[Samba] Re:[Samba]Upgrade broke file locking

[adam at morrison-ind.com]

>>I saw you message in the Samba archives concerning corruption of the
>locking
>>database.  Since we upgraded from rh70 to rh73 a few days ago we are
>ee-ing the same thing.  No change to Samba.  Did you ever manage to resolve 
>this situation?
>No, I haven't found a solution.  Currently, I'm restarting samba
>services each night and cleaning out the connection.tdb (and others in
>the same area) files. 
>This seems to have improved it a bit, but occasionally it does break
>within a 24 hour period.  So it seems to be some sort of leak or buffer

We can go ~4 hours tops :(

>overrun problem, but I can't figure what triggers it.  Unfortunately we
>have > 150 clients clients using this machine, so I can't afford to
>cripple it by turning up debugging, and on the other hand if I farm off
>some of the load, the problem probably won't occur.  Additionally, we
>are using W2K terminal servers which is unusual enough that the
>developers probably don't get much testing done on them.  The issue
>could even be some sort of race condition for multiple clients sharing
>the one samba smbd process (this is what W2K TS does).

Are you using an SMP box?  I'm tempted to reboot into UP mode and see what happens.

I went int source/tdb and did a make to get the tdbtest, tdbtool, and tdbtorture
utilities.  tdbtool can open and parse the locking.tdb file and tdbtorture runs
and returns "OK".  So this shouldn't be a simple tdb and my version of glibc
kind of problem.

>I haven't tried 2.2.6pre1, I've been hoping for a release of the full
>2.2.6.  It seems to me from reading the changelog for 2.2.6pre1 "12)
>Memory fixes caught by Valgrind" could sort out some issues, but I'm
>just guessing. 

>I had considered going back to 2.2.3a for our fileserving (which worked
>mostly OK for us) and joining it to a 2.2.5 PDC.... messy.

Possibly same thing here.  Our 2.2.5 PDC is on a different box and that works great.

>I'm thinking about reposting to the list again, see if I get any
>responses the second time around.