The Question was:
With ase clustering how do you run the cron job on the node that is "active"
only? As I log _every_ cron job, I want to avoid 50%+ of logs showing
failure.
Many thanks to all that responded:
Scott Mutchler
Ivan Hoe
Richard Tame
Robert Mulley
Danielle Georgette
Hoai TRAN
Alex Nord
Rich
Martin Andersson
Kris Chandrasekhar
Con Tassios
Wendy Fong
Larry Clegg
Ricardo Ramos
The answer from 14 of 16 respondents was to run identical crontabs and code
the jobs to exit early if they are not running on the active machine. As,
I was fishing for a more sophisticated solution, I was happy to get two
slick solutions too.
But, the testing effort to implement the slick solutions is quite high.
So, I think that for the next 6 months or so till we upgrade to 5.1,
something like the below will work, when applied identically on both nodes,
will work.
0 6 * * * test -d /app/oracle/local/bin &&
/app/oracle/local/bin/compute_statistics.ksh
This implementation only invokes the script if it's directory has been
mounted. This is different than simply checking for the script to exist,
or the script being executable. Mistakes in spelling and permissions are
common user errors, but hopefully directory will exist unless the mount is
not on the machine.
This seems to me to be a much better solution than I would have had without
all the ideas sent in, THANKS to all the contributers!
Greg
----------------------------------------------------------------------------
---------------
Con Tassios -
I have a cron daemon for each cluster service. The cron binaries and
crontabs reside on the shared storage device which is part of that service.
The ASE script for the service starts/stop the cron daemon. Therefore the
cron daemon for a service moves from one machine to another just like the
filesystems.
----------------------------------------------------------------------------
---------------
Danielle Georgette --
We manage this situation by writing two crontabs for root (or any other
user). One is the plain crontab that contains all the system housekeeping
etc that needs to be done regardless of the node running the cluster
service. The other has all the additions required when it is actively
running the service. We call them root.cluster and root.standalone, and keep
them in /var/spool/crontabs (ie locally) on each machine.
The start script for the service copies the root.cluster crontab to be the
root crontab. The stop script for the same service copies the
root.standalone to be the active root crontab. Then they both grep for the
cron pid and send it a hup.
Voila, works perfectly and is a very simple solution. Just make sure you
make any crontab changes to the relevant root.XXX files and not the running
root cron, because they will be lost on the next cluster transition (it
happened to us and we were rather annoyed at our silliness). There would
probably be a way of keeping the two crontabs in with the cluster info but I
havent done that yet - it would remove the only downside, which is having to
have local copies of the crontabs on each cluster member.
Anyway, hope this helps!
Danielle
ps Robert Mulley, who is also a member of this list, designed this solution,
so some of the kudos go to him for this solution.
----------------------------------------------------------------------------
-------------
Use an environment variable or check a file or mount point to see if Oracle
is running on the node...if you don't find it then let the cron job exit
otherwise it proceeds and does its thing.
When I had an ASE cluster I checked for the mount point since under ASE the
disks are not shared...if I didn't find the mount point then the cron job
exited.
Hope this helps...and get upgraded to V5.0a asap...it is wonderful...
----------------------------------------------------------------------------
-------------
You can still run the cronjob on both machine.
Write the cron job as a script. Have the script check for the activemachine.
Exit for non-active machine.
----------------------------------------------------------------------------
-------------
Richard Tame
The scripts for your cron jobs should be on shared space.
All you need to do is check if the script is executable (test -x), if the
service is on the other node it won't exist and therefore can't be executed.
If you are running a late enough version there is an example of this in the
cron table for 'defragcron'.
[found it]
#1 1 * * * test -x /usr/sbin/defragcron && /usr/sbin/defragcron -p
>>/usr/adm/defragcron.log 2>&
----------------------------------------------------------------------------
-------------
In V40x, you can have your script do "asemgr -d" or "asemgr -d -v
servicename|grep $nodename" where you got the nodename earlier. Check $?
(status) to see if the service is there.
In V50A, you can similarly do a "caa_stat" or "caa_stat servicename" and
parse the output to determine whether the service is on your node.
----------------------------------------------------------------------------
-------------
I would have the script called by cron to check who is the director. Use:
asemgr -d -v
and grep for the name of the host. Then you can decide where to run your
job. Or, df and grep for the name of the nfs partitions.
----------------------------------------------------------------------------
-------------
You could check on what node the service actually runs on, using asemgr -dv
and then grepping for the service name and the hostname.
Or you could just check if the oracle filesystems are mounted in the script.
----------------------------------------------------------------------------
-------------
Have the cron job call a "checkaction" script, used with asemgr so that you
can tell where the database is at anytime.
----------------------------------------------------------------------------
-------------
Write a simple shell script and check for the mount point where the Oracle
database is mounted. If it is not mounted, then you know that the database
is not running and simply exit the script.
Regards,
Hoai
----------------------------------------------------------------------------
-------------
Scott Mutchler
there is an entry in /etc/fdmns called db_domain. I write my cron entries
something like this:
* * * * * [ -d /etc/fdmns/db_domain ] && /usr/local/scriptname
Node B has an identical entry, but since the file domain is not present, the
script exits.
The only caution for using this method is that when you reboot a node - at
least in my configuration - entries get made for all the known filesets
under /etc/fdmns, so we make sure to relocate services back and forth, thus
cleaning up the /etc/fdmns directory. If moving a service is not an option,
the entry can safely be removed by hand on the node NOT running the fileset.
----------------------------------------------------------------------------
-------------
Again THANKS to all who responded. All had good ideas. This list is the
best!
Greg Dora
Gregory.dora_at_citrix.com <mailto:Gregory.dora_at_citrix.com>
Received on Mon Jul 31 2000 - 21:42:02 NZST