Python: Trace GC collection pauses in Performance

[craigds]

Problem Statement

We are finding the occasional transaction in Performance (usually when looking at 'Outlier transactionss (p100)') that pauses for multiple seconds (sometimes >10s) in 'Missing instrumentation', or (more often) in a db query that (we're pretty sure) couldn't possibly be actually taking that long.

This is in a threaded web server, so all sorts of things could be going on:

• thread is not being scheduled due to file I/O or busy CPUs
• the GIL is locked
• Python's garbage collector is running and has paused the interpreter

It would be nice if Sentry could display some of these events in the Performance traces, especially the GC one.

Solution Brainstorm

The GC has some callback hooks that the sentry integration could use to track when GC is occurring: https://docs.python.org/3/library/gc.html#gc.callbacks

Could this be added to the sentry client library and reported as a span in the Performance pages?

[antonpirker]

Hey @craigds !

I think this is a good idea.

What do other people think: Would it be helpful to see spans in Sentrys "Performance" section that show when the garbage collector was running?

Give us your thoughts or a thumbs up so we can see how many people like this idea!