EchoMany response lost about rpc-frmwrk

[zhiming99]

Probably we fixed this issue in commit on 4.12. We haven't seen this issue occur for 8days. the commit removed all the references to fuse_lowlevel_notify_delete on the execution path of unlink. I close it now and if it occurs again, I will reopen this issue.

[zhiming99]

reopened due to this run

[zhiming99]

another try to fix this issue, by commit on 5.31
Actually the response of EchoMany did not lose, and instead, github does not log
the hanging program, which is seemingly like the program hangs at the first
request EchoMany waiting for its response. But actually the last request
EchoStream request was also completed. And most likely this time, the program
hangs endlessly at the call to 'select' on the stmfile because of a race
condition between fs_poll and OnStmRecv. That is, There is a slim chance when
the fs_poll checked to found no data, and then preempted right before setting
a valid notify handle, the OnStmRecv arrived and found there is no notify handle,
and left. and when the fs_poll got back the control and set the pollhandle, the
pollhandle will never be notified, because the OnStmRecv has gone. Now I have
added to stmfile a FillAndNotify to address this issue.

[zhiming99]

reopened by this run. This time the hanging happens in mkDirTest ROS, not the previous stressTest normal. probably, the last fix hit something, but not a knockout.

[zhiming99]

close this issue by this commit on Jun. 11, 2023. Just a few hours after reopened.