client_golang/prometheus/go_collector_latest_test.go

371 lines
10 KiB
Go
Raw Normal View History

Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
// Copyright 2021 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.
//go:build go1.17
// +build go1.17
package prometheus
import (
"math"
"reflect"
"runtime"
"runtime/metrics"
"sync"
"testing"
"github.com/prometheus/client_golang/prometheus/internal"
dto "github.com/prometheus/client_model/go"
)
func TestRmForMemStats(t *testing.T) {
if got, want := len(bestEffortLookupRM(rmForMemStats)), len(rmForMemStats); got != want {
t.Errorf("got %d, want %d metrics", got, want)
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
}
}
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
func expectedBaseMetrics() map[string]struct{} {
metrics := map[string]struct{}{}
b := newBaseGoCollector()
for _, m := range []string{
b.gcDesc.fqName,
b.goInfoDesc.fqName,
b.goroutinesDesc.fqName,
b.gcLastTimeDesc.fqName,
b.threadsDesc.fqName,
} {
metrics[m] = struct{}{}
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
}
return metrics
}
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
func addExpectedRuntimeMemStats(metrics map[string]struct{}) map[string]struct{} {
for _, m := range goRuntimeMemStats() {
metrics[m.desc.fqName] = struct{}{}
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
}
return metrics
}
func addExpectedRuntimeMetrics(metrics map[string]struct{}) map[string]struct{} {
for _, m := range expectedRuntimeMetrics {
metrics[m] = struct{}{}
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
}
return metrics
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
}
func TestGoCollector(t *testing.T) {
for _, tcase := range []struct {
collections uint32
expectedFQNameSet map[string]struct{}
}{
{
collections: 0,
expectedFQNameSet: expectedBaseMetrics(),
},
{
collections: goRuntimeMemStatsCollection,
expectedFQNameSet: addExpectedRuntimeMemStats(expectedBaseMetrics()),
},
{
collections: goRuntimeMetricsCollection,
expectedFQNameSet: addExpectedRuntimeMetrics(expectedBaseMetrics()),
},
{
collections: goRuntimeMemStatsCollection | goRuntimeMetricsCollection,
expectedFQNameSet: addExpectedRuntimeMemStats(addExpectedRuntimeMetrics(expectedBaseMetrics())),
},
} {
if ok := t.Run("", func(t *testing.T) {
goMetrics := collectGoMetrics(t, tcase.collections)
goMetricSet := make(map[string]Metric)
for _, m := range goMetrics {
goMetricSet[m.Desc().fqName] = m
}
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
for i := range goMetrics {
name := goMetrics[i].Desc().fqName
if _, ok := tcase.expectedFQNameSet[name]; !ok {
t.Errorf("found unpexpected metric %s", name)
continue
}
}
// Now iterate over the expected metrics and look for removals.
for expectedName := range tcase.expectedFQNameSet {
if _, ok := goMetricSet[expectedName]; !ok {
t.Errorf("missing expected metric %s in collection", expectedName)
continue
}
}
}); !ok {
return
}
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
}
}
var sink interface{}
func TestBatchHistogram(t *testing.T) {
goMetrics := collectGoMetrics(t, defaultGoCollections)
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
var mhist Metric
for _, m := range goMetrics {
if m.Desc().fqName == "go_gc_heap_allocs_by_size_bytes_total" {
mhist = m
break
}
}
if mhist == nil {
t.Fatal("failed to find metric to test")
}
hist, ok := mhist.(*batchHistogram)
if !ok {
t.Fatal("found metric is not a runtime/metrics histogram")
}
// Make a bunch of allocations then do another collection.
//
// The runtime/metrics API tries to reuse memory where possible,
// so make sure that we didn't hang on to any of that memory in
// hist.
countsCopy := make([]uint64, len(hist.counts))
copy(countsCopy, hist.counts)
for i := 0; i < 100; i++ {
sink = make([]byte, 128)
}
collectGoMetrics(t, defaultGoCollections)
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
for i, v := range hist.counts {
if v != countsCopy[i] {
t.Error("counts changed during new collection")
break
}
}
// Get the runtime/metrics copy.
