Producer timing out

[g-freire]

Description

Hi, I'm having an issue on some of the microservices pods running the library. I'm receiving many timeouts, which I don't receive using other libs(eg. sarama). Does someone now if this is something expected, like a warning ? We are concerned, because we can't lose data and there is timeouts all over the logs. I've already increased the client timeout related configs, downgraded the version and test in different clusters. Any help on it will be very welcomed as we really like this lib and want to use in prod.

How to reproduce

If I publish every second I can't reproduce the error. If I start sleeping between the iteration, I always trigger the timeout when sleeping more than 5-7 min. I'm running it on a Azure cluster (AKS). When we test it locally pointing to the qa cluster on azure we receive the timeouts but we don't lose data. I don't know about the qa system now, because we have many clients and is harder to track the data flow. Same problem with transactional or non-transactional block

Checklist

Please provide the following information:

• confluent-kafka-go and librdkafka version (LibraryVersion()): 1.7 (latest)
• Apache Kafka broker version: 2.7
• Client configuration: ConfigMap{...}
• Operating system: golang image build executing on debian image
• Provide client logs (with "debug": ".." as necessary)
• Provide broker log excerpts
• Critical issue

Configs used on tests


No configs at all

[jliunyu]

Thanks @g-freire for reporting this. Looks this is related to confluentinc/librdkafka#2845 (comment).

But I saw there is a fix related to this confluentinc/librdkafka@0527457.

@edenhill , can you please take a look at this one?

[jliunyu]

Hi @g-freire, sorry for the late response. I conformed with @edenhill , the coming 1.8.0 release will add support for the client to kill idle connections before the LB or broker does, improve things on Azure.
confluentinc/librdkafka#3109