|
|
|
// Copyright 2016 The Prometheus Authors
|
|
|
|
// 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 rules
|
|
|
|
|
|
|
|
import (
|
|
|
|
"context"
|
|
|
|
"errors"
|
|
|
|
"testing"
|
|
|
|
"time"
|
|
|
|
|
|
|
|
"github.com/go-kit/log"
|
|
|
|
"github.com/prometheus/common/model"
|
|
|
|
"github.com/stretchr/testify/require"
|
|
|
|
|
|
|
|
"github.com/prometheus/prometheus/model/histogram"
|
|
|
|
"github.com/prometheus/prometheus/model/labels"
|
|
|
|
"github.com/prometheus/prometheus/model/relabel"
|
|
|
|
"github.com/prometheus/prometheus/model/timestamp"
|
|
|
|
"github.com/prometheus/prometheus/notifier"
|
|
|
|
"github.com/prometheus/prometheus/promql"
|
|
|
|
"github.com/prometheus/prometheus/promql/parser"
|
|
|
|
"github.com/prometheus/prometheus/storage"
|
|
|
|
"github.com/prometheus/prometheus/util/teststorage"
|
|
|
|
"github.com/prometheus/prometheus/util/testutil"
|
|
|
|
)
|
|
|
|
|
|
|
|
var testEngine = promql.NewEngine(promql.EngineOpts{
|
|
|
|
Logger: nil,
|
|
|
|
Reg: nil,
|
|
|
|
MaxSamples: 10000,
|
|
|
|
Timeout: 100 * time.Second,
|
|
|
|
NoStepSubqueryIntervalFn: func(int64) int64 { return 60 * 1000 },
|
|
|
|
EnableAtModifier: true,
|
|
|
|
EnableNegativeOffset: true,
|
|
|
|
EnablePerStepStats: true,
|
|
|
|
})
|
|
|
|
|
|
|
|
func TestAlertingRuleState(t *testing.T) {
|
|
|
|
tests := []struct {
|
|
|
|
name string
|
|
|
|
active map[uint64]*Alert
|
|
|
|
want AlertState
|
|
|
|
}{
|
|
|
|
{
|
|
|
|
name: "MaxStateFiring",
|
|
|
|
active: map[uint64]*Alert{
|
|
|
|
0: {State: StatePending},
|
|
|
|
1: {State: StateFiring},
|
|
|
|
},
|
|
|
|
want: StateFiring,
|
|
|
|
},
|
|
|
|
{
|
|
|
|
name: "MaxStatePending",
|
|
|
|
active: map[uint64]*Alert{
|
|
|
|
0: {State: StateInactive},
|
|
|
|
1: {State: StatePending},
|
|
|
|
},
|
|
|
|
want: StatePending,
|
|
|
|
},
|
|
|
|
{
|
|
|
|
name: "MaxStateInactive",
|
|
|
|
active: map[uint64]*Alert{
|
|
|
|
0: {State: StateInactive},
|
|
|
|
1: {State: StateInactive},
|
|
|
|
},
|
|
|
|
want: StateInactive,
|
|
|
|
},
|
|
|
|
}
|
|
|
|
|
|
|
|
for i, test := range tests {
|
|
|
|
rule := NewAlertingRule(test.name, nil, 0, 0, labels.EmptyLabels(), labels.EmptyLabels(), labels.EmptyLabels(), "", true, nil)
|
|
|
|
rule.active = test.active
|
|
|
|
got := rule.State()
|
|
|
|
require.Equal(t, test.want, got, "test case %d unexpected AlertState, want:%d got:%d", i, test.want, got)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
func TestAlertingRuleTemplateWithHistogram(t *testing.T) {
|
|
|
|
h := histogram.FloatHistogram{
|
|
|
|
Schema: 0,
|
|
|
|
Count: 30,
|
|
|
|
Sum: 1111.1,
|
|
|
|
ZeroThreshold: 0.001,
|
|
|
|
ZeroCount: 2,
|
|
|
|
PositiveSpans: []histogram.Span{
|
|
|
|
{Offset: 0, Length: 1},
|
|
|
|
{Offset: 1, Length: 5},
|
|
|
|
},
|
|
|
|
PositiveBuckets: []float64{1, 1, 2, 1, 1, 1},
|
|
|
|
NegativeSpans: []histogram.Span{
|
|
|
|
{Offset: 1, Length: 4},
|
|
|
|
{Offset: 4, Length: 3},
|
|
|
|
},
|
|
|
|
NegativeBuckets: []float64{-2, 2, 2, 7, 5, 5, 2},
|
|
|
|
}
|
|
|
|
|
|
|
|
q := func(ctx context.Context, qs string, t time.Time) (promql.Vector, error) {
|
|
|
|
return []promql.Sample{{H: &h}}, nil
|
|
|
|
}
|
|
|
|
|
|
|
|
expr, err := parser.ParseExpr("foo")
|
|
|
|
require.NoError(t, err)
|
|
|
|
|
|
|
|
rule := NewAlertingRule(
|
|
|
|
"HistogramAsValue",
|
|
|
|
expr,
|
|
|
|
time.Minute,
|
|
|
|
0,
|
|
|
|
labels.FromStrings("histogram", "{{ $value }}"),
|
|
|
|
labels.EmptyLabels(), labels.EmptyLabels(), "", true, nil,
|
|
|
|
)
|
|
|
|
|
|
|
|
evalTime := time.Now()
|
|
|
|
res, err := rule.Eval(context.TODO(), evalTime, q, nil, 0)
|
|
|
|
require.NoError(t, err)
|
|
|
|
|
|
|
|
require.Len(t, res, 2)
|
|
|
|
for _, smpl := range res {
|
|
|
|
smplName := smpl.Metric.Get("__name__")
|
|
|
|
if smplName == "ALERTS" {
|
|
|
|
result := promql.Sample{
|
|
|
|
Metric: labels.FromStrings(
|
|
|
|
"__name__", "ALERTS",
|
|
|
|
"alertname", "HistogramAsValue",
|
|
|
|
"alertstate", "pending",
|
|
|
|
"histogram", h.String(),
|
|
|
|
),
|
|
|
|
T: timestamp.FromTime(evalTime),
|
|
|
|
F: 1,
|
|
|
|
}
|
|
|
|
testutil.RequireEqual(t, result, smpl)
|
|
|
|
} else {
|
|
|
|
// If not 'ALERTS', it has to be 'ALERTS_FOR_STATE'.
|
|
|
|
require.Equal(t, "ALERTS_FOR_STATE", smplName)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
func TestAlertingRuleLabelsUpdate(t *testing.T) {
|
|
|
|
storage := promql.LoadedStorage(t, `
|
|
|
|
load 1m
|
|
|
|
http_requests{job="app-server", instance="0"} 75 85 70 70 stale
|
|
|
|
`)
|
|
|
|
t.Cleanup(func() { storage.Close() })
|
|
|
|
|
|
|
|
expr, err := parser.ParseExpr(`http_requests < 100`)
|
|
|
|
require.NoError(t, err)
|
|
|
|
|
|
|
|
rule := NewAlertingRule(
|
|
|
|
"HTTPRequestRateLow",
|
|
|
|
expr,
|
|
|
|
time.Minute,
|
|
|
|
0,
|
|
|
|
// Basing alerting rule labels off of a value that can change is a very bad idea.
|
|
|
|
// If an alert is going back and forth between two label values it will never fire.
|
|
|
|
// Instead, you should write two alerts with constant labels.
|
|
|
|
labels.FromStrings("severity", "{{ if lt $value 80.0 }}critical{{ else }}warning{{ end }}"),
|
|
|
|
labels.EmptyLabels(), labels.EmptyLabels(), "", true, nil,
|
|
|
|
)
|
|
|
|
|
|
|
|
results := []promql.Vector{
|
|
|
|
{
|
|
|
|
promql.Sample{
|
|
|
|
Metric: labels.FromStrings(
|
|
|
|
"__name__", "ALERTS",
|
|
|
|
"alertname", "HTTPRequestRateLow",
|
|
|
|
"alertstate", "pending",
|
|
|
|
"instance", "0",
|
|
|
|
"job", "app-server",
|
|
|
|
"severity", "critical",
|
|
|
|
),
|
promql: Separate `Point` into `FPoint` and `HPoint`
In other words: Instead of having a “polymorphous” `Point` that can
either contain a float value or a histogram value, use an `FPoint` for
floats and an `HPoint` for histograms.
This seemingly small change has a _lot_ of repercussions throughout
the codebase.
The idea here is to avoid the increase in size of `Point` arrays that
happened after native histograms had been added.
The higher-level data structures (`Sample`, `Series`, etc.) are still
“polymorphous”. The same idea could be applied to them, but at each
step the trade-offs needed to be evaluated.
The idea with this change is to do the minimum necessary to get back
to pre-histogram performance for functions that do not touch
histograms. Here are comparisons for the `changes` function. The test
data doesn't include histograms yet. Ideally, there would be no change
in the benchmark result at all.
First runtime v2.39 compared to directly prior to this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 542µs ± 1% +38.58% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 617µs ± 2% +36.48% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.36ms ± 2% +21.58% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 8.94ms ± 1% +14.21% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.30ms ± 1% +10.67% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.10ms ± 1% +11.82% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 11.8ms ± 1% +12.50% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 87.4ms ± 1% +12.63% (p=0.000 n=9+9)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 32.8ms ± 1% +8.01% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.6ms ± 2% +9.64% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 117ms ± 1% +11.69% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 876ms ± 1% +11.83% (p=0.000 n=9+10)
```
And then runtime v2.39 compared to after this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 547µs ± 1% +39.84% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 616µs ± 2% +36.15% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.26ms ± 1% +12.20% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 7.95ms ± 1% +1.59% (p=0.000 n=10+8)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.38ms ± 2% +13.49% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.02ms ± 1% +9.80% (p=0.000 n=10+9)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 10.8ms ± 1% +3.08% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 78.1ms ± 1% +0.58% (p=0.035 n=9+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 33.5ms ± 4% +10.18% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.0ms ± 1% +7.98% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 107ms ± 1% +1.92% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 775ms ± 1% -1.02% (p=0.019 n=9+9)
```
In summary, the runtime doesn't really improve with this change for
queries with just a few steps. For queries with many steps, this
commit essentially reinstates the old performance. This is good
because the many-step queries are the one that matter most (longest
absolute runtime).
In terms of allocations, though, this commit doesn't make a dent at
all (numbers not shown). The reason is that most of the allocations
happen in the sampleRingIterator (in the storage package), which has
to be addressed in a separate commit.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
F: 1,
|
|
|
|
},
|
|
|
|
},
|
|
|
|
{
|
|
|
|
promql.Sample{
|
|
|
|
Metric: labels.FromStrings(
|
|
|
|
"__name__", "ALERTS",
|
|
|
|
"alertname", "HTTPRequestRateLow",
|
|
|
|
"alertstate", "pending",
|
|
|
|
"instance", "0",
|
|
|
|
"job", "app-server",
|
|
|
|
"severity", "warning",
|
|
|
|
),
|
promql: Separate `Point` into `FPoint` and `HPoint`
In other words: Instead of having a “polymorphous” `Point` that can
either contain a float value or a histogram value, use an `FPoint` for
floats and an `HPoint` for histograms.
This seemingly small change has a _lot_ of repercussions throughout
the codebase.
The idea here is to avoid the increase in size of `Point` arrays that
happened after native histograms had been added.
The higher-level data structures (`Sample`, `Series`, etc.) are still
“polymorphous”. The same idea could be applied to them, but at each
step the trade-offs needed to be evaluated.
The idea with this change is to do the minimum necessary to get back
to pre-histogram performance for functions that do not touch
histograms. Here are comparisons for the `changes` function. The test
data doesn't include histograms yet. Ideally, there would be no change
in the benchmark result at all.
First runtime v2.39 compared to directly prior to this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 542µs ± 1% +38.58% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 617µs ± 2% +36.48% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.36ms ± 2% +21.58% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 8.94ms ± 1% +14.21% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.30ms ± 1% +10.67% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.10ms ± 1% +11.82% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 11.8ms ± 1% +12.50% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 87.4ms ± 1% +12.63% (p=0.000 n=9+9)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 32.8ms ± 1% +8.01% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.6ms ± 2% +9.64% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 117ms ± 1% +11.69% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 876ms ± 1% +11.83% (p=0.000 n=9+10)
```
And then runtime v2.39 compared to after this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 547µs ± 1% +39.84% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 616µs ± 2% +36.15% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.26ms ± 1% +12.20% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 7.95ms ± 1% +1.59% (p=0.000 n=10+8)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.38ms ± 2% +13.49% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.02ms ± 1% +9.80% (p=0.000 n=10+9)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 10.8ms ± 1% +3.08% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 78.1ms ± 1% +0.58% (p=0.035 n=9+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 33.5ms ± 4% +10.18% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.0ms ± 1% +7.98% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 107ms ± 1% +1.92% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 775ms ± 1% -1.02% (p=0.019 n=9+9)
```
In summary, the runtime doesn't really improve with this change for
queries with just a few steps. For queries with many steps, this
commit essentially reinstates the old performance. This is good
because the many-step queries are the one that matter most (longest
absolute runtime).