s := []metrics.Sample{
{Name: "/gc/heap/allocs-by-size:bytes"},
}
metrics.Read(s)
rmHist := s[0].Value.Float64Histogram()
wantBuckets := internal.RuntimeMetricsBucketsForUnit(rmHist.Buckets, "bytes")
// runtime/metrics histograms always have a +Inf bucket and are lower
// bound inclusive. In contrast, we have an implicit +Inf bucket and
// are upper bound inclusive, so we can chop off the first bucket
// (since the conversion to upper bound inclusive will shift all buckets
// down one index) and the +Inf for the last bucket.
wantBuckets = wantBuckets[1 : len(wantBuckets)-1]
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
// Check to make sure the output proto makes sense.
pb := &dto.Metric{}
hist.Write(pb)
if math.IsInf(pb.Histogram.Bucket[len(pb.Histogram.Bucket)-1].GetUpperBound(), +1) {
t.Errorf("found +Inf bucket")
}
if got := len(pb.Histogram.Bucket); got != len(wantBuckets) {
t.Errorf("got %d buckets in protobuf, want %d", got, len(wantBuckets))
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
}
for i, bucket := range pb.Histogram.Bucket {
// runtime/metrics histograms are lower-bound inclusive, but we're
// upper-bound inclusive. So just make sure the new inclusive upper
// bound is somewhere close by (in some cases it's equal).
wantBound := wantBuckets[i]
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
if gotBound := *bucket.UpperBound; (wantBound-gotBound)/wantBound > 0.001 {
t.Errorf("got bound %f, want within 0.1%% of %f", gotBound, wantBound)
}
// Make sure counts are cumulative. Because of the consistency guarantees
// made by the runtime/metrics package, we're really not guaranteed to get
// anything even remotely the same here.
if i > 0 && *bucket.CumulativeCount < *pb.Histogram.Bucket[i-1].CumulativeCount {
t.Error("cumulative counts are non-monotonic")
}
}
}
func collectGoMetrics(t *testing.T, enabledCollections uint32) []Metric {
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
t.Helper()
c := NewGoCollector(func(o *GoCollectorOptions) {
o.EnabledCollections = enabledCollections
}).(*goCollector)
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
// Collect all metrics.
ch := make(chan Metric)
var wg sync.WaitGroup
var metrics []Metric
wg.Add(1)
go func() {
defer wg.Done()
for metric := range ch {
metrics = append(metrics, metric)
}
}()
c.Collect(ch)
close(ch)
wg.Wait()
return metrics
}
func TestMemStatsEquivalence(t *testing.T) {
var msReal, msFake runtime.MemStats
descs := bestEffortLookupRM(rmForMemStats)
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
samples := make([]metrics.Sample, len(descs))
samplesMap := make(map[string]*metrics.Sample)
for i := range descs {
samples[i].Name = descs[i].Name
samplesMap[descs[i].Name] = &samples[i]
}
// Force a GC cycle to try to reach a clean slate.
runtime.GC()
// Populate msReal.
runtime.ReadMemStats(&msReal)
// Populate msFake and hope that no GC happened in between (:
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
metrics.Read(samples)
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
memStatsFromRM(&msFake, samplesMap)
// Iterate over them and make sure they're somewhat close.
msRealValue := reflect.ValueOf(msReal)
msFakeValue := reflect.ValueOf(msFake)
typ := msRealValue.Type()
for i := 0; i < msRealValue.NumField(); i++ {
fr := msRealValue.Field(i)
ff := msFakeValue.Field(i)
if typ.Field(i).Name == "PauseTotalNs" || typ.Field(i).Name == "LastGC" {
// We don't use those fields for metrics,
// thus we are not interested in having this filled.
continue
}
switch fr.Kind() {
// Fields which we are interested in are all uint64s.
// The only float64 field GCCPUFraction is by design omitted.
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
case reflect.Uint64:
vr := fr.Interface().(uint64)
vf := ff.Interface().(uint64)
if float64(vr-vf)/float64(vf) > 0.05 {
t.Errorf("wrong value for %s: got %d, want %d", typ.Field(i).Name, vf, vr)
}
}
}
}
func TestExpectedRuntimeMetrics(t *testing.T) {
goMetrics := collectGoMetrics(t, goRuntimeMetricsCollection)
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
goMetricSet := make(map[string]Metric)
for _, m := range goMetrics {
goMetricSet[m.Desc().fqName] = m
}
descs := metrics.All()
rmSet := make(map[string]struct{})
// Iterate over runtime-reported descriptions to find new metrics.
