nzw921rx commented on PR #10954:
URL: https://github.com/apache/seatunnel/pull/10954#issuecomment-4543860130

   > Thanks for the contribution. I reviewed the full diff on the current head 
and traced where this option is consumed on the normal Kafka source path.
   > 
   > # What this PR fixes
   > * User pain: very large reader-side poll buffers can amplify memory 
pressure on the Kafka source path.
   > * Fix approach: reduce the default `reader_cache_queue_size` from 1024 to 
2 and clarify the option description.
   > * One-line summary: the memory concern is real, but the current patch 
changes a stable default contract on the main path without handling 
compatibility expectations.
   > 
   > # Runtime chain I checked
   > ```
   > Kafka source config
   >   -> KafkaSourceOptions.READER_CACHE_QUEUE_SIZE 
[KafkaSourceOptions.java:44-51]
   >   -> reader / fetcher queue capacity
   >   -> affects the normal poll-buffering behavior for existing jobs that do 
not override the option
   > ```
   > 
   > # Findings
   > Issue 1: this changes a stable default contract for existing jobs
   > 
   > * Location: `KafkaSourceOptions.java:44-51`
   > * Problem: the default changes from `1024` to `2` on the normal user path 
when the option is omitted.
   > * Risk: this can materially change buffering, throughput, backpressure, 
and poll behavior for already deployed jobs even if users never touched the 
config.
   > * Better fix: either keep the old default and introduce a new safer opt-in 
path, or document this as an incompatible behavior change with explicit 
migration guidance.
   > * Severity: High
   > 
   > # Merge conclusion
   > ### Conclusion: can merge after fixes
   > 1. Blocking items
   > 
   > * Issue 1: the default-value compatibility story needs to be handled 
explicitly before merge.
   > 
   > 2. Suggested non-blocking follow-up
   > 
   > * If the team wants a safer out-of-box profile later, a connector-level 
recommendation in docs would be a better first step than silently changing the 
existing default.
   > 
   > Overall, I agree with the memory-pressure motivation, but the current 
patch is not ready to merge as a silent default flip.
   
   No need to add new parameters, the default behavior of this parameter will 
make users perceive inexplicable OOM, which is safer than throughput runtime. 
Users can explicitly adjust this parameter


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to