricRange implements ViewRequestBuilder.
func (v *viewRequestBuilder) GetMetricRange(fp *clientmodel.Fingerprint, from, through clientmodel.Timestamp) {
heap.Push(&v.operations, getValuesAlongRanges.Get(fp, from, through))
2013-02-08 09:03:26 -08:00
}
var getValuesAtIntervalAlongRanges = newValueAtIntervalAlongRangeList(10 * 1024)
// GetMetricRangeAtInterval implements ViewRequestBuilder.
func (v *viewRequestBuilder) GetMetricRangeAtInterval(fp *clientmodel.Fingerprint, from, through clientmodel.Timestamp, interval, rangeDuration time.Duration) {
heap.Push(&v.operations, getValuesAtIntervalAlongRanges.Get(fp, from, through, interval, rangeDuration))
}
// PopOp implements ViewRequestBuilder.
func (v *viewRequestBuilder) PopOp() op {
return heap.Pop(&v.operations).(op)
}
2013-02-08 09:03:26 -08:00
// HasOp implements ViewRequestBuilder.
func (v *viewRequestBuilder) HasOp() bool {
return v.operations.Len() > 0
2013-02-08 09:03:26 -08:00
}
2013-03-06 17:16:39 -08:00
type view struct {
2013-05-16 07:02:07 -07:00
*memorySeriesStorage
2013-03-06 17:16:39 -08:00
}
func (v view) appendSamples(fingerprint *clientmodel.Fingerprint, samples Values) {
v.memorySeriesStorage.appendSamplesWithoutIndexing(fingerprint, samples)
2013-03-06 17:16:39 -08:00
}
func newView() view {
return view{NewMemorySeriesStorage(MemorySeriesOptions{})}
2013-03-06 17:16:39 -08:00
}
func giveBackOp(op interface{}) bool {
switch v := op.(type) {
case *getValuesAtTimeOp:
return getValuesAtTimes.Give(v)
case *getValuesAtIntervalOp:
return getValuesAtIntervals.Give(v)
case *getValuesAlongRangeOp:
return getValuesAlongRanges.Give(v)
case *getValueRangeAtIntervalOp:
return getValuesAtIntervalAlongRanges.Give(v)
default:
panic("unrecognized operation")
}
}