# Large fetch response size when resuming read from partition

**URL:** <https://forum.confluent.io/t/large-fetch-response-size-when-resuming-read-from-partition/37375>\
**Category:** Java Clients\
**Created:** [1 March 2025 14:54 UTC](https://forum.confluent.io/t/large-fetch-response-size-when-resuming-read-from-partition/37375 "2025-03-01T14:54:24Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![dizider](https://avatars.discourse-cdn.com/v4/letter/d/ecb155/32.png) [@dizider](https://forum.confluent.io/u/dizider)\
**Post date:** [1 March 2025 14:54 UTC](https://forum.confluent.io/t/large-fetch-response-size-when-resuming-read-from-partition/37375/1 "2025-03-01T14:54:24Z")

</div>

Hi,

I’ve Kafka consumer with maximum fetch size per partition (_max.partition.fetch.bytes_ ) set to 5MB. Fetches respect this until the first pause and resume consuming from partition. This cause the fetch request size to increase up to 30MB. I have no clue what is causing this behavior. Does anyone have any idea?

Here is a plot of [_fetch-size-avg_](https://kafka.apache.org/37/generated/consumer_metrics.html) metric for visualization of described behavior.

 ![IMG_2788](https://us1.discourse-cdn.com/flex019/uploads/confluentcommunity/original/2X/5/5a91485610efba8db3d7894f30d8d0341cf6a01d.jpeg)

---

<div class="post-metadata">

**Author:** ![mjsax](https://sea1.discourse-cdn.com/flex019/user_avatar/forum.confluent.io/mjsax/32/3113_2.png) [@mjsax](https://forum.confluent.io/u/mjsax)\
**Post date:** [1 March 2025 22:20 UTC](https://forum.confluent.io/t/large-fetch-response-size-when-resuming-read-from-partition/37375/2 "2025-03-01T22:20:53Z")

</div>

No idea. What version of the clients/broker are you using?

---

<div class="post-metadata">

**Author:** ![dizider](https://avatars.discourse-cdn.com/v4/letter/d/ecb155/32.png) [@dizider](https://forum.confluent.io/u/dizider)\
**Post date:** [2 March 2025 12:54 UTC](https://forum.confluent.io/t/large-fetch-response-size-when-resuming-read-from-partition/37375/3 "2025-03-02T12:54:16Z")

</div>

Thanks for the reply. I’m using broker and client version `3.8`. Maybe, it can be due to a misunderstanding in reporting, because if client has any completed fetch requests to load a partition that has been pause, responses are not discarded, but remain in fetch buffer for future poll execution. Let’s say, client has had 2 inflight requests to fetch partition 0. Then this partition was `paused`, so it leads in client having two completed fetch requests in buffer, which cannot be skipped until `resume` is called. And the metric is updated after the fetch request is consumed by `poll`.

Anyway, I suppose my theory is wrong, but I have no other reasonable explanation for this behavior. I’ll try to reproduce it and look at the network traffic and broker metrics. If you have any other idea, I would be glad to hear it.
