v1.12.0 race condition #968

nicolaasuni · 2022-01-19T18:24:52Z

After upgrading to: github.com/prometheus/client_golang v1.12.0
I am getting "WARNING: DATA RACE" while testing:
https://github.com/nexmoinc/gosrvlib/tree/main/pkg/metrics/prometheus

The race conditions do not appear in version 1.11.0 or any other previous version.

The race condition do not appear when disabling the t.Parallel() in the unit tests (e.g.: https://github.com/nexmoinc/gosrvlib/blob/main/pkg/metrics/prometheus/client_test.go#L153).

The relevant errors are:

==================
WARNING: DATA RACE
Read at 0x00c000598660 by goroutine 59:
 runtime.mapaccess2_faststr()
     /usr/local/go/src/runtime/map_faststr.go:107 +0x0
 runtime/metrics.runtime_readMetrics()
     /usr/local/go/src/runtime/metrics.go:563 +0xb9
 github.com/prometheus/client_golang/prometheus.(*Registry).Gather.func1()
     /home/user/go/pkg/mod/github.com/prometheus/client_golang@v1.12.0/prometheus/registry.go:446 +0x133

Previous write at 0x00c000598660 by goroutine 70:
 runtime.mapassign_faststr()
     /usr/local/go/src/runtime/map_faststr.go:202 +0x0
 runtime.initMetrics()
     /usr/local/go/src/runtime/metrics.go:67 +0x1c6
 github.com/prometheus/client_golang/prometheus.(*Registry).Gather.func1()
     /home/user/go/pkg/mod/github.com/prometheus/client_golang@v1.12.0/prometheus/registry.go:446 +0x133

Goroutine 59 (running) created at:
 github.com/prometheus/client_golang/prometheus.(*Registry).Gather()
     /home/user/go/pkg/mod/github.com/prometheus/client_golang@v1.12.0/prometheus/registry.go:538 +0x1137
 github.com/prometheus/client_golang/prometheus/testutil.GatherAndCount()
     /home/user/go/pkg/mod/github.com/prometheus/client_golang@v1.12.0/prometheus/testutil/testutil.go:139 +0x7a
 github.com/nexmoinc/gosrvlib/pkg/metrics/prometheus.TestIncLogLevelCounter()
     /home/user/go/src/github.com/nexmoinc/gosrvlib/pkg/metrics/prometheus/client_test.go:144 +0x157
 testing.tRunner()
     /usr/local/go/src/testing/testing.go:1259 +0x22f
 testing.(*T).Run·dwrap·21()
     /usr/local/go/src/testing/testing.go:1306 +0x47

Goroutine 70 (finished) created at:
 github.com/prometheus/client_golang/prometheus.(*Registry).Gather()
     /home/user/go/pkg/mod/github.com/prometheus/client_golang@v1.12.0/prometheus/registry.go:538 +0x1137
 github.com/prometheus/client_golang/prometheus/promhttp.HandlerFor.func1()
     /home/user/go/pkg/mod/github.com/prometheus/client_golang@v1.12.0/prometheus/promhttp/http.go:126 +0x11a
 net/http.HandlerFunc.ServeHTTP()
     /usr/local/go/src/net/http/server.go:2047 +0x4d
 github.com/prometheus/client_golang/prometheus/promhttp.InstrumentHandlerInFlight.func1()
     /home/user/go/pkg/mod/github.com/prometheus/client_golang@v1.12.0/prometheus/promhttp/instrument_server.go:40 +0x118
 net/http.HandlerFunc.ServeHTTP()
     /usr/local/go/src/net/http/server.go:2047 +0x4d
 github.com/prometheus/client_golang/prometheus/promhttp.InstrumentHandlerCounter.func1()
     /home/user/go/pkg/mod/github.com/prometheus/client_golang@v1.12.0/prometheus/promhttp/instrument_server.go:117 +0xe2
 net/http.HandlerFunc.ServeHTTP()
     /usr/local/go/src/net/http/server.go:2047 +0x4d
 net/http.Handler.ServeHTTP-fm()
     /usr/local/go/src/net/http/server.go:87 +0x75
 net/http.HandlerFunc.ServeHTTP()
     /usr/local/go/src/net/http/server.go:2047 +0x4d
 github.com/prometheus/client_golang/prometheus/promhttp.InstrumentHandlerRequestSize.func2()
     /home/user/go/pkg/mod/github.com/prometheus/client_golang@v1.12.0/prometheus/promhttp/instrument_server.go:204 +0xb3
 net/http.HandlerFunc.ServeHTTP()
     /usr/local/go/src/net/http/server.go:2047 +0x4d
 github.com/prometheus/client_golang/prometheus/promhttp.InstrumentHandlerResponseSize.func1()
     /home/user/go/pkg/mod/github.com/prometheus/client_golang@v1.12.0/prometheus/promhttp/instrument_server.go:239 +0xec
 net/http.HandlerFunc.ServeHTTP()
     /usr/local/go/src/net/http/server.go:2047 +0x4d
 github.com/prometheus/client_golang/prometheus/promhttp.InstrumentHandlerCounter.func1()
     /home/user/go/pkg/mod/github.com/prometheus/client_golang@v1.12.0/prometheus/promhttp/instrument_server.go:117 +0xe2
 net/http.HandlerFunc.ServeHTTP()
     /usr/local/go/src/net/http/server.go:2047 +0x4d
 github.com/prometheus/client_golang/prometheus/promhttp.InstrumentHandlerDuration.func2()
     /home/user/go/pkg/mod/github.com/prometheus/client_golang@v1.12.0/prometheus/promhttp/instrument_server.go:84 +0xca
 net/http.HandlerFunc.ServeHTTP()
     /usr/local/go/src/net/http/server.go:2047 +0x4d
 github.com/prometheus/client_golang/prometheus/promhttp.InstrumentHandlerInFlight.func1()
     /home/user/go/pkg/mod/github.com/prometheus/client_golang@v1.12.0/prometheus/promhttp/instrument_server.go:40 +0x118
 net/http.HandlerFunc.ServeHTTP()
     /usr/local/go/src/net/http/server.go:2047 +0x4d
 github.com/nexmoinc/gosrvlib/pkg/metrics/prometheus.TestInstrumentHandler()
     /home/user/go/src/github.com/nexmoinc/gosrvlib/pkg/metrics/prometheus/client_test.go:89 +0x3a1
 testing.tRunner()
     /usr/local/go/src/testing/testing.go:1259 +0x22f
 testing.(*T).Run·dwrap·21()
     /usr/local/go/src/testing/testing.go:1306 +0x47
==================
=== CONT  TestInstrumentDB
   testing.go:1152: race detected during execution of test
--- FAIL: TestInstrumentDB (0.03s)
=== CONT  TestIncLogLevelCounter
   testing.go:1152: race detected during execution of test
--- FAIL: TestIncLogLevelCounter (0.04s)
=== CONT  TestInstrumentRoundTripper
   testing.go:1152: race detected during execution of test
--- FAIL: TestInstrumentRoundTripper (0.06s)
=== CONT  TestInstrumentHandler
   testing.go:1152: race detected during execution of test
--- FAIL: TestInstrumentHandler (0.07s)
FAIL

