Is there any reason why Replay will be uploaded without errors?

[Zero-ruan]

Self-Hosted Version

"@sentry/react": "^7.72.0"

CPU Architecture

x86_64

Docker Version

20.10.16

Docker Compose Version

2.6.0

Steps to Reproduce

• Calling replay.start() again after using replay.stop() did not work, but after I added await replay.setupOnce() it became normal. I don’t know if it has anything to do with this.

Expected Result

• There shouldn't be any of these pointless Replays in existence right?

Actual Result

• There are no exceptions in console and network

Event ID

No response

[getsantry]

Assigning to @getsentry/support for routing ⏲️

[getsantry]

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

[Zero-ruan]

getsentry/sentry-javascript#8054 I found this and it seems to be the same question. However I am using ^7.72.0 .

[Zero-ruan]

Will errors caused by browser plug-ins trigger replay? However, the console did not report any errors in the replay.

[billyvg]

Can you confirm you are using self hosted and not SaaS?
Can you post your Sentry.init() configuration?

[Zero-ruan]

Thanks for your reply, here is my init

[billyvg]

You have replaysSessionSampleRate set to 0.1, which means it'll capture 10% of all sessions (which may or may not have an error). ALso I'm not sure how you are using onStartReplay and onStopReplay, which may also be capturing replays w/o errors.

[Zero-ruan]

I will stop Replay when the user switches roles, start Replay after the user switches successfully, and setUser again to start a new Replay to record errors. So as long as replaysSessionSampleRate is used, this without error Replay will definitely exist, right?

[Zero-ruan]

When I refresh, will run Sentry.init and replay.start again. It seems that this will generate a without error Replay. Is there a way to solve this? Because the error I want to record after the user switches roles is in a new Replay, Sentry.init and replay.start or stop are both performed during the role selection step after the user logs in.

[billyvg]

So as long as replaysSessionSampleRate is used, this without error Replay will definitely exist, right?

Correct

You can use replay.startBuffering() instead of replay.start() if you only want replays with an error. startBuffering() will wait for an error to occur, check if it should be sampled, and then create a replay if so.

[Zero-ruan]

Thank you for your reply. When I refresh and re-execute Sentry.init and replay.startBuffering, a replay without errors will still be generated again. The effect is no different from using replay.start. I wonder if I have overlooked something?

[Zero-ruan]

Sorry, this is a private project,I can't provide it, but I can provide my operating principle.
https://github.com/Zero-ruan/sentry/tree/main

[billyvg]

Sorry, this is a private project,I can't provide it, but I can provide my operating principle. https://github.com/Zero-ruan/sentry/tree/main

Just want to double check, is replaysSessionSampleRate set to 0? In this repo it is set to 1.

[Zero-ruan]

I use 1 in my local environment because I found that when using 0.1, replay is not generated for the same type of error report after switching roles, so I need to use 1 to test.

[billyvg]

Right but I thought you did not want replays without any errors? with replaysSessionSampleRate: 0, and using startBuffer() after role switch, you are still getting replays without errors?

[Zero-ruan]

thank you for your reply, I used replaysSessionSampleRate: 0 and startBuffer(). This solves the problem of generating a replay without errors. However, when I make an 'ApolloError: Response not successful: Received status code 400' error, this generates a replay, which is fine. But when the error occurs again after switching roles, the error after switching roles does not generate a replay.

[Zero-ruan]

I have tried different interfaces to generate this error, but no replay will be generated. However, if I make other types of errors such as (xxx is not a function), the replay will be generated again. It seems that it can only record the same type of error once. This is not good. I made an error when calling the interface at place A, and then I switched roles and made an error when calling another interface at place B. It can log the A error. But it does not record errors in the B interface

[Zero-ruan]

Incidentally, I had to use replay.setupOnce() to start replay again after using replay.stop(), using replay.startBuffering() or replay.start() alone didn't work, and I don't see a description of this in the documentation.

[bruno-garcia]

@Zero-ruan could you share a repro of the behavior your experiencing? We could go from there and either identify any bug or suggest a way to use the API to get the behavior you expect instead

[Zero-ruan]

Sure, I updated this link: https://github.com/Zero-ruan/sentry/tree/main

[bruno-garcia]

Thanks @Zero-ruan, we'll take a look soon

[billyvg]

I'm going to close this ticket as the original issue is now fixed. Calling start() after a replay was stopped works as expected currently.

However there is still a problem where after a session has been stopped, we can't "restart" the session to capture onError. This issue has been captured in #77458