default behaviour should be to throw away decision logs instead of running OOM if its unable to send the decision logs

[johanneslarsson]

After a discussion on Slack I realised that OPA would end up in a OOM, if its unable to push its decision logs to the control plane.
https://openpolicyagent.slack.com/archives/CBR63TK2A/p1651663584624969. I haven't verified this is the case yet.

What part of OPA would you like to see improved?

decision_logs.reporting.buffer_size_limit_bytes should probably have default value that suits the smallest containers. I understand that there are use cases where logs are required, but a crash wouldn't help there either.

Describe the ideal solution

I think most people want OPA to rather throw away logs than causing runtime exceptions.

[srenatus]

I haven't verified this is the case yet.

We should try to do that. 🔍 I'd imagine an infinite loop that queries, queries, queries, while having the decision logger configured to send to a port without a service, should trigger this...?

[anderseknert]

Hey Johannes 👋😃

Optimizing the value for minimal resource allocation sounds a bit risky to me — I would not want OPA to discard decisons if it's using like 10% of the memory I've allocated for the process. In an ideal scenario you'd be able to do something like:

decision_logs.reporting.buffer_size_limit_available_mem: 80%

But I'm guessing there's no reliable way of doing that.

An alternative could be to dump logs to disk if possible: #3333

[stale]

This issue has been automatically marked as inactive because it has not had any activity in the last 30 days.

[ashutosh-narkar]

Closing this in favor of #4659

[johanneslarsson]

@ashutosh-narkar the ticket you mentioned as favored is this one?

[ashutosh-narkar]

Sorry @johanneslarsson. I meant this one. Thanks for catching this.