I keep getting MessageSizeTooLargeError, error message gives size much bigger than actual message were given to producer. about aiokafka

[vmaurin]

How do you serialize with avro ? Avro is a format where a schema is required for the producer to serialize and the consumer to deserialize, otherwise is just gibberish bytes. Usually the strategy with kafka is to store schema in some sort of central registery, then put in kafka header the reference to the schema used to produce the message. It is what confluent is doing with their schema registry.

As far as I know aiokafka doesn't provide anything avro related, so the serializer/deserializer must be from your own implementation. Depending on what you are doing, it might be that your serialized message is containing both the schema and the data

[ant0nk]

@vmaurin I'm using kafkit library for serialization and communication with schema registry.

[vmaurin]

Maybe try to dump the message you serialized before passing it to aiokafka ? Otherwise, as far as I can see, the size is checked per message https://github.com/aio-libs/aiokafka/blob/master/aiokafka/producer/producer.py#L411 (even if then messages might be batched)
The formula seems to be : overhead + len(key) + len(value)
Headers seems ignored

[ant0nk]

@vmaurin Can I specify max_request_size for producer bigger than broker's relevant value if I have compression enabled?

[vmaurin]

You mean max.message.bytes on broker/topic ? It might be then it seems to be applied after compression, but then it is also applied to a batch of message, while the check in aiokafka is just for a single message

[ant0nk]

@vmaurin yes, but I use send_and_wait() to send immediately, so I hope batch will not exceed max.message.bytes too.