thumbor / remotecv / 39880988731

94%

Build Type

push

github

Committed by web-flow

Commit Message

Drop the dead Sentry excepthook and flush pyres job failures (#124)

ErrorHandler.install_handler returned early unless self.sentry_client was
set, and nothing set it, so the custom sys.excepthook, the after_fork
callback that re-ran it in each pyres child and the remotecv-version extra
were all dead code. Job failures reached Sentry only through the SDK's
logging integration, and only when the event went out within a few
milliseconds: the pyres child calls os._exit right after logging, which
skips the SDK's atexit flush. The hook is gone, the release falls back to
remotecv@<version> when the SDK detects none of its own (SENTRY_RELEASE,
the git revision, platform variables), and UniqueWorker.process flushes the
SDK before returning.

With 200 ms of send latency injected, a failed job produced no event on sentry-
sdk 1.45.1 or 2.72.0 before this change and one after it. The flush is a no-op
without a DSN and returns as soon as the queue is empty; when Sentry is slow or
down it waits up to shutdown_timeout (2 s by default), which a failing job can
now spend before its child exits and which counts against DETECTOR_TIMEOUT.
Uncaught exceptions keep going through the SDK's default excepthook
integration. This change is stacked on #121 and should merge after it.

Coverage Stats

8 of 8 new or added lines in 2 files covered. (100.0%)

1 existing line in 1 file now uncovered.

606 of 660 relevant lines covered (91.82%)

4.59 hits per line