Consumer.poll goes into infinite loop

[r-priyam]

Description

Consumer.poll client is going into infinite loop when the Kafka is running on Kubernetes. When the cluster is scheduled on the new node is just stuck on the below error:

As you can see in the screenshot the service was just stuck and then it recovered after like 12 hours.

How to reproduce

import socket
from confluent_kafka import Consumer, Message

consumer = Consumer(
	{
            "bootstrap.servers": "",
            "client.id": socket.gethostname(),
            "auto.offset.reset": "earliest",
            "enable.auto.commit": False,
            "group.id": "",
        }
)
consumer.subscribe(topics=[...])
msg: Message = self._consumer.poll(timeout="10') # Goes into infinite loop here and gets stuck

Run the above code and then restart the kafka cluster or change the node in k8s

Checklist

Please provide the following information:

• confluent-kafka-python and librdkafka version (confluent_kafka.version() and confluent_kafka.libversion()): 2.1.1
• Apache Kafka broker version:
• Client configuration: { "bootstrap.servers": "", "client.id": socket.gethostname(), "auto.offset.reset": "earliest", "enable.auto.commit": False, "group.id": "", }
• Operating system: Linux
• Provide client logs (with 'debug': '..' as necessary)
• Provide broker log excerpts
• Critical issue

[pranavrth]

It feels that the host is not reachable or up.

[r-priyam]

@pranavrth the host gets up back in like max 2 mins, it's just k8s scheduling it on the another node which changes the IP, the service container works fine when it's restarted.

[r-priyam]

@pranavrth, any luck, please? Appears that the package is handling this issue at the very low level and not raising an exception?

[r-priyam]

@pranavrth, upon further digging in the source code, we found that we can use "error_cb" in the config to have the callback whenever the error is raised. It would be good if the client has raised this directly instead of having a callback.

[pranavrth]

Can you enable debug logging by using 'debug': 'all' in the config and send us the logs to investigate?