Hi All,
I have a customer who is currently seeing a problem with an SMF service
they've written.
They are running a variety of clustered systems, and as part of their
testing need to be able move from SC 3.2 back to 3.1U4 (as this is
something that can happen in a customer scenario).
With SC 3.2, they have persistent SCSI reservations enabled, and in
their 3.1U4 config, they don't. When moving back, this means that they
have to remove the persistent reservations, in order to avoid cluster
panics.
They have created their own service to do this, which they call
SMUscrub.
It has the following dependencies:
<dependency
name='fs-minimal-SMUscrub'
grouping='require_all'
restart_on='none'
type='service'>
<service_fmri
value='svc:/system/filesystem/minimal' />
</dependency>
<dependent
name='SMUscrub-bootcluster'
grouping='require_all'
restart_on='none'>
<service_fmri
value='svc:/system/cluster/bootcluster:default' />
</dependent>
The start script of the service looks like:
#!/bin/sh
[...disk scrub actions...]
(sleep 5
/usr/sbin/svccfg delete -f svc:/system/device/SMUscrub) &
exit 0
They have found that trying to delete the service in their start
script leads to a failure in other cluster startup dependencies,
that eventually results in the cluster node panicing during the
boot sequence.
If they remove the service delete from their script, then the cluster
startup occurs normally.
We have tried to argue that removing a service whilst it is running
is unwise, but they wish to know where it is documented that they
cannot do this.
If anyone can give me some pointer to documentation that might reinforce
our argument, that would be great.
Thanks,
Andy.
--
Andrew Fox E-mail : andrew.fox at sun.com
Software Support Telephone : +44 (0)1252 420000
Remote Service Delivery Fax : +44 (0)1252 421658
Sun Microsystems EMEA
