Forum Discussion
Tim_a_willis_92
Nimbostratus
NimbostratusLoad balancing read only SQL 2008R2 servers
We have everything setup in the LTM and everything is working right. Traffic is getting balanced equally between our MS SQL server (read only). But the issue we are seeing happens when a server goes out for maintenance. We have a health monitor setup to query a DB and the LTM knows when the SQl server it down. But it seems the web servers connections to the SQL servers are keeping the connection alive.
Here is a copy of the error that is recieved: System.Data.SqlClient.SqlException: A transport-level error has occurred when sending the request to the server. (provider: TCP Provider,error: 0 - An existing connection was forcibly closed by the remote host.)
After this happens, the web servers then move over to the next available node.
the VIP is setup for performance layer 4, web servers are server 2003, and our SQL servers are server 2008R2 setup with n-path.
What_Lies_Bene1Cirrostratus
You probably want to modify the value of the Action on Service Down setting for the Pool. The default is none, if you change it to Reselect LTM will select another Pool Member and ‘rebind’ the client-side connection to a new Pool member.
If authentication is involved however I'd imagine things would still break. If you use persistence for that reason or some other reason, things will break, application wise.
Tim_a_willis_92action on service down is already set to reselect
- What_Lies_Bene1
OK. Is persistence used/required?
- Tim_a_willis_92
no it is not used
- What_Lies_Bene1
Interesting. Would this apply? http://support.f5.com/kb/en-us/solutions/public/8000/100/sol8160.html
Any iRules in play?
Any authentication involved?
- Tim_a_willis_92
No irules in place for this VIP and the LTM is version BIG-IP 11.3.0 Build 3022.0 Hotfix HF3
- What_Lies_Bene1
So, how about changing Action on Service Down to Reject; won't that speed up the recovery, even if it's not transparent?
- Tim_a_willis_92
We have tried that too already
- What_Lies_Bene1
Only just noticed you're using nPath. As the LB is only involved once to make the initial load balancing decision surely whatever you set it'll fail as the F5 isn't acting as a proxy here.
- Tim_a_willis_92
So is it possible that the npath is keeping the connection alive then?