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
for i := range descs {
rmName := descs[i].Name
rmSet[rmName] = struct{}{}
// expectedRuntimeMetrics depends on Go version.
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
expFQName, ok := expectedRuntimeMetrics[rmName]
if !ok {
t.Errorf("found new runtime/metrics metric %s", rmName)
_, _, _, ok := internal.RuntimeMetricsToProm(&descs[i])
if !ok {
t.Errorf("new metric has name that can't be converted, or has an unsupported Kind")
}
continue
}
_, ok = goMetricSet[expFQName]
if !ok {
t.Errorf("existing runtime/metrics metric %s (expected fq name %s) not collected", rmName, expFQName)
continue
}
}
// Now iterate over the expected metrics and look for removals.
cardinality := 0
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
for rmName, fqName := range expectedRuntimeMetrics {
if _, ok := rmSet[rmName]; !ok {
t.Errorf("runtime/metrics metric %s removed", rmName)
continue
}
if _, ok := goMetricSet[fqName]; !ok {
t.Errorf("runtime/metrics metric %s not appearing under expected name %s", rmName, fqName)
continue
}
// While we're at it, check to make sure expected cardinality lines
// up, but at the point of the protobuf write to get as close to the
// real deal as possible.
//
// Note that we filter out non-runtime/metrics metrics here, because
// those are manually managed.
var m dto.Metric
if err := goMetricSet[fqName].Write(&m); err != nil {
t.Errorf("writing metric %s: %v", fqName, err)
continue
}
// N.B. These are the only fields populated by runtime/metrics metrics specifically.
// Other fields are populated by e.g. GCStats metrics.
switch {
case m.Counter != nil:
fallthrough
case m.Gauge != nil:
cardinality++
case m.Histogram != nil:
cardinality += len(m.Histogram.Bucket) + 3 // + sum, count, and +inf
default:
t.Errorf("unexpected protobuf structure for metric %s", fqName)
}
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
}
if t.Failed() {
t.Log("a new Go version may have been detected, please run")
t.Log("\tgo run gen_go_collector_metrics_set.go go1.X")
t.Log("where X is the Go version you are currently using")
}
expectCardinality := expectedRuntimeMetricsCardinality
if cardinality != expectCardinality {
t.Errorf("unexpected cardinality for runtime/metrics metrics: got %d, want %d", cardinality, expectCardinality)
}
Use the runtime/metrics package for the Go collector for 1.17+ (#955) This change introduces use of the runtime/metrics package in place of runtime.MemStats for Go 1.17 or later. The runtime/metrics package was introduced in Go 1.16, but not all the old metrics were accounted for until 1.17. The runtime/metrics package offers several advantages over using runtime.MemStats: * The list of metrics and their descriptions are machine-readable, allowing new metrics to get added without any additional work. * Detailed histogram-based metrics are now available, offering much deeper insights into the Go runtime. * The runtime/metrics API is significantly more efficient than runtime.MemStats, even with the additional metrics added, because it does not require any stop-the-world events. That being said, integrating the package comes with some caveats, some of which were discussed in #842. Namely: * The old MemStats-based metrics need to continue working, so they're exported under their old names backed by equivalent runtime/metrics metrics. * Earlier versions of Go need to continue working, so the old code remains, but behind a build tag. Finally, a few notes about the implementation: * This change includes a whole bunch of refactoring to avoid significant code duplication. * This change adds a new histogram metric type specifically optimized for runtime/metrics histograms. This type's methods also include additional logic to deal with differences in bounds conventions. * This change makes a whole bunch of decisions about how runtime/metrics names are translated. * This change adds a `go generate` script to generate a list of expected runtime/metrics names for a given Go version for auditing. Users of new versions of Go will transparently be allowed to use new metrics, however. Signed-off-by: Michael Anthony Knyszek <mknyszek@google.com>
2022-01-16 19:41:56 +03:00
}
func TestGoCollectorConcurrency(t *testing.T) {
c := NewGoCollector().(*goCollector)
// Set up multiple goroutines to Collect from the
// same GoCollector. In race mode with GOMAXPROCS > 1,
// this test should fail often if Collect is not
// concurrent-safe.
for i := 0; i < 4; i++ {
go func() {
ch := make(chan Metric)
go func() {
// Drain all metrics received until the
// channel is closed.
for range ch {
}
}()
c.Collect(ch)
close(ch)
}()
}
}