GaryCat

Hi

We are putting a system into a customer which runs on 8 weblogic servers and has 'sticky' sessions from ~800 IE users via 2 x Cisco CSS11501S-C-K9 load balancers.

The availability of the servers is checked with a heartbeat every 2 secs and three lost heartbeats will mark a server a being down.

However, if an http request arrives just after it's server has died, the CSS may not have marked it as dead and will still route the request to it.

What we need is for the CSS to recognise that the server is unavailable when it gets the TCP connection failure and to retry the request on an alternative server. The network admin guys say this is not possible on Cisco, but we have been able to do this on our own BigIP LBs

Can anyone help?

stevem2k

[caveat: not a css expert ]

why not chop the connection with

tcp-close fin , rather than the default rst

Avoids waiting for all the packets required for a reset, reduces the 'dying' time.

Also ... if it's that critical why wait 3x2 secs to mark it down ?


Steve

GaryCat

Whoa... that went so far over my head I can't even see it with a telescope.

The 3 x 2 seconds is to prevent the server being marked as offline when it has just been delayed by network issues or server performance etc. If the server is taken out of the load balance group, up to 200 sessions have to be replicated on to the other 3 servers so we need to be sure it's down.

stevem2k

CSS Command Reference (Software Version 7.40) - Keepalive Configuration Mode Commands [Cisco CSS 11500 Series Content Services Switches] - Cisco Systems

If you do sh keepalive you'll see the auto generated rule for the service, you can modify it from there.

Personally I'd use a non-persistant http connection to a weblogic 'alive' page in this case, as you can easily reach a situation where it can open tcp sockets to the machine but not reach the service.

Steve

GaryCat

We're currently using the CSS KeepAlive mechanism to detect a failure of a service and that is working OK.

Our problem relates to those requests that reach CSS from the time a service has failed until its failure is detected by the load balancer's KeepAlive: These requests expose the failure to the clients so the client does not see a 'seamless' failover.

Our application is a standard web application and is using HTTPS protocol. We are looking for a way to configure the load balancer to detect HTTP requests that failed (due to "connection failed" sockets exception) and to retry these requests on another service which hasn't failed.

We have been able to configure our own BigIP load balancer to retry failed HTTP requests, and thus provide seamless failover capability to the end users.

In our BigIp implementation, we catch the load balancer event indicating that the connection to the server/service failed (for BigIp it is LB_FAILED).
In BigIp script code, we try connecting to current server 10 times, and if we fail we change the relevant server/service status to bad (not available).
Then we call the load balancer's reselecting function (which selects other server from the cluster).

We had to use some propriety BigIp scripting in order to implement this.

Our question is how can we achieve the same functionality with the Cisco load balancer ?