fix(logs): add logs flush on crash#1404
Conversation
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
![seer-by-sentry[bot]](https://avatars.githubusercontent.com/in/801464?s=60&v=4){kind=small}
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
…crash # Conflicts: # CHANGELOG.md # tests/test_integration_crashpad.py
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
this eliminates boilerplate noise in the tests but also reduces the chance of forgetting it in a test.
… it since we flush anyway.
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
supervacuus
left a comment
supervacuus
left a comment
There was a problem hiding this comment.
LGTM ❤️
The only thing I ask you to do now is to reduce the logs or move them to the TRACE level.
… cleaning in the pytest configuration
this means when it does its clearing a TEST_CHECK cannot assume that a directory even exists (in which case clear_crash_marker would actually fail versus the case were only the marker file didn't exist).
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
|
@JoshuaMoelans, do you still want to add something here, or is this ready to merge from your point of view? |
|
@supervacuus good to go from my perspective! 🚀 |
* Add logs-on-crash * skip crashpad test on macos * add straight-to-disk writing of log envelopes if crashing * CHANGELOG.md * restore thread join for normal shutdown * directly take log item from httpserver log for crashpad test * update CHANGELOG.md * we know which envelope has the log * we **don't** know which envelope has the log * add more general request fetch from httpserver function * add logs return value checks * fix tsan global data race on logs validation * move database-path cleanup into the cmake build cache fixture this eliminates boilerplate noise in the tests but also reduces the chance of forgetting it in a test. * in the crash-safe logs flush detach the batcher thread and don't wake it since we flush anyway. * don't clean for unit-test crash_marker * don't detach in crash-safe but spin-lock in its flusher to acquire flushing. * add clean parameter to cmake fixture * ensure logs flushing in crash handlers happens within the handler sync bounds. * bound the spin lock in the logs flusher during crash-safe mode * exclude crashed_last_run unit-test from cleaning database paths * use an unsigned integer for the sleep_time of the crash-safe logs flusher spinlock * provide cross-platform cpu_relax and replace syscall sleep with spinners. * isolate `crash_marker` unit test and eliminate exceptions to database cleaning in the pytest configuration * platform path special case in crash_marker unit test * crashed_last_run can no longer assume that a database path exists this means when it does its clearing a TEST_CHECK cannot assume that a directory even exists (in which case clear_crash_marker would actually fail versus the case were only the marker file didn't exist). * clean up log statements * bump CI python to 3.12 * fix CHANGELOG.md --------- Co-authored-by: Mischan Toosarani-Hausberger <mischan@abovevacant.com>
2 participants
Fixes #1403
Part of #1398