In terms of allocations, though, this commit doesn't make a dent at
all (numbers not shown). The reason is that most of the allocations
happen in the sampleRingIterator (in the storage package), which has
to be addressed in a separate commit.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
F: 1,
|
|
|
|
},
|
|
|
|
},
|
|
|
|
{
|
|
|
|
promql.Sample{
|
|
|
|
Metric: labels.FromStrings(
|
|
|
|
"__name__", "ALERTS",
|
|
|
|
"alertname", "HTTPRequestRateLow",
|
|
|
|
"alertstate", "pending",
|
|
|
|
"instance", "0",
|
|
|
|
"job", "app-server",
|
|
|
|
"severity", "critical",
|
|
|
|
),
|
promql: Separate `Point` into `FPoint` and `HPoint`
In other words: Instead of having a “polymorphous” `Point` that can
either contain a float value or a histogram value, use an `FPoint` for
floats and an `HPoint` for histograms.
This seemingly small change has a _lot_ of repercussions throughout
the codebase.
The idea here is to avoid the increase in size of `Point` arrays that
happened after native histograms had been added.
The higher-level data structures (`Sample`, `Series`, etc.) are still
“polymorphous”. The same idea could be applied to them, but at each
step the trade-offs needed to be evaluated.
The idea with this change is to do the minimum necessary to get back
to pre-histogram performance for functions that do not touch
histograms. Here are comparisons for the `changes` function. The test
data doesn't include histograms yet. Ideally, there would be no change
in the benchmark result at all.
First runtime v2.39 compared to directly prior to this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 542µs ± 1% +38.58% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 617µs ± 2% +36.48% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.36ms ± 2% +21.58% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 8.94ms ± 1% +14.21% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.30ms ± 1% +10.67% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.10ms ± 1% +11.82% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 11.8ms ± 1% +12.50% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 87.4ms ± 1% +12.63% (p=0.000 n=9+9)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 32.8ms ± 1% +8.01% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.6ms ± 2% +9.64% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 117ms ± 1% +11.69% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 876ms ± 1% +11.83% (p=0.000 n=9+10)
```
And then runtime v2.39 compared to after this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 547µs ± 1% +39.84% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 616µs ± 2% +36.15% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.26ms ± 1% +12.20% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 7.95ms ± 1% +1.59% (p=0.000 n=10+8)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.38ms ± 2% +13.49% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.02ms ± 1% +9.80% (p=0.000 n=10+9)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 10.8ms ± 1% +3.08% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 78.1ms ± 1% +0.58% (p=0.035 n=9+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 33.5ms ± 4% +10.18% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.0ms ± 1% +7.98% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 107ms ± 1% +1.92% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 775ms ± 1% -1.02% (p=0.019 n=9+9)
```
In summary, the runtime doesn't really improve with this change for
queries with just a few steps. For queries with many steps, this
commit essentially reinstates the old performance. This is good
because the many-step queries are the one that matter most (longest
absolute runtime).
In terms of allocations, though, this commit doesn't make a dent at
all (numbers not shown). The reason is that most of the allocations
happen in the sampleRingIterator (in the storage package), which has
to be addressed in a separate commit.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
F: 1,
|
|
|
|
},
|
|
|
|
},
|
|
|
|
{
|
|
|
|
promql.Sample{
|
|
|
|
Metric: labels.FromStrings(
|
|
|
|
"__name__", "ALERTS",
|
|
|
|
"alertname", "HTTPRequestRateLow",
|
|
|
|
"alertstate", "firing",
|
|
|
|
"instance", "0",
|
|
|
|
"job", "app-server",
|
|
|
|
"severity", "critical",
|
|
|
|
),
|
promql: Separate `Point` into `FPoint` and `HPoint`
In other words: Instead of having a “polymorphous” `Point` that can
either contain a float value or a histogram value, use an `FPoint` for
floats and an `HPoint` for histograms.
This seemingly small change has a _lot_ of repercussions throughout
the codebase.
The idea here is to avoid the increase in size of `Point` arrays that
happened after native histograms had been added.
The higher-level data structures (`Sample`, `Series`, etc.) are still
“polymorphous”. The same idea could be applied to them, but at each
step the trade-offs needed to be evaluated.
The idea with this change is to do the minimum necessary to get back
to pre-histogram performance for functions that do not touch
histograms. Here are comparisons for the `changes` function. The test
data doesn't include histograms yet. Ideally, there would be no change
in the benchmark result at all.
First runtime v2.39 compared to directly prior to this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 542µs ± 1% +38.58% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 617µs ± 2% +36.48% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.36ms ± 2% +21.58% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 8.94ms ± 1% +14.21% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.30ms ± 1% +10.67% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.10ms ± 1% +11.82% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 11.8ms ± 1% +12.50% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 87.4ms ± 1% +12.63% (p=0.000 n=9+9)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 32.8ms ± 1% +8.01% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.6ms ± 2% +9.64% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 117ms ± 1% +11.69% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 876ms ± 1% +11.83% (p=0.000 n=9+10)
```
And then runtime v2.39 compared to after this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 547µs ± 1% +39.84% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 616µs ± 2% +36.15% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.26ms ± 1% +12.20% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 7.95ms ± 1% +1.59% (p=0.000 n=10+8)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.38ms ± 2% +13.49% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.02ms ± 1% +9.80% (p=0.000 n=10+9)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 10.8ms ± 1% +3.08% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 78.1ms ± 1% +0.58% (p=0.035 n=9+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 33.5ms ± 4% +10.18% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.0ms ± 1% +7.98% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 107ms ± 1% +1.92% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 775ms ± 1% -1.02% (p=0.019 n=9+9)
```
In summary, the runtime doesn't really improve with this change for
queries with just a few steps. For queries with many steps, this
commit essentially reinstates the old performance. This is good
because the many-step queries are the one that matter most (longest
absolute runtime).
In terms of allocations, though, this commit doesn't make a dent at
all (numbers not shown). The reason is that most of the allocations
happen in the sampleRingIterator (in the storage package), which has
to be addressed in a separate commit.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
F: 1,
|
|
|
|
},
|
|
|
|
},
|
|
|
|
}
|
|
|
|
|
|
|
|
baseTime := time.Unix(0, 0)
|
|
|
|
for i, result := range results {
|
|
|
|
t.Logf("case %d", i)
|
|
|
|
evalTime := baseTime.Add(time.Duration(i) * time.Minute)
|
promql: Separate `Point` into `FPoint` and `HPoint`
In other words: Instead of having a “polymorphous” `Point` that can
either contain a float value or a histogram value, use an `FPoint` for
floats and an `HPoint` for histograms.
This seemingly small change has a _lot_ of repercussions throughout
the codebase.
The idea here is to avoid the increase in size of `Point` arrays that
happened after native histograms had been added.
The higher-level data structures (`Sample`, `Series`, etc.) are still
“polymorphous”. The same idea could be applied to them, but at each
step the trade-offs needed to be evaluated.
The idea with this change is to do the minimum necessary to get back
to pre-histogram performance for functions that do not touch
histograms. Here are comparisons for the `changes` function. The test
data doesn't include histograms yet. Ideally, there would be no change
in the benchmark result at all.
First runtime v2.39 compared to directly prior to this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 542µs ± 1% +38.58% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 617µs ± 2% +36.48% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.36ms ± 2% +21.58% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 8.94ms ± 1% +14.21% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.30ms ± 1% +10.67% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.10ms ± 1% +11.82% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 11.8ms ± 1% +12.50% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 87.4ms ± 1% +12.63% (p=0.000 n=9+9)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 32.8ms ± 1% +8.01% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.6ms ± 2% +9.64% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 117ms ± 1% +11.69% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 876ms ± 1% +11.83% (p=0.000 n=9+10)
```
And then runtime v2.39 compared to after this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 547µs ± 1% +39.84% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 616µs ± 2% +36.15% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.26ms ± 1% +12.20% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 7.95ms ± 1% +1.59% (p=0.000 n=10+8)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.38ms ± 2% +13.49% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.02ms ± 1% +9.80% (p=0.000 n=10+9)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 10.8ms ± 1% +3.08% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 78.1ms ± 1% +0.58% (p=0.035 n=9+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 33.5ms ± 4% +10.18% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.0ms ± 1% +7.98% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 107ms ± 1% +1.92% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 775ms ± 1% -1.02% (p=0.019 n=9+9)
```
In summary, the runtime doesn't really improve with this change for
queries with just a few steps. For queries with many steps, this
commit essentially reinstates the old performance. This is good
because the many-step queries are the one that matter most (longest
absolute runtime).
In terms of allocations, though, this commit doesn't make a dent at
all (numbers not shown). The reason is that most of the allocations
happen in the sampleRingIterator (in the storage package), which has
to be addressed in a separate commit.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
result[0].T = timestamp.FromTime(evalTime)
|
|
|
|
res, err := rule.Eval(context.TODO(), evalTime, EngineQueryFunc(testEngine, storage), nil, 0)
|
|
|
|
require.NoError(t, err)
|
|
|
|
|
|
|
|
var filteredRes promql.Vector // After removing 'ALERTS_FOR_STATE' samples.
|
|
|
|
for _, smpl := range res {
|
|
|
|
smplName := smpl.Metric.Get("__name__")
|
|
|
|
if smplName == "ALERTS" {
|
|
|
|
filteredRes = append(filteredRes, smpl)
|
|
|
|
} else {
|
|
|
|
// If not 'ALERTS', it has to be 'ALERTS_FOR_STATE'.
|
|
|
|
require.Equal(t, "ALERTS_FOR_STATE", smplName)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
testutil.RequireEqual(t, result, filteredRes)
|
|
|
|
}
|
|
|
|
evalTime := baseTime.Add(time.Duration(len(results)) * time.Minute)
|
|
|
|
res, err := rule.Eval(context.TODO(), evalTime, EngineQueryFunc(testEngine, storage), nil, 0)
|
|
|
|
require.NoError(t, err)
|
|
|
|
require.Empty(t, res)
|
|
|
|
}
|
|
|
|
|
|
|
|
func TestAlertingRuleExternalLabelsInTemplate(t *testing.T) {
|
|
|
|
storage := promql.LoadedStorage(t, `
|
|
|
|
load 1m
|
|
|
|
http_requests{job="app-server", instance="0"} 75 85 70 70
|
|
|
|
`)
|
|
|
|
t.Cleanup(func() { storage.Close() })
|
|
|
|
|
|
|
|
expr, err := parser.ParseExpr(`http_requests < 100`)
|
|
|
|
require.NoError(t, err)
|
|
|
|
|
|
|
|
ruleWithoutExternalLabels := NewAlertingRule(
|
|
|
|
"ExternalLabelDoesNotExist",
|
|
|
|
expr,
|
|
|
|
time.Minute,
|
|
|
|
0,
|
|
|
|
labels.FromStrings("templated_label", "There are {{ len $externalLabels }} external Labels, of which foo is {{ $externalLabels.foo }}."),
|
|
|
|
labels.EmptyLabels(),
|
|
|
|
labels.EmptyLabels(),
|
|
|
|
"",
|
|
|
|
true, log.NewNopLogger(),
|
|
|
|
)
|
|
|
|
ruleWithExternalLabels := NewAlertingRule(
|
|
|
|
"ExternalLabelExists",
|
|
|
|
expr,
|
|
|
|
time.Minute,
|
|
|
|
0,
|
|
|
|
labels.FromStrings("templated_label", "There are {{ len $externalLabels }} external Labels, of which foo is {{ $externalLabels.foo }}."),
|
|
|
|
labels.EmptyLabels(),
|
|
|
|
labels.FromStrings("foo", "bar", "dings", "bums"),
|
|
|
|
"",
|
|
|
|
true, log.NewNopLogger(),
|
|
|
|
)
|
|
|
|
result := promql.Vector{
|
|
|
|
promql.Sample{
|
|
|
|
Metric: labels.FromStrings(
|
|
|
|
"__name__", "ALERTS",
|
|
|
|
"alertname", "ExternalLabelDoesNotExist",
|
|
|
|
"alertstate", "pending",
|
|
|
|
"instance", "0",
|
|
|
|
"job", "app-server",
|
|
|
|
"templated_label", "There are 0 external Labels, of which foo is .",
|
|
|
|
),
|
promql: Separate `Point` into `FPoint` and `HPoint`
In other words: Instead of having a “polymorphous” `Point` that can
either contain a float value or a histogram value, use an `FPoint` for
floats and an `HPoint` for histograms.
