On Fri, Nov 14, 2008 at 4:50 AM, James . <[EMAIL PROTECTED]> wrote: > On Fri, Nov 14, 2008 at 2:55 AM, <[EMAIL PROTECTED]> wrote: >> >> "James .": >>> After disabling the must_new check on the latest version (20081103), I >>> have gone 3 days without seeing this bug (normally, it would take a >>> mere 20-30 minutes before inotify fails), and I dont think its a >>> coincidence. I dont know how this discovery might help but if you need >>> any extra information, im more than happy to post it here. >> >> What are you calling "this bug?" >> If you have read the source, you would know >>> aufs au_new_inode:362:mv[5446]: Un-notified UDBA or repeatedly renamed >>> dir, b90, squashfs, rules.d, hi23, i303. >> is just a warning. >> >> I don't think that aufs is bug-free, but did you try the inotify_user >> patch? >> >> >> J. R. Okajima >> > > Yes I did try the patch also, no luck there. > > My apologies for some lack of explanation. > The warning, as you rightly point out, is just a warning, and it did > not affect the way the system runs. However, the bug with aufs is that > after the 20080929 change where must_new was added to au_new_inode, > aufs began to miss inotify events after some time. This may or may not > actually be related to the warning. > > Some extra details about this bug: > *Initially, the system will work fine, you can do "touch testfile" and > nautilus will see the changes. > *However, after a period of time, that appears to be RANDOM, inotify > messages are no longer received by gio (the glib inotify interface > that can optionally use gamin as the backend). This I have verified > using gamin debugging. > *Despite a random amount of time before the bug appears, I have bee > able to reproduce this bug RELIABLY within 1 hour of running the > system, each and every time. > *Disabling the "must_new" check in inode.c causes the bug to no longer > appear, and the system works fine as if I had gone back to 20080922. > > It is my experience (which i admit is not much) that bugs like these > tend to occur due to some race condition/threads or timer issues. >
I forgot to add also that killing gam_server, killing nautilus or refreshing the inotify watches in any way will cause inotify to work again until after some time, it will stop again. ------------------------------------------------------------------------- This SF.Net email is sponsored by the Moblin Your Move Developer's challenge Build the coolest Linux based applications with Moblin SDK & win great prizes Grand prize is a trip for two to an Open Source event anywhere in the world http://moblin-contest.org/redirect.php?banner_id=100&url=/
