Hide Forgot
I manually set trusted.glusterfs.dht on the root directories of the 3rd replicas for all my bricks to be identical to the other two copies and have not experienced the "Stale NFS file handle" issue since.
This actually on 3.0.8, not available for selection. On several new volumes, trusted.glusterfs.dht was not written to the root of them, causing the fuse client to eventually enter a perpetual loop, repeating: [2011-06-08 12:18:53] D [dht-layout.c:649:dht_layout_dir_mismatch] uspaweb-dht: / - disk layout missing [2011-06-08 12:18:53] D [dht-common.c:274:dht_revalidate_cbk] uspaweb-dht: mismatching layouts for / until unmounted. This would cause the user to receive the "Stale NFS file handle" error message when accessing that mount. I rolled further back to 3.0.7 before I discovered where the problem was coming from. Here's the full tracelog of the client: http://joejulian.name/tracelog-3.0.7-dht_layout_dir_mismatch.log I realize that you've stated in another bug report that the gfid change in 3.1+ obsoletes this xattr, but the fact that it's not replicating to the 3rd replica may still be significant in the current tree.
Hi Joe. I couldn't reproduce this. Created a dist-rep setup with 6/9 bricks with replica=3 using gluster 3.0.8 built from source and tested several times. Every time the 3rd replica gets the trusted.glusterfs.dht gets set on the root dir of the brick. Can you confirm this still exists. And if it does can you give a test case for reproducing this. Thanks.
Joe, Are you still using 3.0.x series? We would like to test mostly 3.2.x or master branch for any of the bugs as of now. (In some cases 3.1.x may be valid). Let us know if it is ok to close this bug considering 3.0.x is not actively worked on. I will make sure we will add this to the test case to test master.
I no longer have a 3.0 installation to test against. I'll go ahead and close this.