From: Chris Adams (cmadams@hiwaay.net)
Date: Sat Sep 11 2004 - 16:22:13 EDT
Once upon a time, Chris Adams <cmadams@hiwaay.net> said:
> I am seeing inconsistencies in the cluster file system in the file
> access time stamp. Under some conditions (which I haven't figured out
> yet), one cluster member will see a different atime than the other. The
> mtime and ctime are the same on both, and the member that last wrote to
> the file sees the correct atime, but the other member (which is the CFS
> server for the domain) sees the atime as the same as the mtime.
The is from a while back - I didn't get any responses from the list, so
I opened a case with HP. It took some digging, but they finally figured
out that the problem was in the CFS and AdvFS parts of the kernel.
Basically, when one cluster member had to generate an internal partial
read (when appending to an existing block, the existing part of the
block is read and then the whole block re-written), the CFS server for
the fileset would treat it the same as a regular access and update the
atime.
I received a patch from HP that fixes this, and it should be fixed in
the next patch kit.
-- Chris Adams <cmadams@hiwaay.net> Systems and Network Administrator - HiWAAY Internet Services I don't speak for anybody but myself - that's enough trouble.
This archive was generated by hypermail 2.1.7 : Sat Apr 12 2008 - 10:50:07 EDT