Camelia, the Perl 6 bug

IRC log for #gluster-dev, 2013-04-20

| Channels | #gluster-dev index | Today | | Search | Google Search | Plain-Text | summary

All times shown according to UTC.

Time Nick Message
02:35 bala joined #gluster-dev
03:05 yinyin joined #gluster-dev
03:50 yinyin joined #gluster-dev
05:18 H___ joined #gluster-dev
05:25 H__ joined #gluster-dev
05:30 mohankumar joined #gluster-dev
06:26 spai joined #gluster-dev
09:57 badone joined #gluster-dev
10:50 bulde joined #gluster-dev
11:51 hagarth joined #gluster-dev
13:14 bulde joined #gluster-dev
16:33 sandeen joined #gluster-dev
16:33 sandeen hey bfoster  :)
16:34 sandeen can anyone enlighten me on the lifecycle/workload for the indices/xattrop/xatterop-* file and other files in that dir?
16:44 foster sandeen: o/
16:44 foster see my mail?
16:53 sandeen this morning yeah
16:54 sandeen I'm not sure why I didn't see any link/unlink/rename/whatever for that inode in the trace
16:54 sandeen but *shrug*
16:54 foster which inode?
16:55 sandeen the one that raced in xfs_iunlink
16:56 foster well all of the link/removes under that directory refer to the same inode
16:57 sandeen I jsut mean we had:
16:57 sandeen xfs_iunlink: dev 8:16 ino 0x83a8b
16:58 sandeen to get to xfs_iunlink, we should have gone through oneof the other tracepoints
16:58 sandeen not sure it matters
16:58 foster oh, right
17:00 sandeen I'm trying a reproducer based on your observations, we'll see
17:06 foster sandeen: couldn't that mean the race is leading to an extra link drop?
17:08 sandeen not sure I follow?
17:08 foster two racers attempting to remove the same link
17:09 foster if the locking is busted, they both drop nlink..?
17:10 sandeen I was trying to figure this out last night
17:10 sandeen even if the locking works right, if we both get into xfs_remove, the one that waited for the lock proceeds & drops the link again, I think
17:10 sandeen but that would have gone past 0?
17:10 sandeen ugh
17:11 sandeen I was also wondering if this race needs to be on removing hardlinked files from 2 different dirs
17:11 sandeen because I think there's more vfs locking above when we remove from a single dir
17:11 sandeen not sure I'm making any sense :/
17:12 foster hrm yeah, I'd have to skim through the code
17:13 foster what's interesting is the xattrop file is its own file (not a link to something else)
17:13 foster and then everything else in that directory is a link to that inode
17:13 foster and the # of links ebbs and flows with activity through the replication framework
17:14 foster and if I followed correctly, the inode effected by the iunlink race is that inode
17:14 sandeen I was trying to do a loop of link/unlink to a target in a dir, with 16 different threads, so i.e. ended up with:
17:14 sandeen # ls -li /mnt/test/testdir
17:14 sandeen ...
17:14 sandeen 70 -rw-r--r--. 8 root root 0 Apr 20 13:08 e698c855-2f3a-4e32-8f40-760fb64eaad4
17:14 sandeen 70 -rw-r--r--. 8 root root 0 Apr 20 13:08 target
17:14 sandeen with many uuid files linked to that target
17:14 sandeen but never going over link 1 I guess
17:14 sandeen well
17:15 sandeen wait
17:15 sandeen I'm tired :)
17:15 sandeen 16 threads linking/unlinking in a tight loop and so on average usually around 8 links existing in the dir I guess
17:15 sandeen anyway, nothing yet
17:16 foster yeah, ok
17:23 mohankumar joined #gluster-dev
17:26 sandeen (I think the xfs_link tracepoint might need to do something with the lengths, we get junk off the ends of filenames)
17:28 sandeen anyway, it'd be nice to know if any other dirs have links to this same file, as well
17:30 foster a quick test showed that it didn't (e.g., do some writes, check the file, find -inum turns up nothing else)
17:31 foster that doesn't mean there isn't something intermittent/non-obvious going on in the background of course
17:33 sandeen (ah that name thing is fixed upstream f616137 xfs: trace xfs_name strings correctly)
18:15 * sandeen has no luck so far
19:41 * sandeen heads to SFO
21:55 sandeen joined #gluster-dev
23:42 badone joined #gluster-dev

| Channels | #gluster-dev index | Today | | Search | Google Search | Plain-Text | summary