Roast of Sentry

[rahulchhabria]

It’s roast time. In honor of our ego being too big, we’re asking you to knock us back down to size with a little feedback. Tell us everything you hate (or find mildly annoying) about our product. After that crushing blow, we’ll try our best to fix it.

To leave feedback, create a new issue in The Roast of Sentry discussion.

More details about the roast coming soon.

[jalaziz]

Well for one, the link 404s 😄

[rahulchhabria]

🤣 you got me. adding this discussion was on my to-do. was going to publish the actual roast site tomorrow. timing is hard 😔. but stay tuned. more coming soon.

[shaedrich]

Admit it, it was bait 😜

[rahulchhabria]

lol i wish i was that crafty. instead, i was like... oh no i better do this before i forget. 😆 but it's up now. check it out. we'll post more on our social channels next week.

[shaedrich]

We've all been there. Good to know that you folks over there at Sentry, who do great work, are only human after all 😉

[jalaziz]

Getting ahead of the site being up, the thing we find most annoying about the product are already issues / discussions:

• Multi-select environments when creating issue alerts #27725
• Greater flexibility when creating alert conditions #54732

Basically, the amount of alert rule duplication needed when using multiple services and environments is a massive PITA without automation.

[max-wittig]

I really appreciate Sentry for what it is and we offer it to internal employees (only internally!) with a multi-org setup
and using the self-hosted version, so the following:

• No OIDC login, with a provider that we specify in the admin setup (no OpenID!!!) so we created https://github.com/siemens/sentry-auth-oidc, which is used by many people. Sadly this causes the people in the whole instance to be able to login as member in all orgs so we had to limit the rights of members in organizations to zero.
• Custom patches in-order to automate Sentry for user management so that we can on-board people based on the groups they use in Gitlab
• Custom patches to disable functionality that is not working inside our environment (air-gapped) like slack, but that people still like to configure and try, which messes Sentry up
• Mostly a pain to update, no readable changelogs, just dumps of commits in the release text without knowing the actual changes
• No way to figure out, which organization is causing Sentry to use 100% CPU. Lack of monitoring
• A release way yanked, which had severe consequences and caused up to drop random data in the DB until we could restore our backup again: Postgres alpine image introduces unique constraint index errors self-hosted#3167
• We had to block certain nginx routes, so that other employees are not able to e.g. disconnect our OIDC provider (and complain that they can't login), increase rights of owners
• 55 containers running
• We would love to offer Sentry to more internal employees, but its just not stable and reliable enough for that regarding updates and monitoring

[frederikspang]

Goes against the 55 containers running - But default self-hosted default setup gave me connection problems with postgresql; Until I added pgbouncer to the mix.

[hwittenborn]

Also until the link manages to come back online...

It would be nicer if it was easier to deploy Sentry. I understand the need for multiple containers since there's so much Sentry needs to do what it does. But it would be really nice if I could simply have a Docker Compose file that didn't need updated via a script. Not sure what this would look like in terms of needed changes, but I would definitely prefer it more 😄.

[stephenlacy]

Native sourcemaps! Very tedious to navigate sourcemaps and dumps for Unreal Engine.

[bruno-garcia]

Dealing with native debug files is definitely tedious. That said we're always on a look out in ways we can make it less so.
Could you share some of your pain points? How do you manage it today?

What platforms are you building for? Windows, Android?

About 'navigating dumps' for Unreal could you expand what you mean by that?

[stephenlacy]

The debug symbols were not matched correctly, they seemed to not point to the error lines.

When working in UE we would love to use it for our errors such that things are divided by:

• Editor: when people are in the editor and they encounter warnings/errors/asserts (Unreal Ensures) that they are tracked
• Game: during playtests, warnings/errors/ensures are tracked

Viewing these errors with callstacks would be a great boost to productivity over the current approach which is:

• take screenshot of the warning/error/ensure
• copy/paste the text (which not a lot of people do)
• describe what you were doing

With the above, we still don't know the frequency of it during Editor sessions and Playtests.

[bruno-garcia]

The debug symbols were not matched correctly, they seemed to not point to the error lines.

Could you please open a ticket about this? A repro would be helpful. I wonder if it's a UE build tool issue or what could be.

Editor: when people are in the editor and they encounter warnings/errors/asserts (Unreal Ensures) that they are tracked
Game: during playtests, warnings/errors/ensures are tracked

We add a tag OOTB for editor vs game. Is it hard to tell apart crashes from one vs the other?

Viewing these errors with callstacks would be a great boost to productivity over the current approach which is:

Sorry I don't follow, which errors don't have callstacks?

Are you using the Unreal Engine SDK? https://github.com/getsentry/sentry-unreal
If so, which version of the SDK?
Or the Crash Report Client?

Also, which platforms are you targeting?
I ask because different 'backends' run depending on the platform.

Which UE Version are you on?

Please feel free to open tickets on https://github.com/getsentry/sentry-unreal so we can debug any issues you're having.
You shouldn't have to look at screenshot to debug issues and if line numbers don't point to the right place, we should be looking at that too.

Thank you for your roast ! :) we can always improve things, with your help

[catdogmat]

Metrics!

They looked like exactly what we needed, but then the beta ended.

[md384]

Same! We were looking for a way to get metrics from our web clients and sentry metrics were working great

[jvineet]

The custom metrics feature was a pretty great "push-metrics" solution, especially considering the ease of sending metrics from internal svcs as well as browsers/web clients etc.

[mamiu]

Self-hosting Sentry is a pain!
Give the on-prem community more love.
There are still tons of ways to monetize on-prem solutions. (Look at Odoo for instance)
I believe it'll grow your userbase significantly, because devs love Sentry!

[frederikspang]

More utilities for CPU and Memory usage! Allocations and CPU Time stats per span, to help diagnose performance even more! (Just a convention for it; https://develop.sentry.dev/sdk/telemetry/traces/span-data-conventions/)