Support partial batch failure

[t-richard]

This allows the Symfony Messenger integration to benefit from "partial batch failures" with SQS.

This is done via a new argument in the service declaration (and proper SQS config ofc).

Bref\Symfony\Messenger\Service\Sqs\SqsConsumer:
        public: true
        autowire: true
        arguments:
            # Pass the transport name used in config/packages/messenger.yaml
            $transportName: 'async'
+           $partialBatchFailure: true

If you want to try this, in `composer.json`

```json { "require": { ... "bref/symfony-messenger": "dev-partial" }, "repositories": [ { "type": "vcs", "url": "https://github.com/t-richard/symfony-messenger" } ] } ```

Closes #57

[tyx]

Thanks @t-richard it looks perfect IMO !

[t-richard]

@tyx any chance you got to test this?

I'd love to get this merged and feedback would be welcomed to confirm it's working as intended :slightly_smiling_face:

[tyx]

Hi ! I don't have so much time for now. But I will test it ASAP for sure.

[t-richard]

I think this can be merged.

I'll try to release a 0.5 next week with this.

Any last thoughts?

[t-richard]

One case that is not covered here is FIFO queues.

AWS docs says about partial batch failure:

If you're using this feature with a FIFO queue, your function should stop processing messages after the first failure and return all failed and unprocessed messages in batchItemFailures. This helps preserve the ordering of messages in your queue.

Maybe we should handle this here too?

[shouze]

One case that is not covered here is FIFO queues.

AWS docs says about partial batch failure:

If you're using this feature with a FIFO queue, your function should stop processing messages after the first failure and return all failed and unprocessed messages in batchItemFailures. This helps preserve the ordering of messages in your queue.

Maybe we should handle this here too?

@t-richard Could be a good idea yes, to ensure that good pratice is applied by design. A good way is to throw a fatal exception to me, WDYT?

[t-richard]

@shouze I pushed a new commit with support to handle FIFO queues properly.

Throwing a fatal exception would fail the whole batch and put us in the current behaviour (ie. no partial batch failure).

What I'm doing here is running markAsFailed on all following messages in the batch if there is a failure.

Let me know what you think.

[shouze]

Indeed @t-richard looks like the right behavior!

[mnapoli]

Thanks! We should be good to merge then, could you rebase (or merge)? There are conflicts preventing to merge.

[t-richard]

Yes, I'll rebase, merge and release.

Do you think we should release this as part of 0.4.5 or 0.5.0 ?

[mnapoli]

Update following discussion: let's go with 0.5.0.