Etienne Chauchot created FLINK-27866:
----------------------------------------
Summary: CassandraOutputFormat and CassandraSink will timeout
twice if the number of maximum requests is reached
Key: FLINK-27866
URL: https://issues.apache.org/jira/browse/FLINK-27866
Project: Flink
Issue Type: Improvement
Components: Connectors / Cassandra
Reporter: Etienne Chauchot
Both Sink and Outputformat have the same logic: they support
_maxConcurrentRequests_ with a semaphore by counting the number of in-flight
write requests and acquire/release semaphore permits with each new request
begin/end. At close time, during the flush they wait for all the requests to be
processed by acquiring all the permits set in the semaphore.
If a given write fails because there are still _maxConcurrentRequests_ in
flight after the given wait period, a _TimeoutException_ will be thrown (while
trying to acquire a permit from the semaphore) . Then Flink will catch this
exception and close the sink/format. During the flush it will try to acquire
permits from the semaphore once again and this will lead to another wait period
and another {_}TimeoutException{_}.
Ideally, we should wait only once and get only one TimeoutException.
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