This seemingly small change has a _lot_ of repercussions throughout
the codebase.
The idea here is to avoid the increase in size of `Point` arrays that
happened after native histograms had been added.
The higher-level data structures (`Sample`, `Series`, etc.) are still
“polymorphous”. The same idea could be applied to them, but at each
step the trade-offs needed to be evaluated.
The idea with this change is to do the minimum necessary to get back
to pre-histogram performance for functions that do not touch
histograms. Here are comparisons for the `changes` function. The test
data doesn't include histograms yet. Ideally, there would be no change
in the benchmark result at all.
First runtime v2.39 compared to directly prior to this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 542µs ± 1% +38.58% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 617µs ± 2% +36.48% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.36ms ± 2% +21.58% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 8.94ms ± 1% +14.21% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.30ms ± 1% +10.67% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.10ms ± 1% +11.82% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 11.8ms ± 1% +12.50% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 87.4ms ± 1% +12.63% (p=0.000 n=9+9)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 32.8ms ± 1% +8.01% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.6ms ± 2% +9.64% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 117ms ± 1% +11.69% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 876ms ± 1% +11.83% (p=0.000 n=9+10)
```
And then runtime v2.39 compared to after this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 547µs ± 1% +39.84% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 616µs ± 2% +36.15% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.26ms ± 1% +12.20% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 7.95ms ± 1% +1.59% (p=0.000 n=10+8)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.38ms ± 2% +13.49% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.02ms ± 1% +9.80% (p=0.000 n=10+9)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 10.8ms ± 1% +3.08% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 78.1ms ± 1% +0.58% (p=0.035 n=9+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 33.5ms ± 4% +10.18% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.0ms ± 1% +7.98% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 107ms ± 1% +1.92% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 775ms ± 1% -1.02% (p=0.019 n=9+9)
```
In summary, the runtime doesn't really improve with this change for
queries with just a few steps. For queries with many steps, this
commit essentially reinstates the old performance. This is good
because the many-step queries are the one that matter most (longest
absolute runtime).
In terms of allocations, though, this commit doesn't make a dent at
all (numbers not shown). The reason is that most of the allocations
happen in the sampleRingIterator (in the storage package), which has
to be addressed in a separate commit.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
F: 1,
|
|
|
|
},
|
|
|
|
promql.Sample{
|
|
|
|
Metric: labels.FromStrings(
|
|
|
|
"__name__", "ALERTS",
|
|
|
|
"alertname", "ExternalLabelExists",
|
|
|
|
"alertstate", "pending",
|
|
|
|
"instance", "0",
|
|
|
|
"job", "app-server",
|
|
|
|
"templated_label", "There are 2 external Labels, of which foo is bar.",
|
|
|
|
),
|
promql: Separate `Point` into `FPoint` and `HPoint`
In other words: Instead of having a “polymorphous” `Point` that can
either contain a float value or a histogram value, use an `FPoint` for
floats and an `HPoint` for histograms.
This seemingly small change has a _lot_ of repercussions throughout
the codebase.
The idea here is to avoid the increase in size of `Point` arrays that
happened after native histograms had been added.
The higher-level data structures (`Sample`, `Series`, etc.) are still
“polymorphous”. The same idea could be applied to them, but at each
step the trade-offs needed to be evaluated.
The idea with this change is to do the minimum necessary to get back
to pre-histogram performance for functions that do not touch
histograms. Here are comparisons for the `changes` function. The test
data doesn't include histograms yet. Ideally, there would be no change
in the benchmark result at all.
First runtime v2.39 compared to directly prior to this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 542µs ± 1% +38.58% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 617µs ± 2% +36.48% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.36ms ± 2% +21.58% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 8.94ms ± 1% +14.21% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.30ms ± 1% +10.67% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.10ms ± 1% +11.82% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 11.8ms ± 1% +12.50% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 87.4ms ± 1% +12.63% (p=0.000 n=9+9)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 32.8ms ± 1% +8.01% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.6ms ± 2% +9.64% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 117ms ± 1% +11.69% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 876ms ± 1% +11.83% (p=0.000 n=9+10)
```
And then runtime v2.39 compared to after this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 547µs ± 1% +39.84% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 616µs ± 2% +36.15% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.26ms ± 1% +12.20% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 7.95ms ± 1% +1.59% (p=0.000 n=10+8)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.38ms ± 2% +13.49% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.02ms ± 1% +9.80% (p=0.000 n=10+9)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 10.8ms ± 1% +3.08% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 78.1ms ± 1% +0.58% (p=0.035 n=9+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 33.5ms ± 4% +10.18% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.0ms ± 1% +7.98% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 107ms ± 1% +1.92% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 775ms ± 1% -1.02% (p=0.019 n=9+9)
```
In summary, the runtime doesn't really improve with this change for
queries with just a few steps. For queries with many steps, this
commit essentially reinstates the old performance. This is good
because the many-step queries are the one that matter most (longest
absolute runtime).
In terms of allocations, though, this commit doesn't make a dent at
all (numbers not shown). The reason is that most of the allocations
happen in the sampleRingIterator (in the storage package), which has
to be addressed in a separate commit.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
F: 1,
|
|
|
|
},
|
|
|
|
}
|
|
|
|
|
|
|
|
evalTime := time.Unix(0, 0)
|
promql: Separate `Point` into `FPoint` and `HPoint`
In other words: Instead of having a “polymorphous” `Point` that can
either contain a float value or a histogram value, use an `FPoint` for
floats and an `HPoint` for histograms.
This seemingly small change has a _lot_ of repercussions throughout
the codebase.
The idea here is to avoid the increase in size of `Point` arrays that
happened after native histograms had been added.
The higher-level data structures (`Sample`, `Series`, etc.) are still
“polymorphous”. The same idea could be applied to them, but at each
step the trade-offs needed to be evaluated.
The idea with this change is to do the minimum necessary to get back
to pre-histogram performance for functions that do not touch
histograms. Here are comparisons for the `changes` function. The test
data doesn't include histograms yet. Ideally, there would be no change
in the benchmark result at all.
First runtime v2.39 compared to directly prior to this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 542µs ± 1% +38.58% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 617µs ± 2% +36.48% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.36ms ± 2% +21.58% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 8.94ms ± 1% +14.21% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.30ms ± 1% +10.67% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.10ms ± 1% +11.82% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 11.8ms ± 1% +12.50% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 87.4ms ± 1% +12.63% (p=0.000 n=9+9)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 32.8ms ± 1% +8.01% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.6ms ± 2% +9.64% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 117ms ± 1% +11.69% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 876ms ± 1% +11.83% (p=0.000 n=9+10)
```
And then runtime v2.39 compared to after this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 547µs ± 1% +39.84% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 616µs ± 2% +36.15% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.26ms ± 1% +12.20% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 7.95ms ± 1% +1.59% (p=0.000 n=10+8)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.38ms ± 2% +13.49% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.02ms ± 1% +9.80% (p=0.000 n=10+9)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 10.8ms ± 1% +3.08% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 78.1ms ± 1% +0.58% (p=0.035 n=9+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 33.5ms ± 4% +10.18% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.0ms ± 1% +7.98% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 107ms ± 1% +1.92% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 775ms ± 1% -1.02% (p=0.019 n=9+9)
```
In summary, the runtime doesn't really improve with this change for
queries with just a few steps. For queries with many steps, this
commit essentially reinstates the old performance. This is good
because the many-step queries are the one that matter most (longest
absolute runtime).
In terms of allocations, though, this commit doesn't make a dent at
all (numbers not shown). The reason is that most of the allocations
happen in the sampleRingIterator (in the storage package), which has
to be addressed in a separate commit.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
result[0].T = timestamp.FromTime(evalTime)
|
|
|
|
result[1].T = timestamp.FromTime(evalTime)
|
|
|
|
|
|
|
|
var filteredRes promql.Vector // After removing 'ALERTS_FOR_STATE' samples.
|
|
|
|
res, err := ruleWithoutExternalLabels.Eval(
|
|
|
|
context.TODO(), evalTime, EngineQueryFunc(testEngine, storage), nil, 0,
|
|
|
|
)
|
|
|
|
require.NoError(t, err)
|
|
|
|
for _, smpl := range res {
|
|
|
|
smplName := smpl.Metric.Get("__name__")
|
|
|
|
if smplName == "ALERTS" {
|
|
|
|
filteredRes = append(filteredRes, smpl)
|
|
|
|
} else {
|
|
|
|
// If not 'ALERTS', it has to be 'ALERTS_FOR_STATE'.
|
|
|
|
require.Equal(t, "ALERTS_FOR_STATE", smplName)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
res, err = ruleWithExternalLabels.Eval(
|
|
|
|
context.TODO(), evalTime, EngineQueryFunc(testEngine, storage), nil, 0,
|
|
|
|
)
|
|
|
|
require.NoError(t, err)
|
|
|
|
for _, smpl := range res {
|
|
|
|
smplName := smpl.Metric.Get("__name__")
|
|
|
|
if smplName == "ALERTS" {
|
|
|
|
filteredRes = append(filteredRes, smpl)
|
|
|
|
} else {
|
|
|
|
// If not 'ALERTS', it has to be 'ALERTS_FOR_STATE'.
|
|
|
|
require.Equal(t, "ALERTS_FOR_STATE", smplName)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
testutil.RequireEqual(t, result, filteredRes)
|
|
|
|
}
|
|
|
|
|
|
|
|
func TestAlertingRuleExternalURLInTemplate(t *testing.T) {
|
|
|
|
storage := promql.LoadedStorage(t, `
|
|
|
|
load 1m
|
|
|
|
http_requests{job="app-server", instance="0"} 75 85 70 70
|
|
|
|
`)
|
|
|
|
t.Cleanup(func() { storage.Close() })
|
|
|
|
|
|
|
|
expr, err := parser.ParseExpr(`http_requests < 100`)
|
|
|
|
require.NoError(t, err)
|
|
|
|
|
|
|
|
ruleWithoutExternalURL := NewAlertingRule(
|
|
|
|
"ExternalURLDoesNotExist",
|
|
|
|
expr,
|
|
|
|
time.Minute,
|
|
|
|
0,
|
|
|
|
labels.FromStrings("templated_label", "The external URL is {{ $externalURL }}."),
|
|
|
|
labels.EmptyLabels(),
|
|
|
|
labels.EmptyLabels(),
|
|
|
|
"",
|
|
|
|
true, log.NewNopLogger(),
|
|
|
|
)
|
|
|
|
ruleWithExternalURL := NewAlertingRule(
|
|
|
|
"ExternalURLExists",
|
|
|
|
expr,
|
|
|
|
time.Minute,
|
|
|
|
0,
|
|
|
|
labels.FromStrings("templated_label", "The external URL is {{ $externalURL }}."),
|
|
|
|
labels.EmptyLabels(),
|
|
|
|
labels.EmptyLabels(),
|
|
|
|
"http://localhost:1234",
|
|
|
|
true, log.NewNopLogger(),
|
|
|
|
)
|
|
|
|
result := promql.Vector{
|
|
|
|
promql.Sample{
|
|
|
|
Metric: labels.FromStrings(
|
|
|
|
"__name__", "ALERTS",
|
|
|
|
"alertname", "ExternalURLDoesNotExist",
|
|
|
|
"alertstate", "pending",
|
|
|
|
"instance", "0",
|
|
|
|
"job", "app-server",
|
|
|
|
"templated_label", "The external URL is .",
|
|
|
|
),
|
promql: Separate `Point` into `FPoint` and `HPoint`
In other words: Instead of having a “polymorphous” `Point` that can
either contain a float value or a histogram value, use an `FPoint` for
floats and an `HPoint` for histograms.
This seemingly small change has a _lot_ of repercussions throughout
the codebase.
The idea here is to avoid the increase in size of `Point` arrays that
happened after native histograms had been added.
The higher-level data structures (`Sample`, `Series`, etc.) are still
“polymorphous”. The same idea could be applied to them, but at each
step the trade-offs needed to be evaluated.
The idea with this change is to do the minimum necessary to get back
to pre-histogram performance for functions that do not touch
histograms. Here are comparisons for the `changes` function. The test
data doesn't include histograms yet. Ideally, there would be no change
in the benchmark result at all.
