Hello,
I have a 1TB ocfs partition mounted on mail server. It was a cluster
before, but as one of the machines failed, the cluster has been taken
by only one machine for ages.
This server has courier-pop and imap, postfix and maildrop to deliver
messages to the mailboxes.
System is debian Etch, ocfs2-tools version 2.1.
Recently, I tryed to insert a new machine into this cluster.
Just right after having the cluster back with two nodes, with both
machines able to write and read from the storage volume, configuration
correct and communication between machines ok, the I/O went so slow
that could not be used. A cd or ls in a directory in the volume took 3
or 4 minutes. Any imap connection timed out.
I haven't copies of the logs and command outputs now, but I checked
some thinks at the moment of the problem. I could ping from one
machine to another (no pakages lost, quick echo). I was able to write
in the discs from both machines. Unplugging one of the nodes cause the
other to show logs on the console and the I/O to be fast again.
Looking at some graphics in cacti and zabbix afterwards, I saw that
the I/O rate was VERY VERY high at the time right after the insertion
of the new node.
I'll do a new attempt today. Are there any specific tests I should run
to figure out what the problem was?
What logs should I show in order you're able to help me?
Does the ocfs service perform any tests when a new node is inserted
into the cluster, what could make the I/O so slow?
Does it do something like creating a local journal, so that the volume
access was so bad at that moment?
Thanks in advance for any help.
--
Krishna