The text was updated successfully, but these errors were encountered:

kakkoyun · 2022-01-20T05:30:46Z

Interesting 🤔 I do not recall any changes that could cause this. We'll look into it.

kakkoyun · 2022-01-20T05:41:14Z

Could you provide more about your current setup? OS? Arch? And the test command that you use?

nicolaasuni · 2022-01-20T07:29:50Z

@kakkoyun Thank you for looking into this.

The testing environment is:

Distributor ID:	Ubuntu
Description:	Ubuntu 20.04.3 LTS
Release:	20.04
Codename:	focal
Linux xxxx 5.4.0-96-generic #109-Ubuntu SMP Wed Jan 12 16:49:16 UTC 2022 x86_64 x86_64 x86_64 GNU/Linux

go version go1.17.6 linux/amd64

I have pushed the failing version on the "dependencies" branch:
https://github.com/nexmoinc/gosrvlib/tree/dependencies

You can test the prometheus package via the command:

gotest -race -failfast -v github.com/nexmoinc/gosrvlib/pkg/metrics/prometheus

beorn7 · 2022-01-20T13:45:59Z

I think the problem is that the newly implemented Go collector doesn't have a goroutine-safe Collect method. In particular, the call metrics.Read needs to be protected, see

client_golang/prometheus/go_collector_go117.go

Line 134 in 0108796

metrics.Read(c.rmSampleBuf)

/cc @mknyszek

mknyszek · 2022-01-20T16:06:12Z

Hm... metrics.Read is serialized (with only itself) internally, though. There may be other parts of Collect that are not properly serialized, though.

beorn7 · 2022-01-20T16:24:13Z

But metrics.Read gets c.rmSampleBuf passed in. So even if metrics.Read itself is serialized, c.rmSampleBuf is later ranged over while another goroutine might write to it.

mknyszek · 2022-01-20T16:34:18Z

Right! Thanks. OK, that should be an easy fix. runtime/metrics has a critical section on the order of microseconds (even with all the histograms). Is it safe to assume Collect is on average called at a much higher period? If so, a simple mutex will likely suffice.

beorn7 · 2022-01-20T18:46:34Z

I would probably go for a simple mutex for now. If there are smarter solutions, we can implement them later.

beorn7 · 2022-01-20T18:51:45Z

And yes, scraping can be assumed to be a relatively rare event (a few times a minute in usual cases, a few times a second in extreme cases).

mknyszek · 2022-01-20T19:08:13Z

Sent PR #969. Tests now pass in -race mode.

nicolaasuni · 2022-01-28T12:22:50Z

When are you planning to release and tag this fix?
Currently I cant update to the new v1.12.0 as it contains the race condition bug.

kakkoyun mentioned this issue Jan 21, 2022

Make the Go 1.17 collector thread-safe #969

Merged

kakkoyun closed this as completed in #969 Jan 21, 2022

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

v1.12.0 race condition #968

v1.12.0 race condition #968

nicolaasuni commented Jan 19, 2022

kakkoyun commented Jan 20, 2022

kakkoyun commented Jan 20, 2022

nicolaasuni commented Jan 20, 2022 •

edited

beorn7 commented Jan 20, 2022

mknyszek commented Jan 20, 2022

beorn7 commented Jan 20, 2022

mknyszek commented Jan 20, 2022

beorn7 commented Jan 20, 2022

beorn7 commented Jan 20, 2022

mknyszek commented Jan 20, 2022

nicolaasuni commented Jan 28, 2022

v1.12.0 race condition #968

v1.12.0 race condition #968

Comments

nicolaasuni commented Jan 19, 2022

kakkoyun commented Jan 20, 2022

kakkoyun commented Jan 20, 2022

nicolaasuni commented Jan 20, 2022 • edited

beorn7 commented Jan 20, 2022

mknyszek commented Jan 20, 2022

beorn7 commented Jan 20, 2022

mknyszek commented Jan 20, 2022

beorn7 commented Jan 20, 2022

beorn7 commented Jan 20, 2022

mknyszek commented Jan 20, 2022

nicolaasuni commented Jan 28, 2022

nicolaasuni commented Jan 20, 2022 •

edited