gh-138122: Validate base frame before caching in remote debugging frame cache #142852

pablogsal · 2025-12-17T03:02:46Z

The frame cache in the remote debugging module was storing frame chains
without validating that they reached the base frame. This could happen
when a frame chain was interrupted or when the process state changed
during reading, resulting in incomplete stacks being cached. Subsequent
samples that hit the cache would then produce flamegraphs that didn't
reach the bottom of the call stack.

The fix passes base_frame_addr through to process_frame_chain() which
already has validation logic to ensure the frame walk terminates at the
expected sentinel frame. By enabling this validation in the caching code
path and tracking whether we've confirmed reaching the base frame, we
now only store complete frame chains in the cache. When extending from
cached data, we trust that the cached frames were already validated at
storage time, maintaining the invariant that cached stacks are always
complete.

An integration test using deeply nested generators that oscillate the
stack depth is added to verify that all sampled stacks contain the entry
point function. This catches regressions where incomplete stacks might
be cached and returned.

Issue: Implement PEP 799 – A dedicated profiling package for organizing Python profiling tools #138122

bedevere-bot · 2025-12-17T03:13:20Z

🤖 New build scheduled with the buildbot fleet by @pablogsal for commit 224dede 🤖

Results will be shown at:

https://buildbot.python.org/all/#/grid?branch=refs%2Fpull%2F142852%2Fmerge

If you want to schedule another build, you need to add the 🔨 test-with-buildbots label again.

Lib/test/test_profiling/test_sampling_profiler/test_integration.py

…ng frame cache The frame cache in the remote debugging module was storing frame chains without validating that they reached the base frame. This could happen when a frame chain was interrupted or when the process state changed during reading, resulting in incomplete stacks being cached. Subsequent samples that hit the cache would then produce flamegraphs that didn't reach the bottom of the call stack. The fix passes base_frame_addr through to process_frame_chain() which already has validation logic to ensure the frame walk terminates at the expected sentinel frame. By enabling this validation in the caching code path and tracking whether we've confirmed reaching the base frame, we now only store complete frame chains in the cache. When extending from cached data, we trust that the cached frames were already validated at storage time, maintaining the invariant that cached stacks are always complete. An integration test using deeply nested generators that oscillate the stack depth is added to verify that all sampled stacks contain the entry point function. This catches regressions where incomplete stacks might be cached and returned.

bedevere-bot · 2025-12-17T14:14:07Z

🤖 New build scheduled with the buildbot fleet by @pablogsal for commit 2cd7686 🤖

Results will be shown at:

https://buildbot.python.org/all/#/grid?branch=refs%2Fpull%2F142852%2Fmerge

If you want to schedule another build, you need to add the 🔨 test-with-buildbots label again.

bedevere-app bot added the awaiting core review label Dec 17, 2025

bedevere-app bot mentioned this pull request Dec 17, 2025

Implement PEP 799 – A dedicated profiling package for organizing Python profiling tools #138122

Open

pablogsal force-pushed the tachyon-frame-cache branch 2 times, most recently from d3ffd62 to 224dede Compare December 17, 2025 03:11

pablogsal added the 🔨 test-with-buildbots Test PR w/ buildbots; report in status section label Dec 17, 2025

bedevere-bot removed the 🔨 test-with-buildbots Test PR w/ buildbots; report in status section label Dec 17, 2025