[
https://issues.apache.org/jira/browse/KAFKA-10888?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17286089#comment-17286089
]
Jason Gustafson commented on KAFKA-10888:
-----------------------------------------
[~junrao] That's a good observation. The partitioner knows the approximate size
of each batch that gets sent, so it could take that into account before
rotating partitions. We were thinking of approaches which involved the
partitioner being aware of how many bytes were inflight, but propagating this
information from the accumulator to the partitioner is kind of messy.
Perhaps it is simpler for the partitioner to track at the level of total bytes
sent to each partition. Maybe we could see this analogously to the tcp window
size. For example, say you start with an expected size of 10 records (or 1k
bytes or whatever). The partitioner keeps writing to the partition until this
many records/bytes have been sent regardless of the batching. If the batch is
not filled when the limit is reached, then we move onto the next partition, but
we increase the batch size. On the other hand, if a batch gets sent before the
limit is reached, then we decrease the size. In this way, we can keep a better
balance between partitions.
> Sticky partition leads to uneven product msg, resulting in abnormal delays
> in some partations
> ----------------------------------------------------------------------------------------------
>
> Key: KAFKA-10888
> URL: https://issues.apache.org/jira/browse/KAFKA-10888
> Project: Kafka
> Issue Type: Bug
> Components: clients, producer
> Affects Versions: 2.4.1
> Reporter: jr
> Priority: Major
> Attachments: image-2020-12-24-21-05-02-800.png,
> image-2020-12-24-21-09-47-692.png, image-2020-12-24-21-10-24-407.png
>
>
> 110 producers ,550 partitions ,550 consumers , 5 nodes Kafka cluster
> The producer uses the nullkey+stick partitioner, the total production rate
> is about 100w tps
> Observed partition delay is abnormal and message distribution is uneven,
> which leads to the maximum production and consumption delay of the partition
> with more messages
> abnormal.
> I cannot find reason that stick will make the message distribution uneven
> at this production rate.
> I can't switch to the round-robin partitioner, which will increase the
> delay and cpu cost. Is thathe stick partationer design cause uneven message
> distribution, or this is abnormal. How to solve it?
> !image-2020-12-24-21-09-47-692.png!
> As shown in the picture, the uneven distribution is concentrated on some
> partitions and some brokers, there seems to be some rules.
> This problem does not only occur in one cluster, but in many high tps
> clusters,
> The problem is more obvious on the test cluster we built.
> !image-2020-12-24-21-10-24-407.png!
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
