2013-02-07 02:49:04 -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.
|
|
|
|
|
2013-01-07 14:24:26 -08:00
|
|
|
package ast
|
|
|
|
|
|
|
|
import (
|
|
|
|
"fmt"
|
2013-09-17 05:59:31 -07:00
|
|
|
"math"
|
2013-04-10 01:44:13 -07:00
|
|
|
"sort"
|
2013-01-07 14:24:26 -08:00
|
|
|
"time"
|
2013-06-25 05:02:27 -07:00
|
|
|
|
|
|
|
clientmodel "github.com/prometheus/client_golang/model"
|
|
|
|
|
|
|
|
"github.com/prometheus/prometheus/utility"
|
2013-01-07 14:24:26 -08:00
|
|
|
)
|
|
|
|
|
2014-02-13 09:48:56 -08:00
|
|
|
// Function represents a function of the expression language and is
|
|
|
|
// used by function nodes.
|
2013-01-07 14:24:26 -08:00
|
|
|
type Function struct {
|
|
|
|
name string
|
|
|
|
argTypes []ExprType
|
|
|
|
returnType ExprType
|
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
|
|
|
callFn func(timestamp clientmodel.Timestamp, view *viewAdapter, args []Node) interface{}
|
2013-01-07 14:24:26 -08:00
|
|
|
}
|
|
|
|
|
2014-02-13 09:48:56 -08:00
|
|
|
// CheckArgTypes returns a non-nil error if the number or types of
|
|
|
|
// passed in arg nodes do not match the function's expectations.
|
2013-01-07 14:24:26 -08:00
|
|
|
func (function *Function) CheckArgTypes(args []Node) error {
|
|
|
|
if len(function.argTypes) != len(args) {
|
2014-02-13 09:48:56 -08:00
|
|
|
return fmt.Errorf(
|
|
|
|
"wrong number of arguments to function %v(): %v expected, %v given",
|
|
|
|
function.name, len(function.argTypes), len(args),
|
|
|
|
)
|
2013-01-07 14:24:26 -08:00
|
|
|
}
|
|
|
|
for idx, argType := range function.argTypes {
|
|
|
|
invalidType := false
|
|
|
|
var expectedType string
|
|
|
|
if _, ok := args[idx].(ScalarNode); argType == SCALAR && !ok {
|
|
|
|
invalidType = true
|
|
|
|
expectedType = "scalar"
|
|
|
|
}
|
|
|
|
if _, ok := args[idx].(VectorNode); argType == VECTOR && !ok {
|
|
|
|
invalidType = true
|
|
|
|
expectedType = "vector"
|
|
|
|
}
|
|
|
|
if _, ok := args[idx].(MatrixNode); argType == MATRIX && !ok {
|
|
|
|
invalidType = true
|
|
|
|
expectedType = "matrix"
|
|
|
|
}
|
|
|
|
if _, ok := args[idx].(StringNode); argType == STRING && !ok {
|
|
|
|
invalidType = true
|
|
|
|
expectedType = "string"
|
|
|
|
}
|
|
|
|
|
|
|
|
if invalidType {
|
2014-02-13 09:48:56 -08:00
|
|
|
return fmt.Errorf(
|
|
|
|
"wrong type for argument %v in function %v(), expected %v",
|
|
|
|
idx, function.name, expectedType,
|
|
|
|
)
|
2013-01-07 14:24:26 -08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
2013-06-25 05:02:27 -07:00
|
|
|
// === time() clientmodel.SampleValue ===
|
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
|
|
|
func timeImpl(timestamp clientmodel.Timestamp, view *viewAdapter, args []Node) interface{} {
|
2013-06-25 05:02:27 -07:00
|
|
|
return clientmodel.SampleValue(time.Now().Unix())
|
2013-01-07 14:24:26 -08:00
|
|
|
}
|
|
|
|
|
2013-04-10 01:44:13 -07:00
|
|
|
// === delta(matrix MatrixNode, isCounter ScalarNode) Vector ===
|
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
|
|
|
func deltaImpl(timestamp clientmodel.Timestamp, view *viewAdapter, args []Node) interface{} {
|
2013-01-07 14:24:26 -08:00
|
|
|
matrixNode := args[0].(MatrixNode)
|
2013-05-28 08:36:53 -07:00
|
|
|
isCounter := args[1].(ScalarNode).Eval(timestamp, view) > 0
|
2013-01-07 14:24:26 -08:00
|
|
|
resultVector := Vector{}
|
|
|
|
|
|
|
|
// If we treat these metrics as counters, we need to fetch all values
|
|
|
|
// in the interval to find breaks in the timeseries' monotonicity.
