Hi Team,
I'm utilizing Kafka Streams to handle data from Kafka topics, running
multiple instances with the same application ID. This enables distributed
processing of Kafka data across these instances. Furthermore, I've
implemented state stores with time windows and session windows. To retrieve
windows efficiently, I've established a remote query mechanism between
Kafka Streams instances. By leveraging the queryMetadataForKey method on
streams, I can retrieve the hostname where a specific key was processed and
where the corresponding window data resides in the state store.
*streams.queryMetadataForKey(storeName, recordKey, new
DataKeySerilizer()).activeHost();*
This functionality has been running smoothly for quite some time, up until
we upgraded our Kafka and Kafka Streams versions to 3.7.0. Since the
upgrade, we've noticed some unexpected behavior that we didn't encounter
with the previous versions.
- The queryMetadataForKey method is returning "unavailable" as the
hostname, despite having two Kafka Streams instances in a running state.
The issue seems to persist intermittently, and restarting the Kafka Streams
instances temporarily resolves it. However, the problem resurfaces after
some time.
- Additionally, we've observed that the state store of Kafka Streams
instances is not evenly distributed as it was before the upgrade.
Previously, if a Kafka topic had 10 partitions and two Kafka Streams
instances were running, these 10 partitions would be evenly shared between
the two instances. However, this behavior seems to have changed after the
upgrade.
When can this issue happen?
How can I fix the issue ?
I would like to express my gratitude in advance for any assistance provided.
--
Thank's&Regard's,
Prasad,
91-9030546248.
