September 20, 2006 at 1:04 am
We are evaluating mirroring vs clustering for a large database application and have some basic clustering questions....
When implementing failover clustering, must there be an even number of servers (2,4,6,8,etc) or can you configure say, 3?
Can Active/Active and Active/Passive clustering be "mixed"? Someone here has the idea of having 2 Active/Active servers mixed with a single server running in Passive mode. The reason is for performance and load balancing.
When coding an application, does the application point to the cluster (which chooses an instance) or does the application point directly to a particular instance? Or both? We're not clear on how this is handled.
In the doc explaining Failover Clustering, it talks about transferring an instance from the failed server to the secondary server. How is this done? Does the secondary server simply initialize a fresh copy of the failed instance?
On failure, what happens to the transactions on the failed server? Do the users get an error/warning message?
THANKS IN ADVANCE!
September 21, 2006 at 6:09 am
I cannot answer Active/Active questions since I have no experience with them but I don't think you can mix active/active with active/passive. There would be an IP conflict and a server name conflict I think.
Active/Passive: The users connect to a shared server name ONLY. They never connect to the active server running SQL Server because if there is a failover you would loose your connection. If there is a failover or a problem on the active server SQL Server is constantly checking via a is alive heartbeat between the boxes and if the active one does not respond within so many seconds SQL Server will start up on the passive node. The user community will loose their connection and will have to reconnect. As far as transactions SQL Server will do its best to roll it back or recover it based on the transaction. This happens just like if someone reboots a normal SQL Server in the middle of activity. Basically the disk(s) are shared on a SAN, not a hard drive specific to one or the other server so the disks are available to both nodes.
September 21, 2006 at 7:47 am
You may certainly set up a 3 node cluster, provided that you get the right OS and SQL versions.
You could use Active/Active/Passive so that the Passive node is a standby for the services running on either of the 2 active nodes.
As Markus states, the failover happens when the cluster service decides that a node running an instance of SQL is no longer functional. It approximately yanks the disks, IP addresses, and network names away from the naughty server, moves them to another node, and starts SQL server on that node.
I don't think that it is fair to say that SQL server "will do its best" to roll transactions forward or backwards. It will actually do exactly what it is supposed to do based on the ACID rules.
jg
Viewing 3 posts - 1 through 2 (of 2 total)
You must be logged in to reply to this topic. Login to reply