Fix/check context cancel logging

[abdullahalrifat]

This PR adds a unit test to verify that Check correctly handles explicit context cancellation before query execution.

• Introduced a testServer wrapper with a beforeCheck hook to simulate cancellation during request handling.

• Ensured that the expected error (context canceled) is returned and properly logged.

• Validated that the error metric CheckRequestCancelledErr is incremented.

This strengthens test coverage around graceful handling of cancelled requests and ensures consistent logging/metrics behavior.

[tjons]

Hi @abdullahalrifat - thanks for the contribution. I left a review with some suggested changes, but I'd be happy to help you get this in. That said, I'm curious how you found this behavior? Did it arise in a production deployment? Some background about how this context might be cancelled would be helpful - are you seeing context cancellations from Envoy?

Once we better understand that and you are able to make the suggested changes, I will need you to squash your commits and rebase on top of the latest main as well.

Thanks for contributing!

[abdullahalrifat]

Hi @abdullahalrifat - thanks for the contribution. I left a review with some suggested changes, but I'd be happy to help you get this in. That said, I'm curious how you found this behavior? Did it arise in a production deployment? Some background about how this context might be cancelled would be helpful - are you seeing context cancellations from Envoy?

Once we better understand that and you are able to make the suggested changes, I will need you to squash your commits and rebase on top of the latest main as well.

Thanks for contributing!

When I deployed OPA with Envoy in a Kubernetes cluster, I noticed that some incoming requests occasionally failed. These failures could happen for various reasons, such as networking issues. On the client side, this wasn’t a problem since they had a retry mechanism in place, but OPA was still logging these as failures.

When analyzing OPA failure logs, we generally classify them into categories like:

• Rule conflicts (caused by bad policies)

• Query timeouts (when evaluation takes longer than expected)

However, the context_cancelled errors are different — they aren’t actual OPA failures, since OPA itself isn’t at fault. I wanted to separate these out to ensure that OPA failure logging accurately reflects real issues.

[tjons]

This LGTM, @abdullahalrifat - can you please rebase on the latest main and squash your commits?