A couple of days ago, we noticed the following message appearing in syslog:
Aug 14 15:09:35 zadok smbd[16067]: tdb(/var/lock/samba/locking.tdb):
expand_file write of 8192 bytes failed (No space left on device)
Similar messages would show up for several of the files under
/var/lock/samba.
Running df showed that /run/lock was full. Not having ever paid any
attention to /run/lock before, I thought that this was normal, but RiXtEr
clued me in on #samba.
There was a difference in the reported size of the lock files in
/var/lock/samba depending on whether we used 'ls -l' or 'du -s'.
Two files
were about 4GB in size as reported by 'ls' and well under 5MB (the size
of
/run/lock) as reported by 'du'. The big files were locking.tdb and
smbXsrv_open_global.tdb,
Restarted Samba brought those files down to...very small. But we've noticed
that they grow quite quickly, seemingly whenever a new user logs in (but
we're really nt sure of that).
Right now, /run/lock is sitting at 54% (2.7MB). This is just in a testing
environment, with four workstations and twenty test users. Some of the
production environments that we are aiming to roll out real soon would have
more than 100 workstations and several hundred users.
The Samba4 server is running Ubuntu 12.04LTS and the Sernet packages from
the PPA. Our configuration is quite simple, just a Samba4 PDC/AD with a
share for redirected user folders using the built-in LDAP and DNS.
We will have two more test environments up and running shortly and could
use them to test, compare, and contrast. Any suggestions for what to try
are welcome. If we see that one of the test environments doesn't exhibit
these symptoms, we'll try to home in on the difference.
Should this be posted to samba-technical?
Mark