RE: piranha fos setup

Thanks so much for the response.  I'll try your attached patch.  Obviously
my configs were just for basic testing - my 
start_cmd and stop_cmd that simply touch and remove files, would be replaced
with the startup and shutdown commands for the db, and my test_ha.sh would
be replaced by a valid db check.  Are you saying that the send_program only
executes on the failover box periodically, to perform a validation check on
the primary?  If this is the case, my test below would have to be slightly
modified so that the test looking for a semaphore, (or existence of a file),
would execute on the partner.  Guess I was thinking this validation
test_ha.sh script ran periodically on the primary node.

Just to clarify another point - with either the C code changes below or the
patch you supplied, I won't see fos trying to start or stop X.X.X.100:80
monitoring services, correct?

Brian

-----Original Message-----
From: Sébastien Bonnet [mailto:[email protected]]
Sent: Friday, December 05, 2003 3:49 AM
To: [email protected]
Subject: Re: piranha fos setup


Hi Brian,

> It appears that fos wants to monitor port 80 for cluster health,
> even though I've omitted this port designation and am using
> send_program within my lvs.cf 

If my memory's right, there's a bug which prevent using send_program
correctly. If you have C programming skills, edit fos.c. Near line 300,
you will find :
*arg++ = strip_quotes (send_program);

replace it by (program becomes str):
+		*arg++ = strip_quotes (send_str);


But, I wonder if I haven't also made changes in nanny.c. If the above
fix does not work, try the attached patch, which does fix !

>         send_program = "/etc/sysconfig/ha/test_ha.sh"
>         start_cmd = "/bin/touch /tmp/primary"
>         stop_cmd = "/bin/rm /tmp/primary"

> 
> test_ha.sh:
> 
> #!/bin/sh
> if [ -f /tmp/primary ]; then
>    echo "OK"
> else
>    echo "FAIL"
> fi

Ouch ! I guess you have understood how it works. start_cmd and stop_cmd
should be set to the commands used to start and stop your database. 

For example :
start_cmd = "/bin/service postgresql start"
stop_cmd = "/bin/service postgresql stop"

The script should try to connect to the other host and check the DB is
available. Keep in mind the script (actually the test nanny has to run)
is only started regurarly on the backup node (the one the DB is not
running on).

If you have replication between both nodes, I understand your start and
stop commands (you want both DBs to run simultaneously), but the script
you still check the availability of the DB on the other host.

If you set it up the way you did, you won't have failover because
piranha is not able to know the DB on the other node has crashed !

Hope this help

-- 
Sébastien Bonnet
  Centre de contacts - Experian France