Hi.
Last Wednesday we hit this bug about "No space left on device" while
"df
-i" and "df -h" shows about 33% and 65% utilization.
We found this
http://www.mail-archive.com/ocfs2-users at oss.oracle.com/msg03060.html and
other similar posts.
We know that we heave hundreds of log files from apache, some of this
files grows up to ~1 GB size. We guess that we are very fragmented
because files grows slowly during day (we rotate logs every night).
We resolve this by compressing old logs - this reduce fragmentation and
free dozen GB's :)
Now we want to predict this kind of situation. It is possible to check
how many "inode allocator" space are free on every node? And when
it's
low how to check a chance go grow up "inode allocator" in current
condition?
ps: Version: Debian 4.0 and OCFS2 from official package v 1.2.1.
I know about some improvements in current version (about stealing
inode_alloc from other nodes etc.) and about newer package in Debian
5.x, but today I can't shut down cluster for so long and can't risk any
problems during and after upgrade.
ps2: sorry for my poor english
Best regards
--
-----------------------------------------------------------------------
S?awomir Parysek
Horyzont Technologie Internetowe Web: http://www.horyzont.net
tel. +4861 858 22 55 fax. + 4861 858 22 56
ul. P??wiejska 24 61-888 Pozna?, Polska
