chore: handle interpreters with multiple ranges

[korniltsev]

alpine:3.20 python3.12 has a separate _PyEval_EvalFrameDefault _PyEval_EvalFrameDefault.cold

readelf -s -W /usr/lib/debug/usr/lib/libpython3.12.so.1.0.debug | grep EvalFrameDef
  2143: 0000000000171780   362 FUNC    LOCAL  DEFAULT    9 _PyEval_EvalFrameDefault.localalias
  2234: 0000000000088a9a 56543 FUNC    LOCAL  DEFAULT    9 _PyEval_EvalFrameDefault.cold
 43791: 0000000000171780   362 FUNC    GLOBAL DEFAULT    9 _PyEval_EvalFrameDefault

Therefore ebpf code only assumes the interpreter function is only 362 bytes long and fails to select python unwinder as the meat of the interpreter is in the cold one.

This PR is a second attempt of #414 (comment)

Specifically

Add support for handling interpreter functions with two ranges. Both in userspace and kernel. Without actually providing second range just yet.

This is a preparation for an actual fix of #416 which will be submitted as a followup

[korniltsev]

hehe I think hit the 4.19 kernel insn limit.

kprobe/unwind_native has 4111 instructions

[korniltsev]

I removed 4.19 tests in the light of #440 Edit: rebased on top of main

[christos68k]

Can we add a comment here (or a few lines later, when we check for len(offetRanges) == 2) clarifying why len(offsetRanges) can be 1 or 2 (I know it's in the eBPF code but it's not obvious here)?

[korniltsev]

Thanks. I've updated the comment in the ebpf code and duplicated it into go code.

[korniltsev]

@christos68k Thanks for the review. I will address your comments on Monday.

[christos68k]

LGTM (didn't test)

[korniltsev]

@christos68k Thank you for your review

(didn't test)

One can test it like this - add an extra range manually
For example use python from the current alpine:3.20 docker image

diff --git a/interpreter/python/python.go b/interpreter/python/python.go
index 7a1a6cc..1e25e37 100644
--- a/interpreter/python/python.go
+++ b/interpreter/python/python.go
@@ -858,5 +858,6 @@ func findInterpreterRanges(info *interpreter.LoaderInfo) ([]util.Range, error) {
        // TODO(korniltsev): find cold ranges
        // see tools/coredump/testdata/amd64/alpine320-nobuildid.json
        // https://github.com/open-telemetry/opentelemetry-ebpf-profiler/issues/416
+       interpRanges = append(interpRanges, util.Range{Start: 0x8a934, End: 0x8a934 + 56456}) // obtained with readelf from the debug elf
        return interpRanges, nil
 }

Before:

After:

I have a proposal branch that recovers cold ranges automatically here main...grafana:opentelemetry-ebpf-profiler:python-interpreter-cold-chunk-fix-part2
The branch is big, I will be submitting the proposal in two separate followups later. (the first chunk is #447 )

[fabled]

This adds ebpf code and makes it slower. And there's no users for this yet. Could this be commented out? Perhaps also split this out to a function to make this a bit more readable. Then the commented code could be in a #if 0 instead.

[fabled]

If you anticipate the follow up PR to happen soon, we can already leave this test enabled. Maybe still separate the test to a helper function?

[korniltsev]

Extracted to a separate function. I prefer to leave the test enabled.

[fabled]

I'd rather add the skip field to the json file perhaps as a string, where the string is the link to the ticket explaining why it fails.

[fabled]

perhaps rename the file to python3-alpine320-nobuildid.json or similar to include the interpreter in the name

[fabled]

rename this too to include python312