search

getsentry / sentry

Developer-first error tracking and performance monitoring
https://sentry.io
Other
39.14k stars 4.2k forks source link

The number of crashes on the release page and the discover page differ significantly #70570

Open PayneFuRC opened 6 months ago

PayneFuRC commented 6 months ago

Environment

SaaS (https://sentry.io/)

Steps to Reproduce

  1. There are 9.1k crashes in release page image

  2. There are 3.8k crashes in discover page image

  3. There's no dropped crashes

  4. There's no filter rules

Expected Result

These two data sets should be the same.

Actual Result

These two data sets differ significantly.

Product Area

Releases

Link

No response

DSN

No response

Version

No response

getsantry[bot] commented 6 months ago

Assigning to @getsentry/support for routing ⏲️

PayneFuRC commented 6 months ago

And the data from other versions is the opposite, the number of crashes on the release page is lower. If the logic of beforeSend filtered it out, this situation should not exist. image (1) image (2)

getsantry[bot] commented 6 months ago

Routing to @getsentry/product-owners-releases for triage ⏲️

nhsiehgit commented 6 months ago

Hi @PayneFuRC

Thanks for the report! It looks like our releases page may be deriving this count from somewhere unexpectedly different than our discover page.

I might err on trusting the discover query for the crash rate for now while we dig to understand where this discrepancy is stemming from.

will create an internal ticket for us to track

nhsiehgit commented 6 months ago

Hi @PayneFuRC

Dug into this a bit more So it looks like the releases page is filtering these events for 'session.status' === 'crashed' as opposed to simply error.unhandled:true

So the releases count actually will be the more proper count! I'll work with the team to see how we can surface this in a more clear manner.

PayneFuRC commented 4 months ago

@nhsiehgit How can I filter for such events? It seems like there's no search condition('session.status' === 'crashed') for this. Will you support this kind of search condition in the future?

getsantry[bot] commented 4 months ago

Routing to @getsentry/product-owners-discover for triage ⏲️

nhsiehgit commented 4 months ago

@PayneFuRC

I'm asking around now. It seems like discover only queries for Errors/Transaction events, whereas sessions may simply be metrics with no underlying events

that being said, aside from the 2 pages querying different datasets, it sounds like there may also be a bug in how the electron SDK is reporting sessions 🤔 Apologies for the non-answer here. I'll try to update this issue as we get more information

PayneFuRC commented 4 months ago

@nhsiehgit I have also submitted a ticket to Sentry/Electron regarding the session crash report. Please help track this issue as well.

https://github.com/getsentry/sentry-electron/issues/941

chipzzz commented 3 months ago

@PayneFuRC , i'm having somewhat a similar problem with python sdk where events in Discover timechart as well events in the events list are almost always missing when we send them in. The only thing we see accurately increasing is the Since issue began count which always shows the correct count of all the events we send in.

This seems only apparent at low volume. I have no idea why Discover time chart and event list always shows less events than what was forwarded in. It's bizarre.

https://github.com/getsentry/self-hosted/issues/3235