Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
No data about Coverage
0.0% Duplication on New Code

See analysis details on SonarCloud

ping @dstepanov

io.micronaut.configuration.kafka.annotation.ErrorStrategyValue.NONE is still in the code. It was added as a deprecated enum on #441 as part of a fix for #372.

I didn't remove this yet because it's still used for code in the module and it's not clear to me how to deal with it. Please advise, or contribute as appropriate. This PR is meant to remove all deprecated code for Micronaut Framework 5.

ping @dstepanov

io.micronaut.configuration.kafka.annotation.ErrorStrategyValue.NONE is still in the code. It was added as a deprecated enum on #441 as part of a fix for #372.

I didn't remove this yet because it's still used for code in the module and it's not clear to me how to deal with it. Please advise, or contribute as appropriate. This PR is meant to remove all deprecated code for Micronaut Framework 5.

Apologies for the late reply on this - for some context, I was one of the original contributors of the error strategies. We built this into micronaut as we were fixing #372 however there is still some lacking behaviour around handling batches (the current issue exists for batches today I believe).

One of the things that we discussed was not breaking existing code; take this below for example:

    @Override
    public void handle(final KafkaListenerException exception) {
        exception.getConsumerRecord().ifPresent(consumerRecord -> {
            TopicPartition topicPartition = new TopicPartition(consumerRecord.topic(), consumerRecord.partition());
            exception.getKafkaConsumer().seek(topicPartition, consumerRecord.offset());
        });
    }

This would be required if the error strategy is set to NONE to ensure that #372 isn't an issue for losing records in a poll. By setting RETRY_ON_ERROR this code that I showed is fundamentally defunct but it would change the behaviour of the kafka listener, as such, it was deprecated and the default behaviour should really be not to lose data (however this is probably not a decision I'm empowered to make on this project) since the other strategies will either skip over the problematic message. There is currently no dead letter queue error strategy unfortunately.

Also at some point I do hope to contribute to error strategies for batching, but I believe there's quite a few things that aren't working quite right for batch at the moment, e.g. redelivery not redelivering messages when batch=true and redelivery=true. I should probably raise an issue for this too but overall, this part in the documentation really does need to be recognised:

I'll ping on discord and see whether I can support this over my break from work during the festive season.

I hope this helps with the removal of deprecated code!