Failed CRC check should not panic

[christian-ehmig]

Describe the bug

If a CRC check fails

rabbitmq-stream-go-client/pkg/stream/server_frame.go

Line 359 in ab4d470

panic("Error during CRC")

panics. Shouldn't this be handeled in a better way?

Any user of your stream client would need to recover() from such a situation.

We are using the "reliable client" and provoke some crash scenarios like taking down one rmq cluster node. While the reconnect seems to work, an enabled CRC check panics in such situations and therefore stops all consumers.

Reproduction steps

1. Provoke a CRC checksum error for consumers

Expected behavior

Log the error - reconnect / reset message buffer etc.

Additional context

No response

[Gsantomaggio]

Hi @christian-ehmig
Thank you for reporting the issue.

CRC failure should be a race condition.

• Do you think you could reproduce the issue?
• Can you add more information about your environment?

[Gsantomaggio]

@hiimjako, Any thoughts on that?

[hiimjako]

I agree that the client shouldn't panic.
I'm not sure about the root cause of the CRC issue, whether it's an actual message problem or some sort of race condition. However, I think we could remove the panic, log the problem and reset the status as suggested.

[christian-ehmig]

Thanks for your quick response on this. I can try to reproduce this locally with a docker-compose setup, producers and consumers. However, in our test case, we use a single consumer on a stream queue via a single connection. I doubt it's related to a race condition.

The CRC check fails if we "unplug the cable" (kill the virtual machine) of one RMQ cluster node (3 in total).

[Gsantomaggio]

What I had in mind for the .NET client was a policy in case CRC fails.
Like:

• Ignore
• Close the consumer
• Close the connection ( with all the consumers )

I need to think a bit about that but I agree that panic is too much

[Gsantomaggio]

@christian-ehmig FYI: at the moment, we are focusing on #367
As soon as possible, we will be back. Feel free to propose a PR if you have a fix in mind. thank you

[Gsantomaggio]

We can close the TCP client when the CRC is corrupted, hoping that is a temporary problem.
That is better than panic.
We will do it for the 1.5