Hey Sushant,
Thanks for the KIP. Overall, the KIP looks like a good enhancement to the
Share Consumer interface and helps unlock a significant use case. A couple
of questions or suggestions:

SJ1: If a record is a "poison pill" (a record that consistently causes a
consumer to require a very long processing timeout), a malicious or buggy
consumer could be programmed to continuously renew the lock to prevent it
from ever reaching a failure state on the consumer and being moved to a
REJECTED or ARCHIVED state in the broker. Unlimited renewals can create a
self-inflicted, persistent Head-of-Line (HOL) blocking issue for the entire
share consumer group. Are there broker-side limits (e.g., a maximum number
of renewals or a maximum cumulative lock duration) that can be configured
to force a difficult record to eventually transition?

SJ2: This is somehow linked to the comment `SJ1`. Would it make sense to
improve visibility into lock renewal activity for better debugging and
operational awareness? For example, exposing broker-level metrics or
diagnostics that help operators identify cases of frequent or ineffective
renewals could be valuable in understanding consumer or record-level
issues. These metrics may be essential for operators to debug long-running
jobs, identify bottlenecks, and properly configure consumer timeouts.

Regards,
Sanskar

On Tue, Oct 7, 2025 at 3:46 PM Sushant Mahajan <[email protected]> wrote:

> I’d like to start the discussion for KIP-1222: Acquisition lock timeout
> renewal in share consumer explicit mode.
>
> JIRA: https://issues.apache.org/jira/browse/KAFKA-19742
> KIP Wiki:
>
> https://cwiki.apache.org/confluence/display/KAFKA/KIP-1222%3A+Acquisition+lock+timeout+renewal+in+share+consumer+explicit+mode
>
> Regards,
> Sushant Mahajan
>

Reply via email to