Ow Mun Heng <Ow.Mun.Heng at wdc.com> writes:
> On Fri, 2008-02-22 at 21:31 +0000, Christopher Browne wrote:
>> Craig James <craig_james at emolecules.com> writes:
>> > Ow Mun Heng wrote:
>> >> I came to work today and seems like the slave server died. (power trip?
>> >> No it was not connected to a UPS :-()
>> >> I've not been able to locate/determine if the slave is really dead or
>> >> otherwise and it's the weekend in Asia and there's no one in the office
>> >> till Next week.
>> >> As of now, the master is still trying to contact the slave (slon is
>> >> still running on the master) and log_1 and log_2 is filling up.
>> >> And yesterday, I just created a job to manually force the log_switch
>> >> to
>> >> occur. So, right now, I'm at a loss as to what i can do.
>> >
>> > Just kill all of the Slony daemons.  Next week when the other server
>> > is back, start them again.  It will figure out what it missed, and
>> > will catch up with no problems.
>> 
>> That's not quite accurate...
>> 
>> If you kill ALL the daemons, and don't have *something* maintaining
>> the creation of SYNCs (e.g. - a script running the "generate_sync()"
>> stored function), then there will be one really gigantic SYNC covering
>> the interval of [time slon for origin died] until [time slon for
>> origin restarted].
>
> I remember reading that in the docs, but I took the advise anyway and
> killed the slon daemon a few hours ago.

You'll have a Pretty Big SYNC at some point, then.  No reason for it
not to work, but it'll be big...

>> a) Set up generate_sync() cron job, and kill all slons.
>
> Where is this generate_sync() anyway? I only saw a
> generate_sync_event(interval) stored function in the cluster DB.

Ah.  Working from memory on that.  generate_sync_event() would be the
one.

>> b) Increase the various sync parms for the slon for the origin node;
>> -s 60000 and -t 120000 will mean you SYNC once per minute, when things
>> are busy.
>> 
>> That reduces the work level a bit., either way.
>
> Thanks. I've already restarted the process on the origin with 
> slon -c2 -d2 -s60000 -t 120000 (actually I was already using -s60000)
>
> I guess my concern now is that the slon logs are filling up and going
> past the 2GB threshold (for both log_1 and 2 which would mean that
> there's not going to be much help in gettting things back to speed when
> next week comes)

Yeah, it'll be a lot of catching up.

You'll get a feel for whether or not you have enough hardware to be
able to catch up at all, which is a useful thing to know.
-- 
output = ("cbbrowne" "@" "acm.org")
http://cbbrowne.com/info/finances.html
"Administering  a Linux  server  is no  more  difficult than  properly
running Windows NT."  -- Infoworld, November 24, 1997