Attempt to fix two corner cases which result in deadlocks.#11
Attempt to fix two corner cases which result in deadlocks.#11Earnestly wants to merge 1 commit intoaktau:masterfrom Earnestly:master
Conversation
This patch attempts to fix two corner cases which can result in hhpc becoming deadlocked, <#9>. The first checks for the case where nanosleep() is interrupted which would set working to 0 and as hhpc never recovers from this case, the program will lockup indefinitely. Instead we resume where we left of if encountering EINTR. The second fix is a bit of a hack; on rare occasions the select() call in waitForMotion can seemingly block and practically never recover (I didn't wait for more than one hour). Instead I added a timeout to the select call, based on the gIdleTimeout value if it is greater than 1, otherwise it uses a 3 second delay. (This implies you can set a timeout of 2 seconds and have a 2 second timeout.) This doesn't really solve the issue of why select is getting stuck but does allow hhpc to recover and prevent it from holding the mouse hostage.
|
Blegh, nevermind. Now there's a huge spike of cpu usage about 1 minute in and The high cpu usage impacts the X server especially so it's likely related to Hopefully you will have better insight about this situation than I. |
|
Hi. I don't know why I never got any notification about this (I checked my mailbox, I assume one of my old filters removed it). I'm sorry about that. I saw this PR because I got a notification about another PR that I just merged. I haven't been running hhpc myself for a while (currently on a Chromebook SSH'ing into the boxen I manage). But I'd be happy to review this if you're still interested in this. I know it's been a very long time. Let me know, and sorry again. |
|
Ah no worries about it, I've forgotten all about it as well |
2 participants
This patch attempts to fix two corner cases which can result in hhpc
becoming deadlocked, #9.
The first checks for the case where nanosleep() is interrupted which
would set working to 0 and as hhpc never recovers from this case, the
program will lockup indefinitely. Instead we resume where we left of if
encountering EINTR.
The second fix is a bit of a hack; on rare occasions the select() call
in waitForMotion can seemingly block and practically never recover (I
didn't wait for more than one hour).
Instead I added a timeout to the select call, based on the gIdleTimeout
value if it is greater than 1, otherwise it uses a 3 second delay.
(This implies you can set a timeout of 2 seconds and have a 2 second
timeout.)
This doesn't really solve the issue of why select is getting stuck but
does allow hhpc to recover and prevent it from holding the mouse
hostage.