First runtime v2.39 compared to directly prior to this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 542µs ± 1% +38.58% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 617µs ± 2% +36.48% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.36ms ± 2% +21.58% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 8.94ms ± 1% +14.21% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.30ms ± 1% +10.67% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.10ms ± 1% +11.82% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 11.8ms ± 1% +12.50% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 87.4ms ± 1% +12.63% (p=0.000 n=9+9)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 32.8ms ± 1% +8.01% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.6ms ± 2% +9.64% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 117ms ± 1% +11.69% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 876ms ± 1% +11.83% (p=0.000 n=9+10)
```
And then runtime v2.39 compared to after this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 547µs ± 1% +39.84% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 616µs ± 2% +36.15% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.26ms ± 1% +12.20% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 7.95ms ± 1% +1.59% (p=0.000 n=10+8)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.38ms ± 2% +13.49% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.02ms ± 1% +9.80% (p=0.000 n=10+9)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 10.8ms ± 1% +3.08% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 78.1ms ± 1% +0.58% (p=0.035 n=9+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 33.5ms ± 4% +10.18% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.0ms ± 1% +7.98% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 107ms ± 1% +1.92% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 775ms ± 1% -1.02% (p=0.019 n=9+9)
```
In summary, the runtime doesn't really improve with this change for
queries with just a few steps. For queries with many steps, this
commit essentially reinstates the old performance. This is good
because the many-step queries are the one that matter most (longest
absolute runtime).
In terms of allocations, though, this commit doesn't make a dent at
all (numbers not shown). The reason is that most of the allocations
happen in the sampleRingIterator (in the storage package), which has
to be addressed in a separate commit.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
F: 1,
|
|
|
|
},
|
|
|
|
promql.Sample{
|
|
|
|
Metric: labels.FromStrings(
|
|
|
|
"__name__", "ALERTS",
|
|
|
|
"alertname", "ExternalURLExists",
|
|
|
|
"alertstate", "pending",
|
|
|
|
"instance", "0",
|
|
|
|
"job", "app-server",
|
|
|
|
"templated_label", "The external URL is http://localhost:1234.",
|
|
|
|
),
|
promql: Separate `Point` into `FPoint` and `HPoint`
In other words: Instead of having a “polymorphous” `Point` that can
either contain a float value or a histogram value, use an `FPoint` for
floats and an `HPoint` for histograms.
This seemingly small change has a _lot_ of repercussions throughout
the codebase.
The idea here is to avoid the increase in size of `Point` arrays that
happened after native histograms had been added.
The higher-level data structures (`Sample`, `Series`, etc.) are still
“polymorphous”. The same idea could be applied to them, but at each
step the trade-offs needed to be evaluated.
The idea with this change is to do the minimum necessary to get back
to pre-histogram performance for functions that do not touch
histograms. Here are comparisons for the `changes` function. The test
data doesn't include histograms yet. Ideally, there would be no change
in the benchmark result at all.
First runtime v2.39 compared to directly prior to this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 542µs ± 1% +38.58% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 617µs ± 2% +36.48% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.36ms ± 2% +21.58% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 8.94ms ± 1% +14.21% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.30ms ± 1% +10.67% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.10ms ± 1% +11.82% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 11.8ms ± 1% +12.50% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 87.4ms ± 1% +12.63% (p=0.000 n=9+9)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 32.8ms ± 1% +8.01% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.6ms ± 2% +9.64% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 117ms ± 1% +11.69% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 876ms ± 1% +11.83% (p=0.000 n=9+10)
```
And then runtime v2.39 compared to after this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 547µs ± 1% +39.84% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 616µs ± 2% +36.15% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.26ms ± 1% +12.20% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 7.95ms ± 1% +1.59% (p=0.000 n=10+8)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.38ms ± 2% +13.49% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.02ms ± 1% +9.80% (p=0.000 n=10+9)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 10.8ms ± 1% +3.08% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 78.1ms ± 1% +0.58% (p=0.035 n=9+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 33.5ms ± 4% +10.18% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.0ms ± 1% +7.98% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 107ms ± 1% +1.92% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 775ms ± 1% -1.02% (p=0.019 n=9+9)
```
In summary, the runtime doesn't really improve with this change for
queries with just a few steps. For queries with many steps, this
commit essentially reinstates the old performance. This is good
because the many-step queries are the one that matter most (longest
absolute runtime).
In terms of allocations, though, this commit doesn't make a dent at
all (numbers not shown). The reason is that most of the allocations
happen in the sampleRingIterator (in the storage package), which has
to be addressed in a separate commit.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
F: 1,
|
|
|
|
},
|
|
|
|
}
|
|
|
|
|
|
|
|
evalTime := time.Unix(0, 0)
|
promql: Separate `Point` into `FPoint` and `HPoint`
In other words: Instead of having a “polymorphous” `Point` that can
either contain a float value or a histogram value, use an `FPoint` for
floats and an `HPoint` for histograms.
This seemingly small change has a _lot_ of repercussions throughout
the codebase.
The idea here is to avoid the increase in size of `Point` arrays that
happened after native histograms had been added.
The higher-level data structures (`Sample`, `Series`, etc.) are still
“polymorphous”. The same idea could be applied to them, but at each
step the trade-offs needed to be evaluated.
The idea with this change is to do the minimum necessary to get back
to pre-histogram performance for functions that do not touch
histograms. Here are comparisons for the `changes` function. The test
data doesn't include histograms yet. Ideally, there would be no change
in the benchmark result at all.
First runtime v2.39 compared to directly prior to this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 542µs ± 1% +38.58% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 617µs ± 2% +36.48% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.36ms ± 2% +21.58% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 8.94ms ± 1% +14.21% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.30ms ± 1% +10.67% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.10ms ± 1% +11.82% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 11.8ms ± 1% +12.50% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 87.4ms ± 1% +12.63% (p=0.000 n=9+9)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 32.8ms ± 1% +8.01% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.6ms ± 2% +9.64% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 117ms ± 1% +11.69% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 876ms ± 1% +11.83% (p=0.000 n=9+10)
```
And then runtime v2.39 compared to after this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 547µs ± 1% +39.84% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 616µs ± 2% +36.15% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.26ms ± 1% +12.20% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 7.95ms ± 1% +1.59% (p=0.000 n=10+8)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.38ms ± 2% +13.49% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.02ms ± 1% +9.80% (p=0.000 n=10+9)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 10.8ms ± 1% +3.08% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 78.1ms ± 1% +0.58% (p=0.035 n=9+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 33.5ms ± 4% +10.18% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.0ms ± 1% +7.98% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 107ms ± 1% +1.92% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 775ms ± 1% -1.02% (p=0.019 n=9+9)
```
In summary, the runtime doesn't really improve with this change for
queries with just a few steps. For queries with many steps, this
commit essentially reinstates the old performance. This is good
because the many-step queries are the one that matter most (longest
absolute runtime).
In terms of allocations, though, this commit doesn't make a dent at
all (numbers not shown). The reason is that most of the allocations
happen in the sampleRingIterator (in the storage package), which has
to be addressed in a separate commit.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
result[0].T = timestamp.FromTime(evalTime)
|
|
|
|
result[1].T = timestamp.FromTime(evalTime)
|
|
|
|
|
|
|
|
var filteredRes promql.Vector // After removing 'ALERTS_FOR_STATE' samples.
|
|
|
|
res, err := ruleWithoutExternalURL.Eval(
|
|
|
|
context.TODO(), evalTime, EngineQueryFunc(testEngine, storage), nil, 0,
|
|
|
|
)
|
|
|
|
require.NoError(t, err)
|
|
|
|
for _, smpl := range res {
|
|
|
|
smplName := smpl.Metric.Get("__name__")
|
|
|
|
if smplName == "ALERTS" {
|
|
|
|
filteredRes = append(filteredRes, smpl)
|
|
|
|
} else {
|
|
|
|
// If not 'ALERTS', it has to be 'ALERTS_FOR_STATE'.
|
|
|
|
require.Equal(t, "ALERTS_FOR_STATE", smplName)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
res, err = ruleWithExternalURL.Eval(
|
|
|
|
context.TODO(), evalTime, EngineQueryFunc(testEngine, storage), nil, 0,
|
|
|
|
)
|
|
|
|
require.NoError(t, err)
|
|
|
|
for _, smpl := range res {
|
|
|
|
smplName := smpl.Metric.Get("__name__")
|
|
|
|
if smplName == "ALERTS" {
|
|
|
|
filteredRes = append(filteredRes, smpl)
|
|
|
|
} else {
|
|
|
|
// If not 'ALERTS', it has to be 'ALERTS_FOR_STATE'.
|
|
|
|
require.Equal(t, "ALERTS_FOR_STATE", smplName)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
testutil.RequireEqual(t, result, filteredRes)
|
|
|
|
}
|
|
|
|
|
|
|
|
func TestAlertingRuleEmptyLabelFromTemplate(t *testing.T) {
|
|
|
|
storage := promql.LoadedStorage(t, `
|
|
|
|
load 1m
|
|
|
|
http_requests{job="app-server", instance="0"} 75 85 70 70
|
|
|
|
`)
|
|
|
|
t.Cleanup(func() { storage.Close() })
|
|
|
|
|
|
|
|
expr, err := parser.ParseExpr(`http_requests < 100`)
|
|
|
|
require.NoError(t, err)
|
|
|
|
|
|
|
|
rule := NewAlertingRule(
|
|
|
|
"EmptyLabel",
|
|
|
|
expr,
|
|
|
|
time.Minute,
|
|
|
|
0,
|
|
|
|
labels.FromStrings("empty_label", ""),
|
|
|
|
labels.EmptyLabels(),
|
|
|
|
labels.EmptyLabels(),
|
|
|
|
"",
|
|
|
|
true, log.NewNopLogger(),
|
|
|
|
)
|
|
|
|
result := promql.Vector{
|
|
|
|
promql.Sample{
|
|
|
|
Metric: labels.FromStrings(
|
|
|
|
"__name__", "ALERTS",
|
|
|
|
"alertname", "EmptyLabel",
|
|
|
|
"alertstate", "pending",
|
|
|
|
"instance", "0",
|
|
|
|
"job", "app-server",
|
|
|
|
),
|
promql: Separate `Point` into `FPoint` and `HPoint`
In other words: Instead of having a “polymorphous” `Point` that can
either contain a float value or a histogram value, use an `FPoint` for
floats and an `HPoint` for histograms.
This seemingly small change has a _lot_ of repercussions throughout
the codebase.
The idea here is to avoid the increase in size of `Point` arrays that
happened after native histograms had been added.
The higher-level data structures (`Sample`, `Series`, etc.) are still
“polymorphous”. The same idea could be applied to them, but at each
step the trade-offs needed to be evaluated.
The idea with this change is to do the minimum necessary to get back
to pre-histogram performance for functions that do not touch
histograms. Here are comparisons for the `changes` function. The test
data doesn't include histograms yet. Ideally, there would be no change
in the benchmark result at all.
First runtime v2.39 compared to directly prior to this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 542µs ± 1% +38.58% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 617µs ± 2% +36.48% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.36ms ± 2% +21.58% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 8.94ms ± 1% +14.21% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.30ms ± 1% +10.67% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.10ms ± 1% +11.82% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 11.8ms ± 1% +12.50% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 87.4ms ± 1% +12.63% (p=0.000 n=9+9)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 32.8ms ± 1% +8.01% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.6ms ± 2% +9.64% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 117ms ± 1% +11.69% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 876ms ± 1% +11.83% (p=0.000 n=9+10)
```
And then runtime v2.39 compared to after this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 547µs ± 1% +39.84% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 616µs ± 2% +36.15% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.26ms ± 1% +12.20% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 7.95ms ± 1% +1.59% (p=0.000 n=10+8)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.38ms ± 2% +13.49% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.02ms ± 1% +9.80% (p=0.000 n=10+9)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 10.8ms ± 1% +3.08% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 78.1ms ± 1% +0.58% (p=0.035 n=9+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 33.5ms ± 4% +10.18% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.0ms ± 1% +7.98% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 107ms ± 1% +1.92% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 775ms ± 1% -1.02% (p=0.019 n=9+9)
```
In summary, the runtime doesn't really improve with this change for
queries with just a few steps. For queries with many steps, this
commit essentially reinstates the old performance. This is good
because the many-step queries are the one that matter most (longest
absolute runtime).
In terms of allocations, though, this commit doesn't make a dent at
all (numbers not shown). The reason is that most of the allocations
happen in the sampleRingIterator (in the storage package), which has
to be addressed in a separate commit.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
F: 1,
|
|
|
|
},
|
|
|
|
}
|
|
|
|
|
|
|
|
evalTime := time.Unix(0, 0)
|
promql: Separate `Point` into `FPoint` and `HPoint`
In other words: Instead of having a “polymorphous” `Point` that can
either contain a float value or a histogram value, use an `FPoint` for
floats and an `HPoint` for histograms.