|
|
|
|
// I.e. if a counter resets, we want to ignore that reset.
|
|
|
|
var matrixValue Matrix
|
2013-05-28 08:36:53 -07:00
|
|
|
if isCounter {
|
2013-03-21 10:06:15 -07:00
|
|
|
matrixValue = matrixNode.Eval(timestamp, view)
|
2013-01-07 14:24:26 -08:00
|
|
|
} else {
|
2013-03-21 10:06:15 -07:00
|
|
|
matrixValue = matrixNode.EvalBoundaries(timestamp, view)
|
2013-01-07 14:24:26 -08:00
|
|
|
}
|
|
|
|
for _, samples := range matrixValue {
|
2013-05-27 02:10:40 -07:00
|
|
|
// No sense in trying to compute a delta without at least two points. Drop
|
|
|
|
// this vector element.
|
|
|
|
if len(samples.Values) < 2 {
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
|
2013-06-25 05:02:27 -07:00
|
|
|
counterCorrection := clientmodel.SampleValue(0)
|
|
|
|
lastValue := clientmodel.SampleValue(0)
|
2013-01-07 14:24:26 -08:00
|
|
|
for _, sample := range samples.Values {
|
|
|
|
currentValue := sample.Value
|
2013-05-28 08:36:53 -07:00
|
|
|
if isCounter && currentValue < lastValue {
|
2013-01-07 14:24:26 -08:00
|
|
|
counterCorrection += lastValue - currentValue
|
|
|
|
}
|
|
|
|
lastValue = currentValue
|
|
|
|
}
|
|
|
|
resultValue := lastValue - samples.Values[0].Value + counterCorrection
|
2013-04-12 17:45:58 -07:00
|
|
|
|
2014-02-22 13:29:32 -08:00
|
|
|
targetInterval := args[0].(*MatrixSelector).interval
|
2013-04-12 17:45:58 -07:00
|
|
|
sampledInterval := samples.Values[len(samples.Values)-1].Timestamp.Sub(samples.Values[0].Timestamp)
|
|
|
|
if sampledInterval == 0 {
|
|
|
|
// Only found one sample. Cannot compute a rate from this.
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
// Correct for differences in target vs. actual delta interval.
|
|
|
|
//
|
|
|
|
// Above, we didn't actually calculate the delta for the specified target
|
|
|
|
// interval, but for an interval between the first and last found samples
|
|
|
|
// under the target interval, which will usually have less time between
|
|
|
|
// them. Depending on how many samples are found under a target interval,
|
|
|
|
// the delta results are distorted and temporal aliasing occurs (ugly
|
|
|
|
// bumps). This effect is corrected for below.
|
2013-06-25 05:02:27 -07:00
|
|
|
intervalCorrection := clientmodel.SampleValue(targetInterval) / clientmodel.SampleValue(sampledInterval)
|
2013-04-12 17:45:58 -07:00
|
|
|
resultValue *= intervalCorrection
|
|
|
|
|
2013-06-25 05:02:27 -07:00
|
|
|
resultSample := &clientmodel.Sample{
|
2013-01-07 14:24:26 -08:00
|
|
|
Metric: samples.Metric,
|
|
|
|
Value: resultValue,
|
2013-03-28 09:05:06 -07:00
|
|
|
Timestamp: timestamp,
|
2013-01-07 14:24:26 -08:00
|
|
|
}
|
|
|
|
resultVector = append(resultVector, resultSample)
|
|
|
|
}
|
|
|
|
return resultVector
|
|
|
|
}
|
|
|
|
|
2013-04-10 01:44:13 -07:00
|
|
|
// === rate(node *MatrixNode) Vector ===
|
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
|
|
|
func rateImpl(timestamp clientmodel.Timestamp, view *viewAdapter, args []Node) interface{} {
|
2013-01-07 14:24:26 -08:00
|
|
|
args = append(args, &ScalarLiteral{value: 1})
|
2013-03-21 10:06:15 -07:00
|
|
|
vector := deltaImpl(timestamp, view, args).(Vector)
|
2013-01-21 16:49:00 -08:00
|
|
|
|
|
|
|
// TODO: could be other type of MatrixNode in the future (right now, only
|
2014-02-22 13:29:32 -08:00
|
|
|
// MatrixSelector exists). Find a better way of getting the duration of a
|
2013-01-21 16:49:00 -08:00
|
|
|
// matrix, such as looking at the samples themselves.
