In-Depth
From the Trenches: The Case of the Disappearing Printer Shares
What do you do when the standard approach to problem-solving doesn't work?
Working in technical support allows you to experience
problems across a wide range of products. Inevitably,
you find that technical issues are reported on certain
products more than others. I've gained an appreciation
for those particular problems that involve products I
personally don't know much about. The tale I spin here
revolves around Microsoft Cluster Server.
The Environment
The phone rang, and I braced for impact. I learned that
my customer was having an "unusual problem" with his NT
Cluster Server version 1.0 build 224.0. This cluster server
has 90 printer shares defined on it and a total of 200
to 250 shares available. The 90 physical printers have
HP JetDirect cards in them. This cluster server has been
in place for two weeks.
The Problem
The customer has found that when he views the cluster
server via UNC name, five or six printer shares mysteriously
disappear. No errors are reported, and no trigger event
can be identified. Poof! The printer shares just
plain disappear. Fortunately, the customer has developed
his own workaround: to manually recreate and reconfigure
the printer share. Over the two weeks of operation, printer
shares have disappeared three or four times. Each time
the print shares disappear, the problem involves different
printer shares, and none that have previously vanished.
When my customer connects to one of the Windows NT servers
within the cluster, the printers are still there—the shares
just disappear on the cluster server.
The Investigation
You'd think that the printer shares can't grow legs and
just walk away. So I began the search to find the cause
of their disappearance. I performed the normal steps of
reviewing both my local copy of TechNet, and the Microsoft
Support Web Site. Neither presented a solution. I even
consulted with other Microsoft gurus, and they're all
stumped.
After exhausting my available resources, a colleague
advised that if he had this call, he would place a call
to Microsoft. That sounded like a good plan to me, so
I opened my call at Microsoft Technical Support. The technical
support representative walked us through verifying that
the Cluster Server configuration was correct. For example,
we made sure that clients connected to the network name
when connecting to the cluster. Also, in checking Cluster
Administrator, we noticed that the actual printer resource
was disappearing as well as the share point.
It turns out that this was a known issue, internal to
Microsoft. The Microsoft Support engineer explained that
an internal document was being prepared for publication
describing our problem. At the time of this writing, Microsoft
has published the article as Q225081, "Cluster Resources
Quorum Size Defaults to 64K."
The problem lay within the cluster quorum. The quorum
is where the cluster service maintains its current status.
If a failover occurs, the receiving node checks the quorum
for any updates and the current status. The quorum is
also used to designate one of the nodes as the "master"
in the event of a network failure, where the cluster service
is unable to determine which node is failing. All services
would then be failed over to the current owner of the
quorum resource.
The key to our problem was our quorum log size. This
was set at the default of 64K-somewhat small. This new
article indicates that when using the small default quorum
value, problems may occur if the number of shares or the
transactions increase significantly. For example, the
support representative explained that one of the problems
you might see is that some of the cluster resources may
disappear. Pay dirt! We now understood what was happening
to cause our printer shares to disappear.
The Solution
The workaround was to do the following:
- Enter Cluster Administrator.
- Right click on the cluster name on the top left.
- Click Properties | Quorum.
- Change the log size from 64K to 4,094K (just under
4M).
- Click the Apply button.
After making these changes and monitoring the server
for two weeks, we found no additional printer share resources
disappearing. My customer considered the mystery officially
solved.
The Lesson
Specifically in regards to Cluster Server, I learned
that we should always make sure that our Quorum log size
has been increased from the default value!
Generally speaking though, you should expect the unexpected.
We can all follow the standard troubleshooting steps in
trying to resolve a problem. We can all take the time
to ask the right questions and listen closely to our customers.
You'd expect that this would lead to a clear problem resolution
in most cases. Because of the unexpected though, I still
never go to a customer site without my magic-imbued rubber
chicken in my work bag.
More importantly, be prepared to call Microsoft for
support if nothing else pans out. Fortunately my customer
and I were able to get that quick solution we all hope
for.
About the Author
Rob Vazzana, MCSE, MCP+Internet, CBE, works as a Senior Systems Consultant for an international PC integrator/enterprise support provider, which specializes in server and network technology support. Rob specializes in Banyan VINES and Windows NT. He likes technology so much that he's built a computer in his car so he can use a cell phone to get on the Internet. Visit his Web site at http://surf.to/maytag.