Preventing a package from halting during a network outage

[Dave Welliver]

We have a planned network outage scheduled and I don't want any packages to shut down when the server loses network connectivity. What do I need to do to prevent the packages from failing over to other nodes in the cluster ?

[Patrick Wallek]

You can disable package switching with:

# cmmodpkg -d packagename

I'm not sure how you can prevent the package from shutting down in the event of network loss though.

[Stephen Doud]

If your package has a SUBNET dependency, you will not be able to prevent Serviceguard from halting the package.
Use cmviewcl -v -p to check to see if the package has a SUBNET dependency.

If the package does not have a SUBNET dependency, then SG will not halt the package when the network ceases to function unless heartbeat traffic is interrupted - which is another matter entirely.

[TwoProc]

Usually - if you create a file called "monitor" in /etc/cmcluster, it will not take the package down. Of course, that depends, in your package as to whether or not that feature is defined and where the file should be created that does this.

[TwoProc]

Umm, Dave, that suggestion I gave *DOES* survive network loss. But, I've checked it out and the file would be "

"/etc/cmcluster/your_package_name/monitor"

If you create that file (just put it there), it really shouldn't switch, no matter what. And, I see no reason why it wouldn't survive a network loss, as I've thrown every thing in the world at it for this purpose; bring down the database, drop mount points, etc. , possibly even bringing down the network cards (I believe I actually did this, but I'm not positive any longer as it was quite a while back).

I'm just refreshing this as I'm pretty certain I've given you THE ANSWER, yet from the points (not that I'm complaining about the point assignments, you can just put zero for this one to show you that's not why I'm posting this), I don't think you realize that you've probably got the solution already.

[Thomas J. Harrold]

I've never heard of the /etc/cmcluster//monitor file. Is this an undocumented feature?

-tjh

[Steven E. Protter]

Shalom Dave,

If the package includes a floating IP address and the network goes down I see no way for the package not to crash.

Best shutting it down yourself before the outage. These outages are useful for reboot and system maintenance anyway, perhaps take advantage.

SEP

[PeterWolfe]

>I've never heard of the
>/etc/cmcluster//monitor file. Is this an
>undocumented feature?

I sounds like John is describing "maintenance
mode". Most of the Enterprise Cluster Master
Toolkit (ECMT) example failover scripts support a
maintenance mode. The scripts that monitor the
application (e.g. Oracle, Tomcat, Apache, etc)
have a feature where if maintenance mode is
enabled (by setting MAINTENANCE_FLAG=yes) then
touching a specific file in the package directory
will disable application monitoring by the monitor
script. You can stop the application and the
package will not fail over at this point.

However, this is a feature of the specific control
script/monitoring script. It's not influencing
in any way the detection of network failures
(since SG does that itself - not the control
script).