This seemingly small change has a _lot_ of repercussions throughout
the codebase.
The idea here is to avoid the increase in size of `Point` arrays that
happened after native histograms had been added.
The higher-level data structures (`Sample`, `Series`, etc.) are still
“polymorphous”. The same idea could be applied to them, but at each
step the trade-offs needed to be evaluated.
The idea with this change is to do the minimum necessary to get back
to pre-histogram performance for functions that do not touch
histograms. Here are comparisons for the `changes` function. The test
data doesn't include histograms yet. Ideally, there would be no change
in the benchmark result at all.
First runtime v2.39 compared to directly prior to this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 542µs ± 1% +38.58% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 617µs ± 2% +36.48% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.36ms ± 2% +21.58% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 8.94ms ± 1% +14.21% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.30ms ± 1% +10.67% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.10ms ± 1% +11.82% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 11.8ms ± 1% +12.50% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 87.4ms ± 1% +12.63% (p=0.000 n=9+9)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 32.8ms ± 1% +8.01% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.6ms ± 2% +9.64% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 117ms ± 1% +11.69% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 876ms ± 1% +11.83% (p=0.000 n=9+10)
```
And then runtime v2.39 compared to after this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 547µs ± 1% +39.84% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 616µs ± 2% +36.15% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.26ms ± 1% +12.20% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 7.95ms ± 1% +1.59% (p=0.000 n=10+8)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.38ms ± 2% +13.49% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.02ms ± 1% +9.80% (p=0.000 n=10+9)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 10.8ms ± 1% +3.08% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 78.1ms ± 1% +0.58% (p=0.035 n=9+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 33.5ms ± 4% +10.18% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.0ms ± 1% +7.98% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 107ms ± 1% +1.92% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 775ms ± 1% -1.02% (p=0.019 n=9+9)
```
In summary, the runtime doesn't really improve with this change for
queries with just a few steps. For queries with many steps, this
commit essentially reinstates the old performance. This is good
because the many-step queries are the one that matter most (longest
absolute runtime).
In terms of allocations, though, this commit doesn't make a dent at
all (numbers not shown). The reason is that most of the allocations
happen in the sampleRingIterator (in the storage package), which has
to be addressed in a separate commit.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
result[0].T = timestamp.FromTime(evalTime)
|
|
|
|
|
|
|
|
var filteredRes promql.Vector // After removing 'ALERTS_FOR_STATE' samples.
|
|
|
|
res, err := rule.Eval(
|
|
|
|
context.TODO(), evalTime, EngineQueryFunc(testEngine, storage), nil, 0,
|
|
|
|
)
|
|
|
|
require.NoError(t, err)
|
|
|
|
for _, smpl := range res {
|
|
|
|
smplName := smpl.Metric.Get("__name__")
|
|
|
|
if smplName == "ALERTS" {
|
|
|
|
filteredRes = append(filteredRes, smpl)
|
|
|
|
} else {
|
|
|
|
// If not 'ALERTS', it has to be 'ALERTS_FOR_STATE'.
|
|
|
|
require.Equal(t, "ALERTS_FOR_STATE", smplName)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
testutil.RequireEqual(t, result, filteredRes)
|
|
|
|
}
|
|
|
|
|
|
|
|
func TestAlertingRuleQueryInTemplate(t *testing.T) {
|
|
|
|
storage := promql.LoadedStorage(t, `
|
|
|
|
load 1m
|
|
|
|
http_requests{job="app-server", instance="0"} 70 85 70 70
|
|
|
|
`)
|
|
|
|
t.Cleanup(func() { storage.Close() })
|
|
|
|
|
|
|
|
expr, err := parser.ParseExpr(`sum(http_requests) < 100`)
|
|
|
|
require.NoError(t, err)
|
|
|
|
|
|
|
|
ruleWithQueryInTemplate := NewAlertingRule(
|
|
|
|
"ruleWithQueryInTemplate",
|
|
|
|
expr,
|
|
|
|
time.Minute,
|
|
|
|
0,
|
|
|
|
labels.FromStrings("label", "value"),
|
|
|
|
labels.FromStrings("templated_label", `{{- with "sort(sum(http_requests) by (instance))" | query -}}
|
|
|
|
{{- range $i,$v := . -}}
|
|
|
|
instance: {{ $v.Labels.instance }}, value: {{ printf "%.0f" $v.Value }};
|
|
|
|
{{- end -}}
|
|
|
|
{{- end -}}
|
|
|
|
`),
|
|
|
|
labels.EmptyLabels(),
|
|
|
|
"",
|
|
|
|
true, log.NewNopLogger(),
|
|
|
|
)
|
|
|
|
evalTime := time.Unix(0, 0)
|
|
|
|
|
|
|
|
startQueryCh := make(chan struct{})
|
|
|
|
getDoneCh := make(chan struct{})
|
|
|
|
slowQueryFunc := func(ctx context.Context, q string, ts time.Time) (promql.Vector, error) {
|
|
|
|
if q == "sort(sum(http_requests) by (instance))" {
|
|
|
|
// This is a minimum reproduction of issue 10703, expand template with query.
|
|
|
|
close(startQueryCh)
|
|
|
|
select {
|
|
|
|
case <-getDoneCh:
|
|
|
|
case <-time.After(time.Millisecond * 10):
|
|
|
|
// Assert no blocking when template expanding.
|
|
|
|
require.Fail(t, "unexpected blocking when template expanding.")
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return EngineQueryFunc(testEngine, storage)(ctx, q, ts)
|
|
|
|
}
|
|
|
|
go func() {
|
|
|
|
<-startQueryCh
|
|
|
|
_ = ruleWithQueryInTemplate.Health()
|
|
|
|
_ = ruleWithQueryInTemplate.LastError()
|
|
|
|
_ = ruleWithQueryInTemplate.GetEvaluationDuration()
|
|
|
|
_ = ruleWithQueryInTemplate.GetEvaluationTimestamp()
|
|
|
|
close(getDoneCh)
|
|
|
|
}()
|
|
|
|
_, err = ruleWithQueryInTemplate.Eval(
|
|
|
|
context.TODO(), evalTime, slowQueryFunc, nil, 0,
|
|
|
|
)
|
|
|
|
require.NoError(t, err)
|
|
|
|
}
|
|
|
|
|
|
|
|
func BenchmarkAlertingRuleAtomicField(b *testing.B) {
|
|
|
|
b.ReportAllocs()
|
|
|
|
rule := NewAlertingRule("bench", nil, 0, 0, labels.EmptyLabels(), labels.EmptyLabels(), labels.EmptyLabels(), "", true, nil)
|
|
|
|
done := make(chan struct{})
|
|
|
|
go func() {
|
|
|
|
for i := 0; i < b.N; i++ {
|
|
|
|
rule.GetEvaluationTimestamp()
|
|
|
|
}
|
|
|
|
close(done)
|
|
|
|
}()
|
|
|
|
b.RunParallel(func(pb *testing.PB) {
|
|
|
|
for pb.Next() {
|
|
|
|
rule.SetEvaluationTimestamp(time.Now())
|
|
|
|
}
|
|
|
|
})
|
|
|
|
<-done
|
|
|
|
}
|
|
|
|
|
|
|
|
func TestAlertingRuleDuplicate(t *testing.T) {
|
|
|
|
storage := teststorage.New(t)
|
|
|
|
defer storage.Close()
|
|
|
|
|
|
|
|
opts := promql.EngineOpts{
|
|
|
|
Logger: nil,
|
|
|
|
Reg: nil,
|
|
|
|
MaxSamples: 10,
|
|
|
|
Timeout: 10 * time.Second,
|
|
|
|
}
|
|
|
|
|
|
|
|
engine := promql.NewEngine(opts)
|
|
|
|
ctx, cancelCtx := context.WithCancel(context.Background())
|
|
|
|
defer cancelCtx()
|
|
|
|
|
|
|
|
now := time.Now()
|
|
|
|
|
|
|
|
expr, _ := parser.ParseExpr(`vector(0) or label_replace(vector(0),"test","x","","")`)
|
|
|
|
rule := NewAlertingRule(
|
|
|
|
"foo",
|
|
|
|
expr,
|
|
|
|
time.Minute,
|
|
|
|
0,
|
|
|
|
labels.FromStrings("test", "test"),
|
|
|
|
labels.EmptyLabels(),
|
|
|
|
labels.EmptyLabels(),
|
|
|
|
"",
|
|
|
|
true, log.NewNopLogger(),
|
|
|
|
)
|
|
|
|
_, err := rule.Eval(ctx, now, EngineQueryFunc(engine, storage), nil, 0)
|
|
|
|
require.Error(t, err)
|
|
|
|
require.EqualError(t, err, "vector contains metrics with the same labelset after applying alert labels")
|
|
|
|
}
|
|
|
|
|
|
|
|
func TestAlertingRuleLimit(t *testing.T) {
|
|
|
|
storage := promql.LoadedStorage(t, `
|
|
|
|
load 1m
|
|
|
|
metric{label="1"} 1
|
|
|
|
metric{label="2"} 1
|
|
|
|
`)
|
|
|
|
t.Cleanup(func() { storage.Close() })
|
|
|
|
|
|
|
|
tests := []struct {
|
|
|
|
limit int
|
|
|
|
err string
|
|
|
|
}{
|
|
|
|
{
|
|
|
|
limit: 0,
|
|
|
|
},
|
|
|
|
{
|
|
|
|
limit: -1,
|
|
|
|
},
|
|
|
|
{
|
|
|
|
limit: 2,
|
|
|
|
},
|
|
|
|
{
|
|
|
|
limit: 1,
|
|
|
|
err: "exceeded limit of 1 with 2 alerts",
|
|
|
|
},
|
|
|
|
}
|
|
|
|
|
|
|
|
expr, _ := parser.ParseExpr(`metric > 0`)
|
|
|
|
rule := NewAlertingRule(
|
|
|
|
"foo",
|
|
|
|
expr,
|
|
|
|
time.Minute,
|
|
|
|
0,
|
|
|
|
labels.FromStrings("test", "test"),
|
|
|
|
labels.EmptyLabels(),
|
|
|
|
labels.EmptyLabels(),
|
|
|
|
"",
|
|
|
|
true, log.NewNopLogger(),
|
|
|
|
)
|
|
|
|
|
|
|
|
evalTime := time.Unix(0, 0)
|
|
|
|
|
|
|
|
for _, test := range tests {
|
|
|
|
switch _, err := rule.Eval(context.TODO(), evalTime, EngineQueryFunc(testEngine, storage), nil, test.limit); {
|
style: Replace `else if` cascades with `switch`
Wiser coders than myself have come to the conclusion that a `switch`
statement is almost always superior to a statement that includes any
`else if`.
The exceptions that I have found in our codebase are just these two:
* The `if else` is followed by an additional statement before the next
condition (separated by a `;`).
* The whole thing is within a `for` loop and `break` statements are
used. In this case, using `switch` would require tagging the `for`
loop, which probably tips the balance.
Why are `switch` statements more readable?
For one, fewer curly braces. But more importantly, the conditions all
have the same alignment, so the whole thing follows the natural flow
of going down a list of conditions. With `else if`, in contrast, all
conditions but the first are "hidden" behind `} else if `, harder to
spot and (for no good reason) presented differently from the first
condition.
I'm sure the aforemention wise coders can list even more reasons.
In any case, I like it so much that I have found myself recommending
it in code reviews. I would like to make it a habit in our code base,
without making it a hard requirement that we would test on the CI. But
for that, there has to be a role model, so this commit eliminates all
`if else` occurrences, unless it is autogenerated code or fits one of
the exceptions above.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
case err != nil:
|
|
|
|
require.EqualError(t, err, test.err)
|
style: Replace `else if` cascades with `switch`
Wiser coders than myself have come to the conclusion that a `switch`
statement is almost always superior to a statement that includes any
`else if`.
The exceptions that I have found in our codebase are just these two:
* The `if else` is followed by an additional statement before the next
condition (separated by a `;`).
* The whole thing is within a `for` loop and `break` statements are
used. In this case, using `switch` would require tagging the `for`
loop, which probably tips the balance.
Why are `switch` statements more readable?
For one, fewer curly braces. But more importantly, the conditions all
have the same alignment, so the whole thing follows the natural flow
of going down a list of conditions. With `else if`, in contrast, all
conditions but the first are "hidden" behind `} else if `, harder to
spot and (for no good reason) presented differently from the first
condition.
I'm sure the aforemention wise coders can list even more reasons.
