panics in k6 extension only skip an iteration instead of failing the test or iteration

[geekflyer]

Brief summary

Currently when a k6 extension panics, the k6 runtime "catches" the panic, simply skips the iteration and logs an error message like:

"time="2022-01-26T14:22:12-05:00" level=error msg="a panic occurred in VU code but was caught: record on line 1373: wrong number of fields" executor=constant-vus scenario=default"

There is no reasonable way to programmatically catch or report extension panics in k6 and apart from the log message they are essentially like a "silent" error.
For example if a load test has 10 iterations and all of them panic, k6 still considers the load test as a whole as successful.

k6 version

v0.35.0

OS

cross platform

Docker version and image (if applicable)

No response

Steps to reproduce the problem

create an extension method that instantly panics and call this method from a JS script

Expected behaviour

any (combination?) of these:

• A panic bubbles up to the k6 JS layer as JS error and can be handled or ignored there
• A panic stops the iteration but makes the iteration count as an "error" somewhere
• A panic fails the k6 process with a non-zero exit code immediately

Actual behaviour

the k6 runtime "catches" the panic, simply skips the iteration and logs an error message like "time="2022-01-26T14:22:12-05:00" level=error msg="a panic occurred in VU code but was caught: record on line 1373: wrong number of fields" executor=constant-vus scenario=default". Otherwise it continues "as if" the iteration or loadtest was successful.

[na--]

Very good points... 🤔

A panic bubbles up to the k6 JS layer as JS error and can be handled or ignored there

I don't think this is possible, since a panic in VU code will generally interrupt the whole JS VM execution 😕 And even if it's technically possible, if there's a chance it leads to undefined behavior, we should probably avoid this approach unless we are absolutely sure it can work reliably.

• A panic stops the iteration but makes the iteration count as an "error" somewhere
• A panic fails the k6 process with a non-zero exit code immediately

These are somewhat connected to #877 and #1769. If Go panics emit an interrupted_iteration{cause:panic} metric, we can easily add a threshold that aborts the script based on that sub-metric 🤔

[olegbespalov]

Hey @geekflyer !

We adjusted the behavior of how k6 treats panic. It should abort the execution. Could you please validate if that resolves your needs?

Cheers