what happens if the service is down ??

[rbatchu]

HI

we have 3 web servers load balanced using a CSS.

The content which get to CSS is balanced among these three services.

Suppose a request came in from Customer A and css redireicts that to say service#1 and the process is going on and Customer A is is waiting for the output from service#1.

meanwhile if service#1 dies. what happens to the process??

Will the Customer A gets the output or it just hangs ??

will that process is redirected to another sercvice.

[Gilles Dufour]

when the service goes down, the flows associated with this service are removed. So, next packet from client will be RESETed by the CSS so the client

should/will reopen a connection and will be loadbalanced to another server.

Gilles.

[drussell]

Is this a recent change?

There is a flow-reset-reject command under the rule to send a TCP reset if the address is unreachable - what is meant by unreachable is not defined. Is it the lack of a routing entry or an invalid ARP entry.

I had understood that the CSS will not take down sessions to services which have failed or are suspended. Letting any existing sessions complete after suspending a service is a gracefull way to take a server down.

[Gilles Dufour]

I'm glad you corrected me.

I gave an answer based on something I read and apparently it was erroneous.

I went into the lab and confirmed what you mentioned.

The CSS will not touch existing flows when a service goes down or when you suspend it manually.

The flow-reset-reject command is described at

http://www.cisco.com/en/US/products/hw/contnetw/ps789/prod_release_note09186a00800ba0c6.html

I'm still trying to clarify this part.

Gilles.

[Gilles Dufour]

ok - here you go for the flow-reset-reject.

It only works with layer 5 rule (url command) and it does send

a reset to active connection when the service goes down.

From internal doc and verified with lab test.

Gilles

[rhholmes]

Would this be applicable in the following scenario?

The content rule is a layer 5 rule (advanced-balance ssl, application ssl) and there are three services. If one of the services is suspended, and the content rule has the flow-reset-reject command configured, would all clients receive a RST?  This would terminate their flow and force a new connection to one of the services that is still active. 

Basically, looking for a means to override the graceful shutdown because it takes too long.

Thanks,

Rob