In any case, I like it so much that I have found myself recommending
it in code reviews. I would like to make it a habit in our code base,
without making it a hard requirement that we would test on the CI. But
for that, there has to be a role model, so this commit eliminates all
`if else` occurrences, unless it is autogenerated code or fits one of
the exceptions above.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
case test.err != "":
|
|
|
|
t.Errorf("Expected errror %s, got none", test.err)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
func TestQueryForStateSeries(t *testing.T) {
|
|
|
|
testError := errors.New("test error")
|
|
|
|
|
|
|
|
type testInput struct {
|
|
|
|
selectMockFunction func(sortSeries bool, hints *storage.SelectHints, matchers ...*labels.Matcher) storage.SeriesSet
|
|
|
|
expectedSeries storage.Series
|
|
|
|
expectedError error
|
|
|
|
}
|
|
|
|
|
|
|
|
tests := []testInput{
|
|
|
|
// Test for empty series.
|
|
|
|
{
|
|
|
|
selectMockFunction: func(sortSeries bool, hints *storage.SelectHints, matchers ...*labels.Matcher) storage.SeriesSet {
|
|
|
|
return storage.EmptySeriesSet()
|
|
|
|
},
|
|
|
|
expectedSeries: nil,
|
|
|
|
expectedError: nil,
|
|
|
|
},
|
|
|
|
// Test for error series.
|
|
|
|
{
|
|
|
|
selectMockFunction: func(sortSeries bool, hints *storage.SelectHints, matchers ...*labels.Matcher) storage.SeriesSet {
|
|
|
|
return storage.ErrSeriesSet(testError)
|
|
|
|
},
|
|
|
|
expectedSeries: nil,
|
|
|
|
expectedError: testError,
|
|
|
|
},
|
|
|
|
// Test for mock series.
|
|
|
|
{
|
|
|
|
selectMockFunction: func(sortSeries bool, hints *storage.SelectHints, matchers ...*labels.Matcher) storage.SeriesSet {
|
|
|
|
return storage.TestSeriesSet(storage.MockSeries(
|
|
|
|
[]int64{1, 2, 3},
|
|
|
|
[]float64{1, 2, 3},
|
|
|
|
[]string{"__name__", "ALERTS_FOR_STATE", "alertname", "TestRule", "severity", "critical"},
|
|
|
|
))
|
|
|
|
},
|
|
|
|
expectedSeries: storage.MockSeries(
|
|
|
|
[]int64{1, 2, 3},
|
|
|
|
[]float64{1, 2, 3},
|
|
|
|
[]string{"__name__", "ALERTS_FOR_STATE", "alertname", "TestRule", "severity", "critical"},
|
|
|
|
),
|
|
|
|
expectedError: nil,
|
|
|
|
},
|
|
|
|
}
|
|
|
|
|
|
|
|
testFunc := func(tst testInput) {
|
|
|
|
querier := &storage.MockQuerier{
|
|
|
|
SelectMockFunction: tst.selectMockFunction,
|
|
|
|
}
|
|
|
|
|
|
|
|
rule := NewAlertingRule(
|
|
|
|
"TestRule",
|
|
|
|
nil,
|
|
|
|
time.Minute,
|
|
|
|
0,
|
|
|
|
labels.FromStrings("severity", "critical"),
|
|
|
|
labels.EmptyLabels(), labels.EmptyLabels(), "", true, nil,
|
|
|
|
)
|
|
|
|
|
|
|
|
alert := &Alert{
|
|
|
|
State: 0,
|
|
|
|
Labels: labels.EmptyLabels(),
|
|
|
|
Annotations: labels.EmptyLabels(),
|
|
|
|
Value: 0,
|
|
|
|
ActiveAt: time.Time{},
|
|
|
|
FiredAt: time.Time{},
|
|
|
|
ResolvedAt: time.Time{},
|
|
|
|
LastSentAt: time.Time{},
|
|
|
|
ValidUntil: time.Time{},
|
|
|
|
}
|
|
|
|
|
|
|
|
series, err := rule.QueryforStateSeries(context.Background(), alert, querier)
|
|
|
|
|
|
|
|
require.Equal(t, tst.expectedSeries, series)
|
|
|
|
require.Equal(t, tst.expectedError, err)
|
|
|
|
}
|
|
|
|
|
|
|
|
for _, tst := range tests {
|
|
|
|
testFunc(tst)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// TestSendAlertsDontAffectActiveAlerts tests a fix for https://github.com/prometheus/prometheus/issues/11424.
|
|
|
|
func TestSendAlertsDontAffectActiveAlerts(t *testing.T) {
|
|
|
|
rule := NewAlertingRule(
|
|
|
|
"TestRule",
|
|
|
|
nil,
|
|
|
|
time.Minute,
|
|
|
|
0,
|
|
|
|
labels.FromStrings("severity", "critical"),
|
|
|
|
labels.EmptyLabels(), labels.EmptyLabels(), "", true, nil,
|
|
|
|
)
|
|
|
|
|
|
|
|
// Set an active alert.
|
|
|
|
lbls := labels.FromStrings("a1", "1")
|
|
|
|
h := lbls.Hash()
|
|
|
|
al := &Alert{State: StateFiring, Labels: lbls, ActiveAt: time.Now()}
|
|
|
|
rule.active[h] = al
|
|
|
|
|
|
|
|
expr, err := parser.ParseExpr("foo")
|
|
|
|
require.NoError(t, err)
|
|
|
|
rule.vector = expr
|
|
|
|
|
|
|
|
// The relabel rule reproduced the bug here.
|
|
|
|
opts := notifier.Options{
|
|
|
|
QueueCapacity: 1,
|
|
|
|
RelabelConfigs: []*relabel.Config{
|
|
|
|
{
|
|
|
|
SourceLabels: model.LabelNames{"a1"},
|
|
|
|
Regex: relabel.MustNewRegexp("(.+)"),
|
|
|
|
TargetLabel: "a1",
|
|
|
|
Replacement: "bug",
|
|
|
|
Action: "replace",
|
|
|
|
},
|
|
|
|
},
|
|
|
|
}
|
|
|
|
nm := notifier.NewManager(&opts, log.NewNopLogger())
|
|
|
|
|
|
|
|
f := SendAlerts(nm, "")
|
|
|
|
notifyFunc := func(ctx context.Context, expr string, alerts ...*Alert) {
|
|
|
|
require.Len(t, alerts, 1)
|
|
|
|
require.Equal(t, al, alerts[0])
|
|
|
|
f(ctx, expr, alerts...)
|
|
|
|
}
|
|
|
|
|
|
|
|
rule.sendAlerts(context.Background(), time.Now(), 0, 0, notifyFunc)
|
|
|
|
nm.Stop()
|
|
|
|
|
|
|
|
// The relabel rule changes a1=1 to a1=bug.
|
|
|
|
// But the labels with the AlertingRule should not be changed.
|
|
|
|
testutil.RequireEqual(t, labels.FromStrings("a1", "1"), rule.active[h].Labels)
|
|
|
|
}
|
|
|
|
|
|
|
|
func TestKeepFiringFor(t *testing.T) {
|
|
|
|
storage := promql.LoadedStorage(t, `
|
|
|
|
load 1m
|
|
|
|
http_requests{job="app-server", instance="0"} 75 85 70 70 10x5
|
|
|
|
`)
|
|
|
|
t.Cleanup(func() { storage.Close() })
|
|
|
|
|
|
|
|
expr, err := parser.ParseExpr(`http_requests > 50`)
|
|
|
|
require.NoError(t, err)
|
|
|
|
|
|
|
|
rule := NewAlertingRule(
|
|
|
|
"HTTPRequestRateHigh",
|
|
|
|
expr,
|
|
|
|
time.Minute,
|
|
|
|
time.Minute,
|
|
|
|
labels.EmptyLabels(),
|
|
|
|
labels.EmptyLabels(), labels.EmptyLabels(), "", true, nil,
|
|
|
|
)
|
|
|
|
|
|
|
|
results := []promql.Vector{
|
|
|
|
{
|
|
|
|
promql.Sample{
|
|
|
|
Metric: labels.FromStrings(
|
|
|
|
"__name__", "ALERTS",
|
|
|
|
"alertname", "HTTPRequestRateHigh",
|
|
|
|
"alertstate", "pending",
|
|
|
|
"instance", "0",
|
|
|
|
"job", "app-server",
|
|
|
|
),
|
promql: Separate `Point` into `FPoint` and `HPoint`
In other words: Instead of having a “polymorphous” `Point` that can
either contain a float value or a histogram value, use an `FPoint` for
floats and an `HPoint` for histograms.
This seemingly small change has a _lot_ of repercussions throughout
the codebase.
The idea here is to avoid the increase in size of `Point` arrays that
happened after native histograms had been added.
The higher-level data structures (`Sample`, `Series`, etc.) are still
“polymorphous”. The same idea could be applied to them, but at each
step the trade-offs needed to be evaluated.
The idea with this change is to do the minimum necessary to get back
to pre-histogram performance for functions that do not touch
histograms. Here are comparisons for the `changes` function. The test
data doesn't include histograms yet. Ideally, there would be no change
in the benchmark result at all.
First runtime v2.39 compared to directly prior to this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 542µs ± 1% +38.58% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 617µs ± 2% +36.48% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.36ms ± 2% +21.58% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 8.94ms ± 1% +14.21% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.30ms ± 1% +10.67% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.10ms ± 1% +11.82% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 11.8ms ± 1% +12.50% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 87.4ms ± 1% +12.63% (p=0.000 n=9+9)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 32.8ms ± 1% +8.01% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.6ms ± 2% +9.64% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 117ms ± 1% +11.69% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 876ms ± 1% +11.83% (p=0.000 n=9+10)
```
And then runtime v2.39 compared to after this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 547µs ± 1% +39.84% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 616µs ± 2% +36.15% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.26ms ± 1% +12.20% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 7.95ms ± 1% +1.59% (p=0.000 n=10+8)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.38ms ± 2% +13.49% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.02ms ± 1% +9.80% (p=0.000 n=10+9)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 10.8ms ± 1% +3.08% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 78.1ms ± 1% +0.58% (p=0.035 n=9+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 33.5ms ± 4% +10.18% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.0ms ± 1% +7.98% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 107ms ± 1% +1.92% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 775ms ± 1% -1.02% (p=0.019 n=9+9)
```
In summary, the runtime doesn't really improve with this change for
queries with just a few steps. For queries with many steps, this
commit essentially reinstates the old performance. This is good
because the many-step queries are the one that matter most (longest
absolute runtime).
In terms of allocations, though, this commit doesn't make a dent at
all (numbers not shown). The reason is that most of the allocations
happen in the sampleRingIterator (in the storage package), which has
to be addressed in a separate commit.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
F: 1,
|
|
|
|
},
|
|
|
|
},
|
|
|
|
{
|
|
|
|
promql.Sample{
|
|
|
|
Metric: labels.FromStrings(
|
|
|
|
"__name__", "ALERTS",
|
|
|
|
"alertname", "HTTPRequestRateHigh",
|
|
|
|
"alertstate", "firing",
|
|
|
|
"instance", "0",
|
|
|
|
"job", "app-server",
|
|
|
|
),
|
promql: Separate `Point` into `FPoint` and `HPoint`
In other words: Instead of having a “polymorphous” `Point` that can
either contain a float value or a histogram value, use an `FPoint` for
floats and an `HPoint` for histograms.
This seemingly small change has a _lot_ of repercussions throughout
the codebase.
The idea here is to avoid the increase in size of `Point` arrays that
happened after native histograms had been added.
The higher-level data structures (`Sample`, `Series`, etc.) are still
“polymorphous”. The same idea could be applied to them, but at each
step the trade-offs needed to be evaluated.
The idea with this change is to do the minimum necessary to get back
to pre-histogram performance for functions that do not touch
histograms. Here are comparisons for the `changes` function. The test
data doesn't include histograms yet. Ideally, there would be no change
in the benchmark result at all.
First runtime v2.39 compared to directly prior to this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 542µs ± 1% +38.58% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 617µs ± 2% +36.48% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.36ms ± 2% +21.58% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 8.94ms ± 1% +14.21% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.30ms ± 1% +10.67% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.10ms ± 1% +11.82% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 11.8ms ± 1% +12.50% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 87.4ms ± 1% +12.63% (p=0.000 n=9+9)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 32.8ms ± 1% +8.01% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.6ms ± 2% +9.64% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 117ms ± 1% +11.69% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 876ms ± 1% +11.83% (p=0.000 n=9+10)
```
And then runtime v2.39 compared to after this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 547µs ± 1% +39.84% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 616µs ± 2% +36.15% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.26ms ± 1% +12.20% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 7.95ms ± 1% +1.59% (p=0.000 n=10+8)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.38ms ± 2% +13.49% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.02ms ± 1% +9.80% (p=0.000 n=10+9)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 10.8ms ± 1% +3.08% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 78.1ms ± 1% +0.58% (p=0.035 n=9+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 33.5ms ± 4% +10.18% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.0ms ± 1% +7.98% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 107ms ± 1% +1.92% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 775ms ± 1% -1.02% (p=0.019 n=9+9)
```
In summary, the runtime doesn't really improve with this change for
queries with just a few steps. For queries with many steps, this
commit essentially reinstates the old performance. This is good
because the many-step queries are the one that matter most (longest
absolute runtime).
