Conference decwet::winnt-clusters

738.0. "[DEC Cluster] Strange Cluster Problem with SQL" by ALFAM7::SIEBOLD (Thomas, TSO Munich - ALPHAholics can not be VAXinated!) Fri Apr 11 1997 07:24

the following was encountered during the demonstartion of a Digital WNT Cluster
at CEBIT97 on Microsoft's Backoffice booth. They demonstrated SQL and SQL
failover on our cluster at this fair.

Assume 2 Alphaservers running NT V4.0 with SP2 and Cluster V1.1 (final).
They are connected to a RAID310 with 2 raidsets. Both servers run SQL server,
each SQL server is primary for a database on the respective raidsets. One of 
the servers has a database (bench) local (not on a shared drive.

The promoter usually demonstrated SQL access from the client through ISQL and
through access from EXCEL and ACCESS. Sometimes he ran a benchmark program
to the local DB. Whenever he showed SQL failover there was no other access
active.

So what happened was:

After several failovers all of a sudden (but reproducable at will after 2-3
failovers) one of the DBs on the shared devices appeared TWICE in Enterprise
Manager on one of the systems. The DBs even appeared twice in the manage
SQL DBs applet from Clusteradmin! They also appeared twice in the output when 
you do the sp_fallback_help.
To repair we rebootet the system where the database where appearing twice,
which fixed the situation.

I know this sounds rather odd, but it really happened regularly.

I have collected:
			SQL error log file
			the event log file
			the sp_fallback_help output
			the cluster trace log
			the select * from sysdatabases output

from both systems.

I will not include it here but will send it from my exchange system directly to
Carl, Will, and Donna.

Thanks
	Thomas

This also happen to my Cluster setup!   I have encountered same problem
during my demo to customer.  

Somebody pls help us out and just tell me what leads to the problem!

Rgds,
:| 
[Posted by WWW Notes gateway]

A status/problem solution update would be appreciated.

If you need more data ( I have sent all the eveidence I have to Donna, Carl and
Will.

Thomas

I tested this scenario today manually failing over a
group which contained 2 enrolled SQL Server databases 
~7 times and did not see a database name duplicated.

I was testing with the V1.1SP1 FT kit.

Thus, if this problem was truly readily reproducible,
then I believe it is fixed in the current FT kit.

Deb,

did you try it with:

	- an open ACCESS or EXCEL connectioan via ODBC?
	- a batch file running on a client doing a SQL query in an endless loop

According to Carl ( he has the query command file and has analyzed the problem)
SQL server was not able to 'shutdown' a client connection (being bombarded by
the endless loop from the client, and failover manager didn't care whether  SQL
server was 'ready' for failover. It just did it!

thanks
	Thomas  

Thomas,

No, I didn't try it the way you described in .4.
I just checked the PTR database but was unable to
quickly find which PTR matched the problem you've
described.  

I think I should quietly bow out of this conversation and
let Davis or Carl provide the real answer...

- Deb