getsentry
Make sentry::init return Result<T> to prevent sentry panicking during outages
Hi,
I think sentry experienced an outage yesterday. We discovered this when one of our production services went down. After some debugging we found that a sentry service was returning a 502 which (I think) made sentry panic. See the logs snippet below where we get the html of a 502 error page from sentry
2024-10-16T19:10:45.950Z
[sentry] enabled sentry client for DSN
[sentry] dropping client guard -> disposing client
[sentry] client close; request transport to shut down
[sentry] Get response: `
<html><head>
<meta http-equiv="content-type" content="text/html;charset=utf-8">
<title>502 Server Error</title>
</head>
<body text=#000000 bgcolor=#ffffff>
<h1>Error: Server Error</h1>
<h2>The server encountered a temporary error and could not complete your request.<p>Please try again in 30 seconds.</h2>
<h2></h2>
</body></html>
`
I've spent some time digging around the source but haven't yet pinpointed where the http call is made. What I wanted to check is that if sentry::init does panic on getting a 502 (from somwhere) if you'd be open to changing the return type to a Result to make it easier to handle that failure mode?
If this is amenable I'd be more than happy to make a contribution but given this would be a breaking change I wanted to ask before writing any code.
Let me know what you think.
Thanks,
I spent a few hours digging into this further this afternoon. I don't think my initial diagnosis is correct. As far as I can tell the log message of the 502 (Get response) is fine with the status being a 502 and does not panic.
I don't understand why we were seeing the dropping client guard -> disposing client and client close; request transport to shut down messages however. I think this is the more likely root cause of the issue we experienced.
This probably means that the title and initial issue are no longer relevant but I want to leave this issue open because I'd appreciate some help in ensuring we don't experience this again
Hi @benjaminjellis @ethiril As shown in the log, it seems that the client guard was dropped (and therefore the client closed) when you didn't expect it to. However, this should not cause a panic. If the client is closed, the Sentry APIs should simply no-op. Therefore, I'm not sure that the panic you experienced is caused by Sentry. I'm more than happy to investigate further, but I would need you to provide some more context on what happened and how you're using the SDK (e.g. how you're initializing it and which integrations/APIs you're using)
Closing this issue as it's been inactive for 3 weeks and it's not actionable for us if we don't have more information. It doesn't seem to be a bug in the SDK. If you're affected by this, please open a new issue with more context.