cancel
Showing results for 
Search instead for 
Did you mean: 

/var full

ericfjchen
Regular Advisor

/var full

This is HP Tru64 5.1A cluster with 2 nodes. One node crash and we find the /var/adm/messages with "File system full: /var". The file system '/var' full will cause one node crash?
5 REPLIES
Ralf Puchner
Honored Contributor

Re: /var full

maybe it is time to open case within the HP support center and ask for crash analyze, all others will be a wild guess.....
Help() { FirstReadManual(urgently); Go_to_it;; }
ericfjchen
Regular Advisor

Re: /var full

Besides /var/adm/messages, what file should we send to HP?
Hein van den Heuvel
Honored Contributor

Re: /var full

I don't have a live cluster ready to check the exact details, but in a cluster good chunks of /var filesystem are unique per member through a {member} link, some are shared. The per-node space is shared with other per-node directories like /dev, /etc,...
I would just use the classic 'who is eating the space' commands like

find -size +2000000c

and looking at recently accessed (log) files.


Hein.
Victor Semaska_3
Esteemed Contributor

Re: /var full

Since (most) system logs are kept in /var/adm/ we decided to set up a separate partition for each member in the cluster. They ended looking like this:

varadm1_domain#varadm 2097152 206659 1883280 10% /var/cluster/members/member1/adm
varadm2_domain#varadm 2097152 200980 1889208 10% /var/cluster/members/member2/adm

That way should a member start generating large amounts of entries in a system log only the /var/adm partition for that member would fill up and not the cluster-wide /var partition.

Vic
There are 10 kinds of people, one that understands binary and one that doesn't.
Johan Brusche
Honored Contributor

Re: /var full


Are you sure "/var full" isn't the consequence of the panic? When the node came back in the cluster it wrote it's memory dump into /var/adm/crash, and that can be very large for busy mutli-gig systems.
Anyway a filesystem full should not panic a system.
(When strict security auditing is in place it could however start a proper shutdown if /var is the place where auditing stores it's logs.)
For crash analysis the HP support engineer needs the vmunix+vmzcore from /var/adm/crash, but best thing to do, is to run "sys_check -escalate", and that will collect all relevant files to diagnose the problem.

Regards,
Johan.

_JB_