Hide Forgot
Created attachment 164 [details] a fix to SIGBUS in control-panel
I encounter lot of warnings while creating and deleting files simultaneously. Brief error message: [2010-03-23 08:19:31] W [fuse-bridge.c:1025:fuse_setattr] glusterfs-fuse: 1984066: SETATTR (null) (fuse_loc_fill() failed) [2010-03-23 07:34:34] W [fuse-bridge.c:585:fuse_lookup] glusterfs-fuse: 36983: LOOKUP 46912585627456/NMI-RCU.txt (fuse_loc_fill() failed) Reproducing error: 1. Extract and delete linux source tar ball. e.g: for i in `seq 10`; do rm -rfv linux-2.6.33.1 & tar jxvf linux-2.6.33.1.tar.bz2 & done Logs, server and client volume files attached.
This seems completely normal to me. With your test it's OK that particular code point is hit. If there is any problem, it is that we can't really tell away such "legitimate" cases from the ones when loc filling fails because of a bug, therefore we get annoyingly many false positive warnings.
Yes, but the applications that are being run are hung. They become irresponsive. I will create a screen session on amazon AWS and send you the logon details.
(In reply to comment #3) > Yes, but the applications that are being run are hung. They become > irresponsive. I will create a screen session on amazon AWS and send you the > logon details. I have already reproduced the configuration you gave, so if running those apps in a way that they exhibit the errant behavior is neither a rocket science, I could maybe do it by myself. (But if setting up the AWS node is easier for you/us, that's of course fine with me too.)
Any updates on this?
(In reply to comment #5) > Any updates on this? Given that the last thing planned was put up by you... are you asking yourself? ;) I told that I reproduced the situation as it's written in the report, and what I see/reported is regular behavior. You told that some apps hang which _is_ a problem. So you either should give info how to reproduce the hangs as well, or set up that AWS node, hangs included.
> Given that the last thing planned was put up by you... are you asking yourself? > ;) > ;-) Nah, asking you :-P. > So you either should give info how to reproduce the hangs as well, or set up > that AWS node, hangs included. If I recall correctly, I have sent a mail to you. I will forward the same again. :-).
Sacchi, Can you check if 3.0.4rc3 fixes the bug?? Because we could reproduce rm -rf hang with 3.0.3rc1 (when you showed us the bug).. But with patch http://patches.gluster.com/patch/3029/ we could solve the hang issue. It should not be technically related to fuse_loc_fill error. afr_readdirp_cbk (), failover code had a missing frame issue without this patch. Please let us know. -Amar
PATCH: http://patches.gluster.com/patch/3135 in master (fuse: fix too early freeing)
PATCH: http://patches.gluster.com/patch/3135 in release-3.0 (fuse: fix too early freeing)