[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference decwet::winnt-clusters

Title:WinNT-Clusters
Notice:Info directories moved to DECWET::SHARE1$:[NT_CLSTR]
Moderator:DECWET::CAPPELLOF
Created:Thu Oct 19 1995
Last Modified:Fri Jun 06 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:863
Total number of notes:3478

837.0. "<DEC cluster> SQL management in cluster" by NNTPD::&quot;[email protected]&quot; (Sven Nilsson) Thu May 29 1997 07:25

Hello.

Cluster 1.0 SP2 running Dual SQL servers.

Building on topic 800. 

Customer fear that
Cluster complicate the SQL Server management
so much that it is a reason NOT to run cluster.

Give me some good advice to bring to Customer.

1. Should he regard the Cluster as 2 SQL standalone SQL servers
    when working with SQL server management stuff like;

Adding/Dropping Logins
Configure SQL server parameters/tuning..
Adding/changing objects on master and  other local databases...
Working with backup devices..
Working with tasks and alerts

2. Work on the shared clustered enrolled database...

Creating/changing objects
tables/views/procedures/users/permissions;

Is that kind of work LOCAL to the database or
should that kind of work be performed twice
Both when the database is hosted by primary
SQL server and once again after a Failover to
the other SQL server?

Changes in size of database and database device is
documented, unenroll - change - enroll also in topic 800..
Is that maybe a general rule, unenroll - change - enroll
working on the clustered database?

What kind of SQL information is "replicated" between
the SQL Servers when running in cluster, other than
which databases are enrolled, primary server..

Other issues in this area to bring confidence to Customer
or in worst case support the fear of the Customer:-)

Lots of thanks in advance!

Regards/Sven in the ceNTer of Sweden
[Posted by WWW Notes gateway]
T.RTitleUserPersonal
Name
DateLines
837.1MPOS01::naiad.mpo.dec.com::mpos01::cerlingI&#039;[email protected]Thu May 29 1997 08:205
	Your question here is similar to 834.  Might be worthwhile to get a 
	single stream of thought going there.

tgc
837.2Some answersDECWET::CAPPELLOFMy other brain is a polymerThu May 29 1997 10:1752
>1. Should he regard the Cluster as 2 SQL standalone SQL servers
>    when working with SQL server management stuff like;
>
>Adding/Dropping Logins
>Configure SQL server parameters/tuning..
>Adding/changing objects on master and  other local databases...
>Working with backup devices..
>Working with tasks and alerts
>
    
    All of the above should be considered LOCAL operations.  You must do
    them on both servers in the cluster if you want similar information on
    both nodes.
    
>2. Work on the shared clustered enrolled database...
>
>Creating/changing objects
>tables/views/procedures/users/permissions;
    
    Not sure on those questions.  It depends on where the information is
    stored.  If the info is stored within the shared database, then you
    only have to do the operation ONCE.  If the information is actually
    stored in the server's master database, then you have to do the
    operation twice (once on each server.)
    We'll try to get a better answer for you.
>
>Changes in size of database and database device is
>documented, unenroll - change - enroll also in topic 800..
>Is that maybe a general rule, unenroll - change - enroll
>working on the clustered database?
    
    Maybe not.  It depends on what kind of information you're changing.
>
>What kind of SQL information is "replicated" between
>the SQL Servers when running in cluster, other than
>which databases are enrolled, primary server..
>
    
    This question is the key to understanding.
    
    When a database on a shared disk is "enrolled", the cluster software
    replicates information about the database location and size. This
    information is stored in spt_fallback tables, which are contained in a
    server's MASTER database.  Basically, the cluster software copies
    information from the "sysdatabases", "sysdevices", and "sysusages"
    system tables into the spt_fallback tables on both servers.  That is
    the only information that is replicated.
    
    SOME other information about the database is contained within the
    database itself (for example, the database schema).  That information
    doesn't need to be replicated because it fails over when the database
    fails over.
837.3"logins" is local, other things are sharedDECWET::CAPPELLOFMy other brain is a polymerThu May 29 1997 13:4925
>2. Work on the shared clustered enrolled database...
>
>Creating/changing objects
>tables/views/procedures/users/permissions;
    
    Ok, a better answer on these questions now:  The answer for everything
    except "users" is easy.  When you make changes to a shared database for
    tables/views/procedures, the changes failover properly because all the
    information is contained within the shared database itself.
    
    Information about users and permissions also fails over correctly IF
    THE LOGIN ENTRY MATCHES EXACTLY ON BOTH SERVERS.  For example, if I
    create a login entry on ServerA with a name of "cja" and uid = 11, I
    must also create a login entry on ServerB with the name "cja" and the
    same uid ( = 11).  Then I can add user "cja" to a shared database and
    assign db permissions within the shared database.  Permissions and
    users failover just fine.
    
    However, if login entry for "cja" on ServerA has a different uid than
    the login entry on ServerB, the user information does not failover
    properly.
    
    Bottom line: make sure the "syslogins" table in the master database
    looks the same on both servers.  Unfortunately, the cluster software
    doesn't help with that.