Hide Forgot
Hi Timoty, Yes, we know that it is an issue with quick-read. Currently in 3.2.x branch the issue is not fixed, but the development tree (branch for release 3.3.0) has the fix. We will evaluate when the fix will be available in 3.2.x branch. Regards, Amar
Dear Gluster Dev Support Team, Thank you for a great product. Backend: Scientific Linux 5.4 X86_64, Intel Quad Core, 16gb RAM App Version: gLuster 3.2.2 Network; 1gigE Servers: 8 Servers with 8 bricks for the distributed striped volume. This morning I noticed that our GlusterFS started to act a bit strange when writing files to the glusterFS mount. I wrote a few text files to the gluster mount and reading the files produced a binary output. Somehow the files had been converted to binary when read back. echo "timothy" > testfile vi /mnt/gluster/testfile ( this produced the output of " ^@^@^@^@^@^@^@^@^@^@^@ " ) od -x /mnt/gluster/testfile ( produced the output of "0000000") Some of the bricks contained more information in the file than others. This has to do with the striped volume. I then continued, with very good advice from the IRC channel on freenode, to disable a option on the volume called " gluster volume set <volname> performance.quick-read off ". I assume this is the translator for performance.quick-read. Immediately upon disabling the quick-read option on the volume I was able to read the data. Thanks Timothy Carr University of Cape Town
CHANGE: http://review.gluster.com/111 (also do some cleanups) merged in release-3.2 by Anand Avati (avati)
fix already exists in master branch. Also its now pushed to release-3.2 branch. In release-3.1 branch we don't want anything because these xattr aggregation logic came with the marker translator (ie, only in code after release-3.2).
awesome. Thanks so much for the hard work. I will put to the test Cheers Timothy
Works fine with both quick-read on & off.