In terms of allocations, though, this commit doesn't make a dent at
all (numbers not shown). The reason is that most of the allocations
happen in the sampleRingIterator (in the storage package), which has
to be addressed in a separate commit.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
F: 1,
|
|
|
|
},
|
|
|
|
},
|
|
|
|
{
|
|
|
|
promql.Sample{
|
|
|
|
Metric: labels.FromStrings(
|
|
|
|
"__name__", "ALERTS",
|
|
|
|
"alertname", "HTTPRequestRateHigh",
|
|
|
|
"alertstate", "firing",
|
|
|
|
"instance", "0",
|
|
|
|
"job", "app-server",
|
|
|
|
),
|
promql: Separate `Point` into `FPoint` and `HPoint`
In other words: Instead of having a “polymorphous” `Point` that can
either contain a float value or a histogram value, use an `FPoint` for
floats and an `HPoint` for histograms.
This seemingly small change has a _lot_ of repercussions throughout
the codebase.
The idea here is to avoid the increase in size of `Point` arrays that
happened after native histograms had been added.
The higher-level data structures (`Sample`, `Series`, etc.) are still
“polymorphous”. The same idea could be applied to them, but at each
step the trade-offs needed to be evaluated.
The idea with this change is to do the minimum necessary to get back
to pre-histogram performance for functions that do not touch
histograms. Here are comparisons for the `changes` function. The test
data doesn't include histograms yet. Ideally, there would be no change
in the benchmark result at all.
First runtime v2.39 compared to directly prior to this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 542µs ± 1% +38.58% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 617µs ± 2% +36.48% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.36ms ± 2% +21.58% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 8.94ms ± 1% +14.21% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.30ms ± 1% +10.67% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.10ms ± 1% +11.82% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 11.8ms ± 1% +12.50% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 87.4ms ± 1% +12.63% (p=0.000 n=9+9)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 32.8ms ± 1% +8.01% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.6ms ± 2% +9.64% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 117ms ± 1% +11.69% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 876ms ± 1% +11.83% (p=0.000 n=9+10)
```
And then runtime v2.39 compared to after this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 547µs ± 1% +39.84% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 616µs ± 2% +36.15% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.26ms ± 1% +12.20% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 7.95ms ± 1% +1.59% (p=0.000 n=10+8)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.38ms ± 2% +13.49% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.02ms ± 1% +9.80% (p=0.000 n=10+9)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 10.8ms ± 1% +3.08% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 78.1ms ± 1% +0.58% (p=0.035 n=9+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 33.5ms ± 4% +10.18% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.0ms ± 1% +7.98% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 107ms ± 1% +1.92% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 775ms ± 1% -1.02% (p=0.019 n=9+9)
```
In summary, the runtime doesn't really improve with this change for
queries with just a few steps. For queries with many steps, this
commit essentially reinstates the old performance. This is good
because the many-step queries are the one that matter most (longest
absolute runtime).
In terms of allocations, though, this commit doesn't make a dent at
all (numbers not shown). The reason is that most of the allocations
happen in the sampleRingIterator (in the storage package), which has
to be addressed in a separate commit.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
F: 1,
|
|
|
|
},
|
|
|
|
},
|
|
|
|
{
|
|
|
|
promql.Sample{
|
|
|
|
Metric: labels.FromStrings(
|
|
|
|
"__name__", "ALERTS",
|
|
|
|
"alertname", "HTTPRequestRateHigh",
|
|
|
|
"alertstate", "firing",
|
|
|
|
"instance", "0",
|
|
|
|
"job", "app-server",
|
|
|
|
),
|
promql: Separate `Point` into `FPoint` and `HPoint`
In other words: Instead of having a “polymorphous” `Point` that can
either contain a float value or a histogram value, use an `FPoint` for
floats and an `HPoint` for histograms.
This seemingly small change has a _lot_ of repercussions throughout
the codebase.
The idea here is to avoid the increase in size of `Point` arrays that
happened after native histograms had been added.
The higher-level data structures (`Sample`, `Series`, etc.) are still
“polymorphous”. The same idea could be applied to them, but at each
step the trade-offs needed to be evaluated.
The idea with this change is to do the minimum necessary to get back
to pre-histogram performance for functions that do not touch
histograms. Here are comparisons for the `changes` function. The test
data doesn't include histograms yet. Ideally, there would be no change
in the benchmark result at all.
First runtime v2.39 compared to directly prior to this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 542µs ± 1% +38.58% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 617µs ± 2% +36.48% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.36ms ± 2% +21.58% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 8.94ms ± 1% +14.21% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.30ms ± 1% +10.67% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.10ms ± 1% +11.82% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 11.8ms ± 1% +12.50% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 87.4ms ± 1% +12.63% (p=0.000 n=9+9)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 32.8ms ± 1% +8.01% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.6ms ± 2% +9.64% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 117ms ± 1% +11.69% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 876ms ± 1% +11.83% (p=0.000 n=9+10)
```
And then runtime v2.39 compared to after this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 547µs ± 1% +39.84% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 616µs ± 2% +36.15% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.26ms ± 1% +12.20% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 7.95ms ± 1% +1.59% (p=0.000 n=10+8)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.38ms ± 2% +13.49% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.02ms ± 1% +9.80% (p=0.000 n=10+9)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 10.8ms ± 1% +3.08% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 78.1ms ± 1% +0.58% (p=0.035 n=9+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 33.5ms ± 4% +10.18% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.0ms ± 1% +7.98% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 107ms ± 1% +1.92% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 775ms ± 1% -1.02% (p=0.019 n=9+9)
```
In summary, the runtime doesn't really improve with this change for
queries with just a few steps. For queries with many steps, this
commit essentially reinstates the old performance. This is good
because the many-step queries are the one that matter most (longest
absolute runtime).
In terms of allocations, though, this commit doesn't make a dent at
all (numbers not shown). The reason is that most of the allocations
happen in the sampleRingIterator (in the storage package), which has
to be addressed in a separate commit.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
F: 1,
|
|
|
|
},
|
|
|
|
},
|
|
|
|
// From now on the alert should keep firing.
|
|
|
|
{
|
|
|
|
promql.Sample{
|
|
|
|
Metric: labels.FromStrings(
|
|
|
|
"__name__", "ALERTS",
|
|
|
|
"alertname", "HTTPRequestRateHigh",
|
|
|
|
"alertstate", "firing",
|
|
|
|
"instance", "0",
|
|
|
|
"job", "app-server",
|
|
|
|
),
|
promql: Separate `Point` into `FPoint` and `HPoint`
In other words: Instead of having a “polymorphous” `Point` that can
either contain a float value or a histogram value, use an `FPoint` for
floats and an `HPoint` for histograms.
This seemingly small change has a _lot_ of repercussions throughout
the codebase.
The idea here is to avoid the increase in size of `Point` arrays that
happened after native histograms had been added.
The higher-level data structures (`Sample`, `Series`, etc.) are still
“polymorphous”. The same idea could be applied to them, but at each
step the trade-offs needed to be evaluated.
The idea with this change is to do the minimum necessary to get back
to pre-histogram performance for functions that do not touch
histograms. Here are comparisons for the `changes` function. The test
data doesn't include histograms yet. Ideally, there would be no change
in the benchmark result at all.
First runtime v2.39 compared to directly prior to this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 542µs ± 1% +38.58% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 617µs ± 2% +36.48% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.36ms ± 2% +21.58% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 8.94ms ± 1% +14.21% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.30ms ± 1% +10.67% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.10ms ± 1% +11.82% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 11.8ms ± 1% +12.50% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 87.4ms ± 1% +12.63% (p=0.000 n=9+9)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 32.8ms ± 1% +8.01% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.6ms ± 2% +9.64% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 117ms ± 1% +11.69% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 876ms ± 1% +11.83% (p=0.000 n=9+10)
```
And then runtime v2.39 compared to after this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 547µs ± 1% +39.84% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 616µs ± 2% +36.15% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.26ms ± 1% +12.20% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 7.95ms ± 1% +1.59% (p=0.000 n=10+8)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.38ms ± 2% +13.49% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.02ms ± 1% +9.80% (p=0.000 n=10+9)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 10.8ms ± 1% +3.08% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 78.1ms ± 1% +0.58% (p=0.035 n=9+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 33.5ms ± 4% +10.18% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.0ms ± 1% +7.98% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 107ms ± 1% +1.92% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 775ms ± 1% -1.02% (p=0.019 n=9+9)
```
In summary, the runtime doesn't really improve with this change for
queries with just a few steps. For queries with many steps, this
commit essentially reinstates the old performance. This is good
because the many-step queries are the one that matter most (longest
absolute runtime).
In terms of allocations, though, this commit doesn't make a dent at
all (numbers not shown). The reason is that most of the allocations
happen in the sampleRingIterator (in the storage package), which has
to be addressed in a separate commit.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
F: 1,
|
|
|
|
},
|
|
|
|
},
|
|
|
|
}
|
|
|
|
|
|
|
|
baseTime := time.Unix(0, 0)
|
|
|
|
for i, result := range results {
|
|
|
|
t.Logf("case %d", i)
|
|
|
|
evalTime := baseTime.Add(time.Duration(i) * time.Minute)
|
promql: Separate `Point` into `FPoint` and `HPoint`
In other words: Instead of having a “polymorphous” `Point` that can
either contain a float value or a histogram value, use an `FPoint` for
floats and an `HPoint` for histograms.
This seemingly small change has a _lot_ of repercussions throughout
the codebase.
The idea here is to avoid the increase in size of `Point` arrays that
happened after native histograms had been added.
The higher-level data structures (`Sample`, `Series`, etc.) are still
“polymorphous”. The same idea could be applied to them, but at each
step the trade-offs needed to be evaluated.
The idea with this change is to do the minimum necessary to get back
to pre-histogram performance for functions that do not touch
histograms. Here are comparisons for the `changes` function. The test
data doesn't include histograms yet. Ideally, there would be no change
in the benchmark result at all.
First runtime v2.39 compared to directly prior to this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 542µs ± 1% +38.58% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 617µs ± 2% +36.48% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.36ms ± 2% +21.58% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 8.94ms ± 1% +14.21% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.30ms ± 1% +10.67% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.10ms ± 1% +11.82% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 11.8ms ± 1% +12.50% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 87.4ms ± 1% +12.63% (p=0.000 n=9+9)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 32.8ms ± 1% +8.01% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.6ms ± 2% +9.64% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 117ms ± 1% +11.69% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 876ms ± 1% +11.83% (p=0.000 n=9+10)
```
And then runtime v2.39 compared to after this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 547µs ± 1% +39.84% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 616µs ± 2% +36.15% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.26ms ± 1% +12.20% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 7.95ms ± 1% +1.59% (p=0.000 n=10+8)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.38ms ± 2% +13.49% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.02ms ± 1% +9.80% (p=0.000 n=10+9)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 10.8ms ± 1% +3.08% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 78.1ms ± 1% +0.58% (p=0.035 n=9+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 33.5ms ± 4% +10.18% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.0ms ± 1% +7.98% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 107ms ± 1% +1.92% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 775ms ± 1% -1.02% (p=0.019 n=9+9)
```
In summary, the runtime doesn't really improve with this change for
queries with just a few steps. For queries with many steps, this
commit essentially reinstates the old performance. This is good
because the many-step queries are the one that matter most (longest
absolute runtime).
