rpc: Improve rpc clnt connection cleanup process #4329
Conversation
|
/run regression |
1 similar comment
|
/run regression |
|
/run regression |
|
0 test(s) failed 1 test(s) generated core 1 test(s) needed retry 1 flaky test(s) marked as success even though they failed |
During the first rpc clnt submission we take the rpc reference and register the call_bail function for the timer thread. The timer thread call call_bail function every 10s basis. In case if a client trigger a shutdown request it try to call rpc_clnt_connection_cleanup to cleanup the rpc connection.The rpc_clnt_connection would not be able to cleanup the rpc connection successfully due to the cleanup_started flag being set by the upper xlator. The rpc reference will be unref only after trigger a call_bail function so basically if somehow call_bail is triggered just before start a shutdown process the application has to wait for 10s to cleanup the rpc connection eventually the process becomes slow. Solution: Unref the rpc object based on the conn->timer/conn->reconnect pointer value as we are doing the same for ping_timer. These pointer are always modified under the critical section so we can assume if pointer is valid it means rpc reference is also valid. Fixes: gluster#4320 credits: Xavi Hernandez <xhernandez@redhat.com> Change-Id: Ib947b8bfcbe1b49e1ed05a50a84de6f92afbca13 Signed-off-by: Mohit Agrawal <moagrawa@redhat.com>
|
/run regression |
|
1 test(s) failed 0 test(s) generated core 3 test(s) needed retry |
|
/run regression |
|
1 test(s) failed 0 test(s) generated core 1 test(s) needed retry |
During the first rpc clnt submission we take the rpc reference and register the call_bail function for the timer thread. The timer thread call call_bail function every 10s basis. In case if a client trigger a shutdown request it try to call rpc_clnt_connection_cleanup to cleanup the rpc connection.The rpc_clnt_connection would not be able to cleanup the rpc connection successfully due to the cleanup_started flag being set by the upper xlator. The rpc reference will be unref only after trigger a call_bail function so basically if somehow call_bail is triggered just before start a shutdown process the application has to wait for 10s to cleanup the rpc connection eventually the process becomes slow.
Solution: Unref the rpc object based on the conn->timer/conn->reconnect pointer value as we are doing the same for ping_timer. These pointer are always modified under the critical section so we can assume if pointer is valid it means rpc reference is also valid.
Fixes: #4320
credits: Xavi Hernandez xhernandez@redhat.com
Change-Id: Ib947b8bfcbe1b49e1ed05a50a84de6f92afbca13