Python stack unwinding fails, but with null error

[Garbett1]

Describe the bug
Parca agent stack unwinding fails to add the unwind table, but also has a null error which is surprising.

To Reproduce
Can't replicate with a minimal example yet, but happens consistently with the specific binary.

Logs

kubectl logs -n parca parca-agent-rb55j | grep 18846

level=debug name=parca-agent ts=2023-11-29T12:45:32.972851349Z caller=cpu.go:536 component=cpu_profiler msg="adding unwind tables" pid=18846
level=error name=parca-agent ts=2023-11-29T12:45:32.975991544Z caller=cpu.go:557 component=cpu_profiler msg="failed to add unwind table" pid=18846 err=null
level=debug name=parca-agent ts=2023-11-29T12:45:33.170072189Z caller=cpu.go:531 component=cpu_profiler msg="prefetching process info" pid=18846
level=debug name=parca-agent ts=2023-11-29T12:45:33.170377246Z caller=cpu.go:536 component=cpu_profiler msg="adding unwind tables" pid=18846
level=error name=parca-agent ts=2023-11-29T12:45:33.173663715Z caller=cpu.go:557 component=cpu_profiler msg="failed to add unwind table" pid=18846 err=null
level=debug name=parca-agent ts=2023-11-29T12:45:35.182409458Z caller=cpu.go:531 component=cpu_profiler msg="prefetching process info" pid=18846
level=debug name=parca-agent ts=2023-11-29T12:45:35.182471246Z caller=cpu.go:536 component=cpu_profiler msg="adding unwind tables" pid=18846
level=error name=parca-agent ts=2023-11-29T12:45:35.184726234Z caller=cpu.go:557 component=cpu_profiler msg="failed to add unwind table" pid=18846 err=null
level=debug name=parca-agent ts=2023-11-29T12:45:44.121114761Z caller=cpu.go:531 component=cpu_profiler msg="prefetching process info" pid=18846
level=debug name=parca-agent ts=2023-11-29T12:45:44.12125953Z caller=cpu.go:536 component=cpu_profiler msg="adding unwind tables" pid=18846
level=error name=parca-agent ts=2023-11-29T12:45:44.123850309Z caller=cpu.go:557 component=cpu_profiler msg="failed to add unwind table" pid=18846 err=null
level=debug name=parca-agent ts=2023-11-29T12:46:20.493137114Z caller=cpu.go:531 component=cpu_profiler msg="prefetching process info" pid=18846
level=debug name=parca-agent ts=2023-11-29T12:46:20.530767211Z caller=info.go:304 component=process_info msg="interpreter information fetched" interpreter=Python version=3.10.13 pid=18846
level=debug name=parca-agent ts=2023-11-29T12:46:20.53203637Z caller=maps.go:851 component=cpu_profiler component=bpf_maps msg="Python Version Offset" pid=18846 version_offset_index=1
level=debug name=parca-agent ts=2023-11-29T12:46:20.532145376Z caller=cpu.go:536 component=cpu_profiler msg="adding unwind tables" pid=18846
level=error name=parca-agent ts=2023-11-29T12:46:20.534201669Z caller=cpu.go:557 component=cpu_profiler msg="failed to add unwind table" pid=18846 err=null
level=debug name=parca-agent ts=2023-11-29T12:46:37.889658547Z caller=cpu.go:531 component=cpu_profiler msg="prefetching process info" pid=18846
level=debug name=parca-agent ts=2023-11-29T12:46:37.889877709Z caller=cpu.go:536 component=cpu_profiler msg="adding unwind tables" pid=18846
level=error name=parca-agent ts=2023-11-29T12:46:37.892896043Z caller=cpu.go:557 component=cpu_profiler msg="failed to add unwind table" pid=18846 err=null

Software (please complete the following information):

• Parca Agent Version: ghcr.io/parca-dev/parca-agent:main-1701258441-d828aaf4

** Workload:**

• Runtime (if applicable): Python 3.10.13
  ** Environment :**

• Linux Distribution: COS

• Linux Version: 5.15.120+

• Arch: amd64

• Kubernetes Version (if applicable): Server Version: version.Info{Major:"1", Minor:"25", GitVersion:"v1.25.13-gke.200", GitCommit:"f24d4fe14dcec505c2b37cc1e5b7024d971f6360", GitTreeState:"clean", BuildDate:"2023-08-25T09:26:18Z", GoVersion:"go1.20.7 X:boringcrypto", Compiler:"gc", Platform:"linux/amd64"}

• Container Runtime (if applicable): containerd 1.6.21

[brancz]

Huh, this is kind of odd.

parca-agent/pkg/profiler/cpu/cpu.go

Lines 551 to 573 in 7df448f

	level.Debug(p.logger).Log("msg", "adding unwind tables", "pid", pid)
	if err := p.bpfMaps.AddUnwindTableForProcess(pid, nil, true); err == nil {
	// Happy path.
	return
	}
	// Error handling,
	switch {
	case errors.Is(err, bpfmaps.ErrNeedMoreProfilingRounds):
	p.metrics.unwindTableAddErrors.WithLabelValues(labelNeedMoreProfilingRounds).Inc()
	level.Debug(p.logger).Log("msg", "PersistUnwindTable called to soon", "err", err)
	case errors.Is(err, os.ErrNotExist):
	p.metrics.unwindTableAddErrors.WithLabelValues(labelProcfsRace).Inc()
	level.Debug(p.logger).Log("msg", "failed to add unwind table due to a procfs race", "pid", pid, "err", err)
	case errors.Is(err, bpfmaps.ErrTooManyExecutableMappings):
	p.metrics.unwindTableAddErrors.WithLabelValues(labelTooManyMappings).Inc()
	level.Warn(p.logger).Log("msg", "failed to add unwind table due to having too many executable mappings", "pid", pid, "err", err)
	case errors.Is(err, buildid.ErrTextSectionNotFound):
	p.processErrorTracker.Track(pid, err)
	default:
	p.metrics.unwindTableAddErrors.WithLabelValues(labelOther).Inc()
	level.Error(p.logger).Log("msg", "failed to add unwind table", "pid", pid, "err", err)
	}

if err == nil we should already be returning before 🤔

[Tatskaari]

Are you falling a fowl of the typed nil variable nonsense?

[brancz]

Yes, definitely possible!