search

Joystream / atlas

Whitelabel consumer and publisher experience for Joystream
https://www.joystream.org
GNU General Public License v3.0
100 stars 44 forks source link

Are we tracking errors properly, are we monitoring? #6417

Open bedeho opened 3 months ago

bedeho commented 3 months ago

Background

I recently saw a video where someone was trying to sell creator tokens, and it consistently failed, they got the "somethign went wrong" dialogue. They were doing it from the portfolio screen, but I'm guessing the root cause was exact same issue as https://github.com/Joystream/atlas/issues/6415. Given that this limitation has been in place since when creator tokens were first shipped, I can only imagine how many times people have seen this error, yet we have been totally unaware of it. The only way we ran into this is by directly experiencing it ourselves, or catching screencasts of others. That is really bad.

Proposal

We already have a sentry dashboard for tracking errors, but I logged in and I saw nothing about it. So either we are failing on actually tracking this error in Sentry, or we are failing to monitor Sentry daily to actually process the issue. We cant wait for 100s of bad user interactions before we incidentally see the error in action first hand.

ikprk commented 3 months ago

I'd say there are few layers to that.

  1. We track a lot of events, I'd say it make it harder to catch important one.
  2. Some issues are not necessary, for example, the failed to send network performance events issue is non-actionable and introduces more spam than value to Sentry.
  3. Some issues are impossible to reproduce just with Sentry error details.
  4. To find the desired issue you have to know its' message that is only in the code. So traversing the feed is tricky without codebase knowledge.
bedeho commented 3 months ago

ok, thank you for the rundown, 1&2 was also my sense. I'm not so concerned about being able to reproduce, first and foremost detection is the key. If we see an error happen time and time again, we can escalate the effort to do more granular logging or error tracking once we know there is something we need to pin down. The real killer is when you don't even know there is a problem.

Can you prepare a plan to fix this?