platinumhamburg commented on PR #3404:
URL: https://github.com/apache/fluss/pull/3404#issuecomment-4716240199

   > The RateLimiter in the job only throttles deletes. The job also issues 
many read-side FS calls — especially `listStatus` over log tablet dirs that can 
hold tens of thousands of segment dirs. Unthrottled, this can saturate the 
underlying store and impact cluster stability. Out of scope for this PR. Could 
you open a tracking issue for this and replace this TODO with a one-line 
pointer to it?
   
   @wuchong Thanks for catching this. I agree this should be fixed in this PR 
because unthrottled remote filesystem reads/lists can be a production stability 
risk on object stores with QPS limits.
   
   I updated the action to use a single shared remote filesystem operation 
limiter instead of limiting deletes only. The new option is 
--remote-fs-op-rate-limit-per-second, and the budget covers remote FS metadata 
reads, manifest reads, listStatus, and delete operations.
   
   For the distributed ScanAndClean stage, the configured rate is split across 
subtasks based on the runtime operator parallelism. This is a best-effort 
job-level target because Flink does not provide a cross-JVM global RateLimiter 
for this action. Scope enumeration uses the same option as well.


-- 
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