Discussion:
Beginner Clustering Question - Application State
(too old to reply)
NvrBst
2010-08-04 01:23:19 UTC
Permalink
I've been wondering what custom service/app requirements are needed to
create an Active/Active Windows 2008 cluster? I've read that only the
shared resources (files) matter but I don't understand if application
state is transfered over to new nodes when a failure occurs, if this
problem means not all custom application are cluster'able, or if it
only affects certain cluster types?

My custom service/app has clients connect, do some stuff, disconnect.
"Do some stuff" depends on what the client has done since they
connect. After a disconnect all information about the client is
cleared.

From the windows 2008 features list I see two cluster types "Network
Load Balance" and "Failover Cluster".

NLB Cluster: I understand this type as it specifically says any
active connections are lost if a failure occurs, which makes sense to
me. So my custom service/application would work here.

Failover Cluster: Active/Passive: Would this work for my custom app/
service? If active connections are not lost then application state
has to be transfered to the new node during failure?

Failover Cluster: Active/Active: I don't see how application state can
be merged between two nodes during a failure which makes me think
state is not transfered. But wouldn't most applications run into this
situation (how else could active connections be maintained/transfered
unless every request is independent?

So basically I'm wondering if the custom application/service needs to
be build with clustering in mind to work with a windows failover
cluster, in relation to application state; or does the cluster service
take care of this for me somehow? I need to setup a shared database
to store client states so I can treat each request independently?
NvrBst
2010-08-04 01:30:57 UTC
Permalink
Post by NvrBst
I've been wondering what custom service/app requirements are needed to
create an Active/Active Windows 2008 cluster?  I've read that only the
shared resources (files) matter but I don't understand if application
state is transfered over to new nodes when a failure occurs, if this
problem means not all custom application are cluster'able, or if it
only affects certain cluster types?
My custom service/app has clients connect, do some stuff, disconnect.
"Do some stuff" depends on what the client has done since they
connect.  After a disconnect all information about the client is
cleared.
From the windows 2008 features list I see two cluster types "Network
Load Balance" and "Failover Cluster".
NLB Cluster:  I understand this type as it specifically says any
active connections are lost if a failure occurs, which makes sense to
me.  So my custom service/application would work here.
Failover Cluster: Active/Passive: Would this work for my custom app/
service?  If active connections are not lost then application state
has to be transfered to the new node during failure?
Failover Cluster: Active/Active: I don't see how application state can
be merged between two nodes during a failure which makes me think
state is not transfered.  But wouldn't most applications run into this
situation (how else could active connections be maintained/transfered
unless every request is independent?
So basically I'm wondering if the custom application/service needs to
be build with clustering in mind to work with a windows failover
cluster, in relation to application state; or does the cluster service
take care of this for me somehow?  I need to setup a shared database
to store client states so I can treat each request independently?
Sorry, also how does a Network Load Balance Cluster differ from a
Active/Active Failover Cluster? From what I understand the only
difference is how active connections are treated during failure? Aka
NLB will terminate the socket connection while Active/Active will re-
route it to a new node. And Active/Active requires application to
handle application state problems?

Loading...