|
2014-02-22 13:29:32 -08:00
|
|
|
interval := args[0].(*MatrixSelector).interval
|
2013-04-16 08:23:59 -07:00
|
|
|
for i := range vector {
|
2013-06-25 05:02:27 -07:00
|
|
|
vector[i].Value /= clientmodel.SampleValue(interval / time.Second)
|
2013-01-21 16:49:00 -08:00
|
|
|
}
|
|
|
|
return vector
|
2013-01-07 14:24:26 -08:00
|
|
|
}
|
|
|
|
|
2013-04-10 01:44:13 -07:00
|
|
|
type vectorByValueSorter struct {
|
|
|
|
vector Vector
|
|
|
|
}
|
|
|
|
|
|
|
|
func (sorter vectorByValueSorter) Len() int {
|
|
|
|
return len(sorter.vector)
|
|
|
|
}
|
|
|
|
|
2013-05-15 22:38:31 -07:00
|
|
|
func (sorter vectorByValueSorter) Less(i, j int) bool {
|
2013-04-10 01:44:13 -07:00
|
|
|
return sorter.vector[i].Value < sorter.vector[j].Value
|
|
|
|
}
|
|
|
|
|
|
|
|
func (sorter vectorByValueSorter) Swap(i, j int) {
|
|
|
|
sorter.vector[i], sorter.vector[j] = sorter.vector[j], sorter.vector[i]
|
|
|
|
}
|
|
|
|
|
|
|
|
// === sort(node *VectorNode) Vector ===
|
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
|
|
|
func sortImpl(timestamp clientmodel.Timestamp, view *viewAdapter, args []Node) interface{} {
|
2013-04-10 01:44:13 -07:00
|
|
|
byValueSorter := vectorByValueSorter{
|
|
|
|
vector: args[0].(VectorNode).Eval(timestamp, view),
|
|
|
|
}
|
|
|
|
sort.Sort(byValueSorter)
|
|
|
|
return byValueSorter.vector
|
|
|
|
}
|
|
|
|
|
|
|
|
// === sortDesc(node *VectorNode) Vector ===
|
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
|
|
|
func sortDescImpl(timestamp clientmodel.Timestamp, view *viewAdapter, args []Node) interface{} {
|
2013-04-10 01:44:13 -07:00
|
|
|
descByValueSorter := utility.ReverseSorter{
|
2014-02-13 09:48:56 -08:00
|
|
|
Interface: vectorByValueSorter{
|
2013-04-10 01:44:13 -07:00
|
|
|
vector: args[0].(VectorNode).Eval(timestamp, view),
|
|
|
|
},
|
|
|
|
}
|
|
|
|
sort.Sort(descByValueSorter)
|
|
|
|
return descByValueSorter.Interface.(vectorByValueSorter).vector
|
|
|
|
}
|
|
|
|
|
|
|
|
// === sampleVectorImpl() Vector ===
|
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
|
|
|
func sampleVectorImpl(timestamp clientmodel.Timestamp, view *viewAdapter, args []Node) interface{} {
|
2013-01-07 14:24:26 -08:00
|
|
|
return Vector{
|
2013-06-25 05:02:27 -07:00
|
|
|
&clientmodel.Sample{
|
|
|
|
Metric: clientmodel.Metric{
|
|
|
|
clientmodel.MetricNameLabel: "http_requests",
|
|
|
|
clientmodel.JobLabel: "api-server",
|
|
|
|
"instance": "0",
|
2013-01-07 14:24:26 -08:00
|
|
|
},
|
|
|
|
Value: 10,
|
2013-03-28 09:05:06 -07:00
|
|
|
Timestamp: timestamp,
