Berkeley DB on Windows Cluster

Question

Does anyone have experience running a Windows service that uses Berkeley DB on a typical active/passive Windows Cluster with shared disks? Is it even possible with Berkeley DB? Just to avoid confusion, I am not talking about replication, but about a failover situation, where one instance of the service gets deactivated, another one activated, and both would access the same (shared) RAID5 disk array which holds the BDB database files.
Karl

Answer 1

Hi Karl,
Berkeley DB has not been tested on Windows clusters
and so it is not an officially supported platform.
Here's my understanding of the scenario:
There are two nodes that are connected to the same
shared disk. One of the nodes is active, and the
other is a passive standby. The Berkeley DB
application is running on one node and accessing data
on the shared disk. When the primary fails, the
Windows cluster services (I believe it used to be
called Wolfpack?) will allow you to failover the
application to the other (passive) node and continue
processing. There are also services to fail back,
once the original node is restarted.Yes, that is correct.
>
I believe that Windows cluster services will control
access to the disk and ensure that BDB running on the
passive node will only access the disk when the
primary fails.Yes. Actually, the passive node has no control over the
shared disk, so the BDB application can only open the
environment after the passive node becomes active and
gains access to the shared disk.
So, I'm tempted to say that this scenario should
work, assuming you're using Windows services
correctly. To my knowledge, there's nothing in
Berkeley DB to prevent this usage. On the other
hand, you'll be relying on the underlying OS services
to give the correct semantics wrt. accessing the
shared storage. For example, if BOTH nodes were to
write to the shared storage, you will most likely get
data corruption, since BDB does not synchronize disk
access ACROSS nodes.That will not happen, as the services on the passive node have
no access to the shared disk.
I'll continue to ask around for a better answer. In
the meantime, I'd encourage you to continue your
testing and keep us posted. If you prefer, you can
contact me directly at ashok dot joshi at oracle dot
com.Thanks for your feedback.
My current theory is that a failover is equivalent to a
BDB application crashing and being restarted.
So one has to re-open the environment and run recovery
before continuing work.
Karl