Subject: RE: MSClustering
Here is a summary of the interaction between NSD, Fault Recovery and MS Cluster Manager. I will be passing this on to the support folks as an update to the Technote you reference. I hope this clarifies the current situation regarding Domino 6 and MS Cluster Manager.
Best Regards, John
What is NSD?
NSD is the core First Failure Data Capture (FFDC) Tool. In the case of an exception or Panic in the Domino code, the Domino exception handler gets control and invokes NSD to collect diagnostic information on the failure. This information includes environmental data, Process Information, Thread/Process call stacks, Local and Shared Memory Dumps, File System Information, etc… When data collection is complete, NSD will clean up memory, files and processes and generally prepare the Domino Server for a clean restart. (This is true for Windows only. On Unix the cleanup is done internally, but conceptually, the principle is the same)
What is Fault Recovery?
The Fault Recovery feature was new to Domino 6. Fault Recovery enables a server that has terminated abnormally to shut itself down and restart automatically, without any administrator intervention. When Domino encounters a situation which is considered a fatal error, diagnostic information is collected via NSD, each Domino process is terminated and all associated resources are released. There is lots of information on Fault Recovery in the Domino Administration Guide and Release Notes.
What happens when MS Cluster Manger is used for Domino Application Fail-over?
If the Domino Server fails (takes an exception) NSD gets control for FFDC as described above
MS Clustering will take away the file system while NSD is running
NSD will NOT be able to capture diagnostic data about the Domino failure
NSD will NOT have an opportunity to clean up the processes
The Domino Server will be left in a “bad” state and the MS Cluster Manager will not be able to fail back. I have heard some describe this as “ghost” Domino processes on the failed server.
There has been some discussion that NSD and Fault Recovery can be disabled on the server. This may allow MS Cluster Manager to perform application failover (although this has NOT been tested or certified by IBM) However, disabling NSD makes Domino fundamentally “unserviceable” because we capture no data concerning product failures. This is the basis of the statement that MS Cluster Manager is not supported for Domino fail-over.
Furthermore, we believe that the fastest and most efficient way to recover from an abnormal termination of the Domino server is Domino Fault Recovery. This feature allows for efficient FFDC, proper clean-up of processes and other resources and efficient restart of the server.
Can MS Cluster Manger be used for Hardware, Network and Operating System Failover?
Yes. There is no known reason why MS Cluster Manager can not be used for Hardware, Network and Operating System failover and continues to be supported. Domino development does not provide scripts to enable this functionality. If technical help is required in writing these scripts contact your IBM Representative to engage IBM Software Services for assistance.
How can I terminate the Domino server on Windows?
Quit at the console or “nserver -q”
This is the cleanest method to shut down the server. All client activity will be quiesced, all databases will be closed and the server will terminate cleanly.
nsd -dumpandkill
This method uses NSD to first take a diagnostic snapshot of the server processes, memory and environment and then forcibly terminates the server. Any databases opened at the time of invocation of the nsd -dumpandkill will require fixup when the server is restarted. The diagnostic information collected can be used by the support teams to analyze the problem when forced the abnormal termination of the server.
nsd -kill
This is similar to the above, but NO diagnostic data is collected. The server is forcibly terminated. Any databases opened at the time of invocation of the nsd -kill will require fixup when the server is restarted.