|
2013-01-07 14:24:26 -08:00
|
|
|
},
|
2013-06-25 05:02:27 -07:00
|
|
|
&clientmodel.Sample{
|
|
|
|
Metric: clientmodel.Metric{
|
|
|
|
clientmodel.MetricNameLabel: "http_requests",
|
|
|
|
clientmodel.JobLabel: "api-server",
|
|
|
|
"instance": "1",
|
2013-01-07 14:24:26 -08:00
|
|
|
},
|
|
|
|
Value: 20,
|
2013-03-28 09:05:06 -07:00
|
|
|
Timestamp: timestamp,
|
2013-01-07 14:24:26 -08:00
|
|
|
},
|
2013-06-25 05:02:27 -07:00
|
|
|
&clientmodel.Sample{
|
|
|
|
Metric: clientmodel.Metric{
|
|
|
|
clientmodel.MetricNameLabel: "http_requests",
|
|
|
|
clientmodel.JobLabel: "api-server",
|
|
|
|
"instance": "2",
|
2013-01-07 14:24:26 -08:00
|
|
|
},
|
|
|
|
Value: 30,
|
2013-03-28 09:05:06 -07:00
|
|
|
Timestamp: timestamp,
|
2013-01-07 14:24:26 -08:00
|
|
|
},
|
2013-06-25 05:02:27 -07:00
|
|
|
&clientmodel.Sample{
|
|
|
|
Metric: clientmodel.Metric{
|
|
|
|
clientmodel.MetricNameLabel: "http_requests",
|
|
|
|
clientmodel.JobLabel: "api-server",
|
|
|
|
"instance": "3",
|
|
|
|
"group": "canary",
|
2013-01-07 14:24:26 -08:00
|
|
|
},
|
|
|
|
Value: 40,
|
2013-03-28 09:05:06 -07:00
|
|
|
Timestamp: timestamp,
|
2013-01-07 14:24:26 -08:00
|
|
|
},
|
2013-06-25 05:02:27 -07:00
|
|
|
&clientmodel.Sample{
|
|
|
|
Metric: clientmodel.Metric{
|
|
|
|
clientmodel.MetricNameLabel: "http_requests",
|
|
|
|
clientmodel.JobLabel: "api-server",
|
|
|
|
"instance": "2",
|
|
|
|
"group": "canary",
|
2013-01-07 14:24:26 -08:00
|
|
|
},
|
|
|
|
Value: 40,
|
2013-03-28 09:05:06 -07:00
|
|
|
Timestamp: timestamp,
|
2013-01-07 14:24:26 -08:00
|
|
|
},
|
2013-06-25 05:02:27 -07:00
|
|
|
&clientmodel.Sample{
|
|
|
|
Metric: clientmodel.Metric{
|
|
|
|
clientmodel.MetricNameLabel: "http_requests",
|
|
|
|
clientmodel.JobLabel: "api-server",
|
|
|
|
"instance": "3",
|
|
|
|
"group": "mytest",
|
2013-01-07 14:24:26 -08:00
|
|
|
},
|
|
|
|
Value: 40,
|
2013-03-28 09:05:06 -07:00
|
|
|
Timestamp: timestamp,
|
2013-01-07 14:24:26 -08:00
|
|
|
},
|
2013-06-25 05:02:27 -07:00
|
|
|
&clientmodel.Sample{
|
|
|
|
Metric: clientmodel.Metric{
|
|
|
|
clientmodel.MetricNameLabel: "http_requests",
|
|
|
|
clientmodel.JobLabel: "api-server",
|
|
|
|
"instance": "3",
|
|
|
|
"group": "mytest",
|
2013-01-07 14:24:26 -08:00
|
|
|
},
|
|
|
|
Value: 40,
|
2013-03-28 09:05:06 -07:00
|
|
|
Timestamp: timestamp,
|
2013-01-07 14:24:26 -08:00
|
|
|
},
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2013-09-17 05:59:31 -07:00
|
|
|
// === scalar(node *VectorNode) Scalar ===
|
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