In terms of allocations, though, this commit doesn't make a dent at
all (numbers not shown). The reason is that most of the allocations
happen in the sampleRingIterator (in the storage package), which has
to be addressed in a separate commit.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
result[0].T = timestamp.FromTime(evalTime)
|
|
|
|
res, err := rule.Eval(context.TODO(), evalTime, EngineQueryFunc(testEngine, storage), nil, 0)
|
|
|
|
require.NoError(t, err)
|
|
|
|
|
|
|
|
var filteredRes promql.Vector // After removing 'ALERTS_FOR_STATE' samples.
|
|
|
|
for _, smpl := range res {
|
|
|
|
smplName := smpl.Metric.Get("__name__")
|
|
|
|
if smplName == "ALERTS" {
|
|
|
|
filteredRes = append(filteredRes, smpl)
|
|
|
|
} else {
|
|
|
|
// If not 'ALERTS', it has to be 'ALERTS_FOR_STATE'.
|
|
|
|
require.Equal(t, "ALERTS_FOR_STATE", smplName)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
testutil.RequireEqual(t, result, filteredRes)
|
|
|
|
}
|
|
|
|
evalTime := baseTime.Add(time.Duration(len(results)) * time.Minute)
|
|
|
|
res, err := rule.Eval(context.TODO(), evalTime, EngineQueryFunc(testEngine, storage), nil, 0)
|
|
|
|
require.NoError(t, err)
|
|
|
|
require.Empty(t, res)
|
|
|
|
}
|
|
|
|
|
|
|
|
func TestPendingAndKeepFiringFor(t *testing.T) {
|
|
|
|
storage := promql.LoadedStorage(t, `
|
|
|
|
load 1m
|
|
|
|
http_requests{job="app-server", instance="0"} 75 10x10
|
|
|
|
`)
|
|
|
|
t.Cleanup(func() { storage.Close() })
|
|
|
|
|
|
|
|
expr, err := parser.ParseExpr(`http_requests > 50`)
|
|
|
|
require.NoError(t, err)
|
|
|
|
|
|
|
|
rule := NewAlertingRule(
|
|
|
|
"HTTPRequestRateHigh",
|
|
|
|
expr,
|
|
|
|
time.Minute,
|
|
|
|
time.Minute,
|
|
|
|
labels.EmptyLabels(),
|
|
|
|
labels.EmptyLabels(), labels.EmptyLabels(), "", true, nil,
|
|
|
|
)
|
|
|
|
|
|
|
|
result := promql.Sample{
|
|
|
|
Metric: labels.FromStrings(
|
|
|
|
"__name__", "ALERTS",
|
|
|
|
"alertname", "HTTPRequestRateHigh",
|
|
|
|
"alertstate", "pending",
|
|
|
|
"instance", "0",
|
|
|
|
"job", "app-server",
|
|
|
|
),
|
promql: Separate `Point` into `FPoint` and `HPoint`
In other words: Instead of having a “polymorphous” `Point` that can
either contain a float value or a histogram value, use an `FPoint` for
floats and an `HPoint` for histograms.
This seemingly small change has a _lot_ of repercussions throughout
the codebase.
The idea here is to avoid the increase in size of `Point` arrays that
happened after native histograms had been added.
The higher-level data structures (`Sample`, `Series`, etc.) are still
“polymorphous”. The same idea could be applied to them, but at each
step the trade-offs needed to be evaluated.
The idea with this change is to do the minimum necessary to get back
to pre-histogram performance for functions that do not touch
histograms. Here are comparisons for the `changes` function. The test
data doesn't include histograms yet. Ideally, there would be no change
in the benchmark result at all.
First runtime v2.39 compared to directly prior to this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 542µs ± 1% +38.58% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 617µs ± 2% +36.48% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.36ms ± 2% +21.58% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 8.94ms ± 1% +14.21% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.30ms ± 1% +10.67% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.10ms ± 1% +11.82% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 11.8ms ± 1% +12.50% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 87.4ms ± 1% +12.63% (p=0.000 n=9+9)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 32.8ms ± 1% +8.01% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.6ms ± 2% +9.64% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 117ms ± 1% +11.69% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 876ms ± 1% +11.83% (p=0.000 n=9+10)
```
And then runtime v2.39 compared to after this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 547µs ± 1% +39.84% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 616µs ± 2% +36.15% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.26ms ± 1% +12.20% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 7.95ms ± 1% +1.59% (p=0.000 n=10+8)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.38ms ± 2% +13.49% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.02ms ± 1% +9.80% (p=0.000 n=10+9)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 10.8ms ± 1% +3.08% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 78.1ms ± 1% +0.58% (p=0.035 n=9+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 33.5ms ± 4% +10.18% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.0ms ± 1% +7.98% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 107ms ± 1% +1.92% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 775ms ± 1% -1.02% (p=0.019 n=9+9)
```
In summary, the runtime doesn't really improve with this change for
queries with just a few steps. For queries with many steps, this
commit essentially reinstates the old performance. This is good
because the many-step queries are the one that matter most (longest
absolute runtime).
In terms of allocations, though, this commit doesn't make a dent at
all (numbers not shown). The reason is that most of the allocations
happen in the sampleRingIterator (in the storage package), which has
to be addressed in a separate commit.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
F: 1,
|
|
|
|
}
|
|
|
|
|
|
|
|
baseTime := time.Unix(0, 0)
|
promql: Separate `Point` into `FPoint` and `HPoint`
In other words: Instead of having a “polymorphous” `Point` that can
either contain a float value or a histogram value, use an `FPoint` for
floats and an `HPoint` for histograms.
This seemingly small change has a _lot_ of repercussions throughout
the codebase.
The idea here is to avoid the increase in size of `Point` arrays that
happened after native histograms had been added.
The higher-level data structures (`Sample`, `Series`, etc.) are still
“polymorphous”. The same idea could be applied to them, but at each
step the trade-offs needed to be evaluated.
The idea with this change is to do the minimum necessary to get back
to pre-histogram performance for functions that do not touch
histograms. Here are comparisons for the `changes` function. The test
data doesn't include histograms yet. Ideally, there would be no change
in the benchmark result at all.
First runtime v2.39 compared to directly prior to this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 542µs ± 1% +38.58% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 617µs ± 2% +36.48% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.36ms ± 2% +21.58% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 8.94ms ± 1% +14.21% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.30ms ± 1% +10.67% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.10ms ± 1% +11.82% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 11.8ms ± 1% +12.50% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 87.4ms ± 1% +12.63% (p=0.000 n=9+9)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 32.8ms ± 1% +8.01% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.6ms ± 2% +9.64% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 117ms ± 1% +11.69% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 876ms ± 1% +11.83% (p=0.000 n=9+10)
```
And then runtime v2.39 compared to after this commit:
```
name old time/op new time/op delta
RangeQuery/expr=changes(a_one[1d]),steps=1-16 391µs ± 2% 547µs ± 1% +39.84% (p=0.000 n=9+8)
RangeQuery/expr=changes(a_one[1d]),steps=10-16 452µs ± 2% 616µs ± 2% +36.15% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_one[1d]),steps=100-16 1.12ms ± 1% 1.26ms ± 1% +12.20% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_one[1d]),steps=1000-16 7.83ms ± 1% 7.95ms ± 1% +1.59% (p=0.000 n=10+8)
RangeQuery/expr=changes(a_ten[1d]),steps=1-16 2.98ms ± 0% 3.38ms ± 2% +13.49% (p=0.000 n=9+10)
RangeQuery/expr=changes(a_ten[1d]),steps=10-16 3.66ms ± 1% 4.02ms ± 1% +9.80% (p=0.000 n=10+9)
RangeQuery/expr=changes(a_ten[1d]),steps=100-16 10.5ms ± 0% 10.8ms ± 1% +3.08% (p=0.000 n=8+10)
RangeQuery/expr=changes(a_ten[1d]),steps=1000-16 77.6ms ± 1% 78.1ms ± 1% +0.58% (p=0.035 n=9+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1-16 30.4ms ± 2% 33.5ms ± 4% +10.18% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=10-16 37.1ms ± 2% 40.0ms ± 1% +7.98% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=100-16 105ms ± 1% 107ms ± 1% +1.92% (p=0.000 n=10+10)
RangeQuery/expr=changes(a_hundred[1d]),steps=1000-16 783ms ± 3% 775ms ± 1% -1.02% (p=0.019 n=9+9)
```
In summary, the runtime doesn't really improve with this change for
queries with just a few steps. For queries with many steps, this
commit essentially reinstates the old performance. This is good
because the many-step queries are the one that matter most (longest
absolute runtime).
In terms of allocations, though, this commit doesn't make a dent at
all (numbers not shown). The reason is that most of the allocations
happen in the sampleRingIterator (in the storage package), which has
to be addressed in a separate commit.
Signed-off-by: beorn7 <beorn@grafana.com>
2 years ago
|
|
|
result.T = timestamp.FromTime(baseTime)
|
|
|
|
res, err := rule.Eval(context.TODO(), baseTime, EngineQueryFunc(testEngine, storage), nil, 0)
|
|
|
|
require.NoError(t, err)
|
|
|
|
|
|
|
|
require.Len(t, res, 2)
|
|
|
|
for _, smpl := range res {
|
|
|
|
smplName := smpl.Metric.Get("__name__")
|
|
|
|
if smplName == "ALERTS" {
|
|
|
|
testutil.RequireEqual(t, result, smpl)
|
|
|
|
} else {
|
|
|
|
// If not 'ALERTS', it has to be 'ALERTS_FOR_STATE'.
|
|
|
|
require.Equal(t, "ALERTS_FOR_STATE", smplName)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
evalTime := baseTime.Add(time.Minute)
|
|
|
|
res, err = rule.Eval(context.TODO(), evalTime, EngineQueryFunc(testEngine, storage), nil, 0)
|
|
|
|
require.NoError(t, err)
|
|
|
|
require.Empty(t, res)
|
|
|
|
}
|
|
|
|
|
|
|
|
// TestAlertingEvalWithOrigin checks that the alerting rule details are passed through the context.
|
|
|
|
func TestAlertingEvalWithOrigin(t *testing.T) {
|
|
|
|
ctx := context.Background()
|
|
|
|
now := time.Now()
|
|
|
|
|
|
|
|
const (
|
|
|
|
name = "my-recording-rule"
|
|
|
|
query = `count(metric{foo="bar"}) > 0`
|
|
|
|
)
|
|
|
|
var (
|
|
|
|
detail RuleDetail
|
|
|
|
lbs = labels.FromStrings("test", "test")
|
|
|
|
)
|
|
|
|
|
|
|
|
expr, err := parser.ParseExpr(query)
|
|
|
|
require.NoError(t, err)
|
|
|
|
|
|
|
|
rule := NewAlertingRule(
|
|
|
|
name,
|
|
|
|
expr,
|
|
|
|
time.Second,
|
|
|
|
time.Minute,
|
|
|
|
lbs,
|
|
|
|
labels.EmptyLabels(),
|
|
|
|
labels.EmptyLabels(),
|
|
|
|
"",
|
|
|
|
true, log.NewNopLogger(),
|
|
|
|
)
|
|
|
|
|
|
|
|
_, err = rule.Eval(ctx, now, func(ctx context.Context, qs string, _ time.Time) (promql.Vector, error) {
|
|
|
|
detail = FromOriginContext(ctx)
|
|
|
|
return nil, nil
|
|
|
|
}, nil, 0)
|
|
|
|
|
|
|
|
require.NoError(t, err)
|
|
|
|
require.Equal(t, detail, NewRuleDetail(rule))
|
|
|
|
}
|
|
|
|
|
|
|
|
func TestAlertingRule_SetNoDependentRules(t *testing.T) {
|
|
|
|
rule := NewAlertingRule(
|
|
|
|
"test",
|
|
|
|
&parser.NumberLiteral{Val: 1},
|
|
|
|
time.Minute,
|
|
|
|
0,
|
|
|
|
labels.FromStrings("test", "test"),
|
|
|
|
labels.EmptyLabels(),
|
|
|
|
labels.EmptyLabels(),
|
|
|
|
"",
|
|
|
|
true, log.NewNopLogger(),
|
|
|
|
)
|
|
|
|
require.False(t, rule.NoDependentRules())
|
|
|
|
|
|
|
|
rule.SetNoDependentRules(false)
|
|
|
|
require.False(t, rule.NoDependentRules())
|
|
|
|
|
|
|
|
rule.SetNoDependentRules(true)
|
|
|
|
require.True(t, rule.NoDependentRules())
|
|
|
|
}
|
|
|
|
|
|
|
|
func TestAlertingRule_SetNoDependencyRules(t *testing.T) {
|
|
|
|
rule := NewAlertingRule(
|
|
|
|
"test",
|
|
|
|
&parser.NumberLiteral{Val: 1},
|
|
|
|
time.Minute,
|
|
|
|
0,
|
|
|
|
labels.FromStrings("test", "test"),
|
|
|
|
labels.EmptyLabels(),
|
|
|
|
labels.EmptyLabels(),
|
|
|
|
"",
|
|
|
|
true, log.NewNopLogger(),
|
|
|
|
)
|
|
|
|
require.False(t, rule.NoDependencyRules())
|
|
|
|
|
|
|
|
rule.SetNoDependencyRules(false)
|
|
|
|
require.False(t, rule.NoDependencyRules())
|
|
|
|
|
|
|
|
rule.SetNoDependencyRules(true)
|
|
|
|
require.True(t, rule.NoDependencyRules())
|
|
|
|
}
|