Time taken for failover

[abhijain1]

Hi,

I am new to VCS and am planning to use it for my COBOL based application. I want to have two servers in my prod environment Primary and Secondary. I want to support Automatic failover without disconnecting the users who access the COBOL application through putty. Could you please help me answer these questions:

(1) When I deploy code in Production, will I deploy only once on the virtual IP or I need to deploy on both Primary and Secondary separately?

(2) Once VCS is running and primary fails, will users get disconnected?

(3) Will the failover to the Secondary seemless or I still need to re-mount the storage, interfaces etc again on the secondary server?

(4) I dont want to use automatic script for failover as data which was in the buffer when server crashed is also critical. Hence if a DR type failover is required, I want to do it manually as I dont want to lose any data. Is it possible?

Many thanks,

A

[CliffordB]

Hi A.

I don't have any feedback for COBOL in particular as I could not find any documentation anywhere in regards to how to cluster COBOL (using any cluster software).   So, here is what I can tell you about clustering generic applications.

1) Deploy code in clusters: It is possible to deploy either on shared storage, or, locally.  I prefer installing locally as it allows rolling upgrades for applications that support such.  Typically:

• Code installed on each node.  If the install is cluster aware, use a virtual IP and/or virtual DNS name.  Never the hostname.
• Data files are configured to be placed on shared storage.  It might be necessary that you may have to fail the data storage to the second node in order to install the software/code on the second node.
• Data files are placed on a separate disk group.
• Zone the storage so all nodes can see the disks.
• Use fully qualified names or make sure that DNS ( or whatever name service you use) understands those names.
• Register/reserve the IP and DNS name in your name service.  You dont want someone else in your company accidently clobbering your IP address.

2) What happens to users on node failure?  This depends on your application.  If there is middelware and a web front-end, then the user normally stays connected or only has to retry the transaction.   If users are connected directly to the host, then they will be disconnected.  They will have to reconnect.  If configured properly, VCS moves the VIP/DNSname to the second host.  Users/applications would reconnect as normal as the IP address is moved.

3) Storage: It is normal to configure VCS to manage everything needed to make an application run, including importing disks, mounting disks, updating NICs with the floating VIP, starting any processes.   These confgiurations are detailed in the admin guide.

4) For local failover, most admins configure the failure between nodes as automatic, but this can be made maunal.   For wide-area (DR) type failover, by default, the failover is not automatic and must be initialized by an operator.  This prevents false positives situations such as loss of network/communications but the server is still running.

Regarding losing data when a hode fails:

• This is application dependent.  No magic here.  If there is a some type of transaction log where data is stored before commit to a database, then that log must also be on shared storage so that is available to the app after failover.  Some applications have the ability to update logs on multiple servers to prevent dataloss.  Very typical of middleware productds.
• For wide-are (DR): data loss is both application depenedent and also tied to whatever data mover one uses to replicate between datacenters.   VCS has several agents to support third-party replication as well as our own data mover, VVR.

Hope this helps.

Cheers