Fix log writing failed. can't be called from trap context (no thread version) #1502
Conversation
Must restart thread since fork does not copy a thread Must reopen pipe, or parent process would receive a signal
| @@ -228,7 +228,7 @@ def work(interval = 5.0, &block) | |||
| break if interval.zero? | |||
| log_with_severity :debug, "Sleeping for #{interval} seconds" | |||
| procline paused? ? "Paused" : "Waiting for #{queues.join(',')}" | |||
| sleep interval | |||
| @signal_handler.handle_signal(interval) | |||
There was a problem hiding this comment.
As mentioned in the other PR, this will only run if the queue was empty. Should we move this out of the unless?
There was a problem hiding this comment.
Hmm, if we move this out of the unless, waiting interval is issued even if we run a job (behavior changes).
I am not getting good idea yet. I will think of how to resolve this. Any idea is welcome.
There was a problem hiding this comment.
can we do a non-blocking select so it doesn't sleep and just returns if there was no signal yet? and keep the sleep here where it was
There was a problem hiding this comment.
applied a change like 4ac5b36.
can we do a non-blocking select so it doesn't sleep and just returns if there was no signal yet? and keep the sleep here where it was
During sleeping, we can not get signal, then. I felt @signal_handler.handle_signal(interval) here is still better, but I added @signal_handler.handle_signal(0) for the case work_one_job returns true.
Also, I added handle_signal in perform_with_fork for during waiting a forked child finishes.
Please note that we still have problems that we can not handle signal if fork_per_job? is false. Hmm.
1 similar comment
sonots
force-pushed
the
fix_signal_trap2
branch
from
6f28013
to
4ac5b36
Compare
1 similar comment
| unless work_one_job(&block) | ||
| if work_one_job(&block) | ||
| @signal_handler.handle_signal(0) | ||
| else | ||
| break if interval.zero? |
There was a problem hiding this comment.
this line here might mess up your logic.. if the interval is zero and the queue is empty, then we wont handle signals...
|
We still have problems that we can not handle signal during processing a job when fork_per_job? returns false. Thread vesion does not have such an issue. |
|
As a final thought, there is no way to resolve the above issue with this no thread version. Let me close. |
Fix #1493, similarly with #1499, but this is no thread version.