Yes, I ended up having to get the
FD and doing a select on it manually. But to me this makes the
blocking read feature of cpg_dispatch useless.
ipc_setup.c:163
qb_ipc_us_ready() returns -EAGAIN on EINTRBut yes, it does check errno and restart on EINTR: cpg.c:372 cpg_dispatch() calls qb_ipcc_event_recv() ipcc.c:358 qb_ipcc_event_recv() calls _check_connection_state_with() ipcc.c:112 _check_connection_state_with() calls qb_ipc_us_ready() ipc_setup.c:161 qb_ipc_us_ready() calls poll() ipcc.c:122 _check_connection_state_with() returns the response (-EAGAIN) ipcc.c:361 qb_ipcc_event_recv() returns the response(-EAGAIN) cpg.c:377
cpg_dispatch() turns the response (-EAGAIN) into CS_ERR_TRY_AGAIN
cpg.c:393 cpg_dispatch() restarts the loop and we go back to cpg.c:372 -Patrick From: Jan Friesse <jfriesse@xxxxxxxxxx>
Sent: 2013-10-23 11:14:28 E
To: Patrick Hemmer <corosync@xxxxxxxxxxxxxxx>,
discuss@xxxxxxxxxxxx
Subject: Re: signaling in cpg_dispatch
Patrick, I'm unsure if I really understand your question. cpg_dispatch doesn't check errno and doesn't do restart. I believe what you want to achieve is alarm handler. Way how to do it, is to use select/poll on cpg/(any other service) fd and call cpg_dispatch only when fd is ready for read. Alarm handler can then set variable which is checked after cpg_dispatch call. If you have only cpg_dispatch, you even don't need poll, but can directly run endless loop. so pseudocode fragment: int want_exit_loop = 0; sigalarm_handler() { want_exit_loop = 1; } ... while (!want_exit_loop) { cpg_dispatch(); } ... If I didn't understood your question correctly, please try reformulate it or even better, send code fragment. Regards, Honza Patrick Hemmer napsal(a):So I'm trying to write an application which uses CPG but am having a problem with cleanly terminating the application. The issue is that when calling cpg_dispatch, if the process gets a signal while it's sitting on the `poll()` call, and there is a signal handler defined, upon return from the signal handler, errno is set to EINTR which causes cpg_dispatch to restart the call to qb_ipcc_event_recv. Basically nothing other than a SIGKILL (or a signal which has no handler and results in the default action of exit) can terminate the app. And aside from terminating, if you wanted to use a SIGALRM to break the poll() and go do other things, you can't. I'm using corosync 2.3.2 and libqb 0.14.4 This assessment correct or am I missing something? -Patrick _______________________________________________ discuss mailing list discuss@xxxxxxxxxxxx http://lists.corosync.org/mailman/listinfo/discuss |
_______________________________________________ discuss mailing list discuss@xxxxxxxxxxxx http://lists.corosync.org/mailman/listinfo/discuss