Yunze Xu

It might be a bug at Pulsar side because the core issue is the reader on `__consumer_offsets-partition-6` failed to read a message and retried for multiple times. > 2022-07-04 17:17:33.0697...

It's caused by the disconnection from Kafka client because Kafka client didn't receive the response after `request.timeout.ms`.

@timmyyuan Could you help with this issue?

For Java client, following the [instructions here](https://github.com/streamnative/kop/blob/master/docs/security.md#oauthbearer). For other languages, I think we need to determine which client we should support. Since it's a **plugin for Kafka client**, we need...

I see. So the targets are more clear now. For Java client: - Complete these TODOs, like the expiration support - Remove the Pulsar dependency to make it lightweight, it...

Did you download the binary from [release page](https://github.com/streamnative/kop/releases) directly? I think you can decompress the NAR file and see if the class exists.

Is this case when partitions were deleted while the metadata still existed? Maybe we can also configure `brokerDeleteInactivePartitionedTopicMetadataEnabled=true` to solve it. I've also encountered the case that only part of...

I've tried it a long time ago. It's nearly impossible because Kafka uses a global unique schema id (integer) while Pulsar doesn't support it.

You can take a look at `KafkaAvroDeserializer#deserialize`, there's an important step that a REST request is sent for the schema string. See `RestService#getId` in Confluent schema registry project: ```java public...

Oh, you're right. In this case, Kafka client should configure our provided serializer/deserializer. The background of my previous task for schema support is to reuse Confluent's provided serializer/deserializer, so it's...