HP OpenVMS Systems

Remote termination mailboxes? Lock Manager?

» close window

The Question is:

 
VMS 7.2 (clustered, but no Galaxies)
We have an application that makes sure processes are
are always up. It uses $CREPRC to start processes
and the termination mbx to detect failure. We
have it set up so that if the management process
restarts, it "reconnects" to the termination mbx.
 
As of VMS 7.x, $CREPRC can handle processes on
different nodes. It looks like there should be
no problem implementing a single manager to handle
the same sort of tasks it does today. However, this
leaves a single point of failure - the node the manager
is running on. We want to implement primary and
secondary managers running on different nodes.
We will use the DLM to handle failover. However,
we aren't sure if we can make the termination
mbx on the secondary mgr work. Obviously there
is something in the code that "emulates" a mbx
for non-local processes (SMISERVER relay?), but
it's not at all obvious that this would continue
to work if the active manager changes nodes.
 
Do you have any suggestions on how to handle this
problem?
 
Thanks,
Marshall
 
 

The Answer is :

 
  The termination mailbox will operate for remote $creprc operations, but
  use of the DLM (or a package based on the DLM) for all process control
  would be generally prefered by the OpenVMS Wizard.  (Assuming that the
  termination mailbox is used solely for the notification of a process
  exit, then once you start to utilize the DLM the termination mailbox
  becomes largely irrelevent.  Use of the DLM also tends to ease cleanup
  processing on node crashes, and failover and primary process election
  processing in general.)
 

  

     

     
answer written or last revised on ( 10-MAY-1999 )
     
» close window