[Gluster-users] Stale NFS file handles related to self-heal

[Peter Linder]

Hello,

When executing "find <gluster-mount> -noleaf -print0 | xargs --null stat 
 >/dev/null" on a directory mounted with gluster NFS, some files get a 
stale nfs file handle error message that doesn't go away for quite some 
time sometimes.

This was discussed yesterday on #gluster with users JoeJulian and 
jdarcy, and I was asked to file a bug report, but I'm posting here first.

I can trigger this quite easily, as my volume has some symlinks 
scattered around that always trigger a  self-heal (bug 3525). It seems 
that when a self heal is triggered, the NFS server sometimes (not every 
time) sends ESTALE for 1-100 or so files that are being accessed by 
find... stat during that time. I haven't yet confirmed that stating 
those files from a different client a second later yields the same 
problem, but for the client that gets the stale nfs handle the problem 
remains for some time, until its cache is cleared i guess.

The native FUSE client does not seem to be affected. I'm using version 
3.2.4, and the volume was just created with that version.