Hi,

Taking your questions one at a time:

1. Architecture feasible?

Yes, it looks good. 

Both the server farm and failover options should work. I understand that in 
each data center there are 4 AR servers in active mode (ie. working as a server 
group)? It wasn't entirely clear to me from your text.

2. How to detect server failure?

I assume you mean one of the servers in the active server farm as the failover 
detection is quite straight-forward. 

To detect that each server in the group is alive you should follow Joe 
DeSouza's advice and use the load balancer's port check mechanism. I am a fan 
of a simple API call as it is a more complete check, but for simplicity and 
agility the port check is superior.

As an aside, I recall previous versions of AR hanging temporarily when a telnet 
was launched against the TCP port of the AR server (the RPC protocol stack was 
intollerant to incomplete transactions). I assume this bug is history and that 
the load balancer's port check doesn't trip up the RPC protocol!

As a further observation I wonder if the status of the servers in the group is 
held in a form/table on the database... would make sense as part of the 
AR_System_Server_Group_Operation_Ranking workflow IMHO. Perhaps someone knows 
the answer?

3. Feeding Admin changes from passive to active?

This is not a good idea.

Your active database is changing all the time and if you decouple the dataguard 
syncronizing mechanism (which is at the datebase or file system level) they 
will diverge. The admin changes you then make in the passive server cannot be 
applied to the active server at the database level without running huge risks 
of messing up the whole database. I am not saying it is impossible (admin 
changes tend to me adding columns, changing workflow objects etc.), just that 
the chances of it going horribly wrong are considerable and you wouldn't get 
any support from BMC.

Alternatives I have seen work include resyncing at an AR level all changed data 
in the active server to the passive server with it's admin changes. If the 
Admin changes and regression tests take say an hour to apply then you need to 
export and import all changes in the last hour. Then you repeat the operation 
for the changes in the time it took to do the last export/import cycle. 
Eventually after a few cycles you will be down to a few changes to a few forms 
only and with a very short outage window you can get the passive server fully 
synced. You activate the passive server that has the admin changes and all the 
data up to date, and off you go! It is not zero-downtime, but it is minimal 
downtime (perhaps a few minutes) and it is supported by BMC!

Complications: Importing data will trigger workflow sensitive to Merge events. 
You need to review this for all forms that are regularly updated. 

Tools like Migrator could help here obviously.


4. Stickiness in load balancer?

If you followed David's advice you will already have your answer.


5. Experience with F5 LB?

Not me.


6. Other critical things?

Load balancers for the mid tier servers need to be part of the solution. Mid 
tier switching to other AR servers requires some care and potentially mutliple 
LBs.

Regularly switch between the active and passive data centers to make sure there 
the failover works. I have customers that run for two months in one center and 
then two months in the other. You  don't want to do this for the first time 
when somemthing breaks!

Watch out for other single points of failure. You are investing a lot in this 
so consider redundant network paths to the database etc. too.


Cheers

Mark



--- On Thu, 22/1/09, Easter, David <[email protected]> wrote:

> From: Easter, David <[email protected]>
> Subject: Re: Issues::Remedy in Load balanced environment
> To: [email protected]
> Date: Thursday, 22 January, 2009, 3:46 PM
> Ankur,
>  
>   A good starting point would be the white paper on using a
> Load
> Balancer within an AR System environment:
>  
> 06-Sep-2007           Using a Hardware Load Balancer with AR System
> 7.1.00 PDF
> <http://www.bmc.com/supportu/documents/40/92/84092/84092.pdf>
> 
>  
> Thanks,
>  
> -David J. Easter
> Sr. Product Manager, Solution Strategy and Development
> BMC Software, Inc.
>  
> The opinions, statements, and/or suggested courses of
> action expressed
> in this E-mail do not necessarily reflect those of BMC
> Software, Inc.
> My voluntary participation in this forum is not intended to
> convey a
> role as a spokesperson, liaison or public relations
> representative for
> BMC Software, Inc.
>  
> ________________________________
> 
> From: Action Request System discussion list(ARSList)
> [mailto:[email protected]] On Behalf Of Ankur Gulati
> Sent: Thursday, January 22, 2009 6:35 AM
> To: [email protected]
> Subject: Issues::Remedy in Load balanced environment
> 
> 
> ** 
> Hi Listers, 
>  
> I've been assigned a task of developing a load balanced
> environment for
> our remedy infrastructure. It has become a headache now
> since I am stuck
> with few things.
>  
> Here's bit of description of the architecture.
>  
> We have planned for having two sites A and B, A will be
> Primary and B a
> Disaster Recovery Site.
> On the top we have Users connecting through Citrix to any
> of the ARS
> application servers through two F5 1600 Load Balancer. The
> Load
> Balancers are in  active - passive mode with a switch above
> them to
> shift the connection incase any of the LB goes down.
> Then we have 4 ARS application server all in Active mode in
> server group
> architecture. These servers have two Database to connect
> to. Databases
> are in Active - Passive mode. Each of these DBs are
> connected to their
> own SAN for storage. We will have a dataguard for
> replication of the
> database.
>  
> DSO is not proposed in this architecture. We have a home
> grown
> application.
> ARS version 7.1.0 on Solaris 10
> Oracle 10g on RHEL 5
> Midtier 7.1.0 on Windows 2003 server
>  
> Questions that concern us are:
>  
> 1. Will this architecture be feasible solution for failover
> and high
> availability?
>  
> 2. How will we find out incase any of the server goes down?
> Is there any
> need for health check script to run on each of the ARS
> Server box to see
> if the server is alive or not?
>  
> 3. To have a zero downtime at the time of application
> upgrade, if we
> take one side completely down then is there an option in
> Load Balancer
> to restrict the users to access the admin server so that we
> can use it
> to do our workflow updates in the passive database. Then
> the dataguard
> will copy the changes from this database to the active. Is
> this scenario
> possible? if not then is there any other option where I can
> achieve zero
> downtime or minimal downtime for workflow and form updates.
> 
>  
> 4. Is there some recommendation for setting the stickiness
> value in the
> load balancer. 
>  
> 5. Has someone used this kind of scenario with F5 Load
> balancer and can
> share how things are configured? Remedy only has a
> documentation with
> Cisco LB.
>  
> 6. Are there any critical things you'd like to list
> which needs to be
> taken care of in this environment.
>  
> Any thoughts are appreciated.....!
>  
> --
> Ankur
> __Platinum Sponsor: RMI Solutions ARSlist: "Where the
> Answers Are"
> html___ 
> 
> _______________________________________________________________________________
> UNSUBSCRIBE or access ARSlist Archives at www.arslist.org
> Platinum Sponsor: RMI Solutions ARSlist: "Where the
> Answers Are"




_______________________________________________________________________________
UNSUBSCRIBE or access ARSlist Archives at www.arslist.org
Platinum Sponsor: RMI Solutions ARSlist: "Where the Answers Are"

Reply via email to