github-actions[bot] commented on code in PR #67559:
URL: https://github.com/apache/doris/pull/67559#discussion_r3946301735


##########
be/src/common/config.cpp:
##########
@@ -714,7 +714,7 @@ DEFINE_mBool(enable_stream_load_commit_txn_on_be, "false");
 DEFINE_Int64(stream_tvf_buffer_size, "1048576"); // 1MB
 
 // request cdc client timeout
-DEFINE_mInt32(request_cdc_client_timeout_ms, "60000");
+DEFINE_mInt32(request_cdc_client_timeout_ms, "600000");

Review Comment:
   [P1] Keep the CDC HTTP budget inside the outer RPC deadline
   
   This is a per-attempt timeout shared by every CDC endpoint, not a 600-second 
end-to-end budget. `send_request_to_cdc_client` runs it inside 
`execute_with_retry(3, 1, ...)`, while FE abandons light calls such as 
`/api/fetchEndOffset` and `/api/close` after 90 seconds and heavy calls after 
600 seconds. The BE closure does not observe the RPC controller/cancellation 
and curl has no abort callback, so a server that accepts a POST but stops 
responding can retain a shared `brpc_heavy` worker for about 1,803 seconds 
after the FE has moved on; even a heavy call cannot report its first 600-second 
timeout before its equal outer deadline expires. With repeated job retries this 
can starve the pool used by unrelated load/network work. Please use 
route-specific light/heavy budgets whose complete retry schedule fits below the 
FE deadline, or propagate the remaining RPC deadline/cancellation into curl and 
stop retries once it expires.



-- 
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]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to