|
|
|
func scalarImpl(timestamp clientmodel.Timestamp, view *viewAdapter, args []Node) interface{} {
|
2013-09-17 05:59:31 -07:00
|
|
|
v := args[0].(VectorNode).Eval(timestamp, view)
|
|
|
|
if len(v) != 1 {
|
|
|
|
return clientmodel.SampleValue(math.NaN())
|
|
|
|
}
|
|
|
|
return clientmodel.SampleValue(v[0].Value)
|
|
|
|
}
|
|
|
|
|
2014-01-30 04:07:26 -08:00
|
|
|
// === count_scalar(vector VectorNode) model.SampleValue ===
|
|
|
|
func countScalarImpl(timestamp clientmodel.Timestamp, view *viewAdapter, args []Node) interface{} {
|
|
|
|
return clientmodel.SampleValue(len(args[0].(VectorNode).Eval(timestamp, view)))
|
|
|
|
}
|
|
|
|
|
2013-01-07 14:24:26 -08:00
|
|
|
var functions = map[string]*Function{
|
2014-01-30 04:07:26 -08:00
|
|
|
"count_scalar": {
|
|
|
|
name: "count_scalar",
|
|
|
|
argTypes: []ExprType{VECTOR},
|
|
|
|
returnType: SCALAR,
|
|
|
|
callFn: countScalarImpl,
|
|
|
|
},
|
2013-01-07 14:24:26 -08:00
|
|
|
"delta": {
|
|
|
|
name: "delta",
|
|
|
|
argTypes: []ExprType{MATRIX, SCALAR},
|
|
|
|
returnType: VECTOR,
|
|
|
|
callFn: deltaImpl,
|
|
|
|
},
|
|
|
|
"rate": {
|
|
|
|
name: "rate",
|
|
|
|
argTypes: []ExprType{MATRIX},
|
|
|
|
returnType: VECTOR,
|
|
|
|
callFn: rateImpl,
|
|
|
|
},
|
|
|
|
"sampleVector": {
|
|
|
|
name: "sampleVector",
|
|
|
|
argTypes: []ExprType{},
|
|
|
|
returnType: VECTOR,
|
|
|
|
callFn: sampleVectorImpl,
|
|
|
|
},
|
2013-09-17 05:59:31 -07:00
|
|
|
"scalar": {
|
|
|
|
name: "scalar",
|
|
|
|
argTypes: []ExprType{VECTOR},
|
|
|
|
returnType: SCALAR,
|
|
|
|
callFn: scalarImpl,
|
|
|
|
},
|
2013-04-10 01:44:13 -07:00
|
|
|
"sort": {
|
|
|
|
name: "sort",
|
|
|
|
argTypes: []ExprType{VECTOR},
|
|
|
|
returnType: VECTOR,
|
|
|
|
callFn: sortImpl,
|
|
|
|
},
|
|
|
|
"sort_desc": {
|
|
|
|
name: "sort_desc",
|
|
|
|
argTypes: []ExprType{VECTOR},
|
|
|
|
returnType: VECTOR,
|
|
|
|
callFn: sortDescImpl,
|
|
|
|
},
|
|
|
|
"time": {
|
|
|
|
name: "time",
|
|
|
|
argTypes: []ExprType{},
|
|
|
|
returnType: SCALAR,
|
|
|
|
callFn: timeImpl,
|
|
|
|
},
|
2013-01-07 14:24:26 -08:00
|
|
|
}
|
|
|
|
|
2014-02-13 09:48:56 -08:00
|
|
|
// GetFunction returns a predefined Function object for the given
|
|
|
|
// name.
|
2013-01-07 14:24:26 -08:00
|
|
|
func GetFunction(name string) (*Function, error) {
|
|
|
|
function, ok := functions[name]
|
|
|
|
if !ok {
|
2014-02-13 09:48:56 -08:00
|
|
|
return nil, fmt.Errorf("couldn't find function %v()", name)
|
2013-01-07 14:24:26 -08:00
|
|
|
}
|
|
|
|
return function, nil
|
|
|
|
}
|