Operating System - HP-UX
1834008 Members
1820 Online
110063 Solutions
New Discussion

SG lan disconnection tests

 
Roberto Volsa
Frequent Advisor

SG lan disconnection tests

Hello all,
simple question about HA test on a cluster.

If i disconnect all lan cards from a node (say node A), which node will be tocked by SG? The A node or the other one?
SG release 11.15, HPUX v2 on Itanium.

BR
Roberto
10 REPLIES 10
G. Vrijhoeven
Honored Contributor

Re: SG lan disconnection tests

Hi Roberto,

If you have a two node cluster and there is no way they can communitcate and you have a cluster lock disk configured the last one to access the cluster lock disk will be tocked.
I do not know a way to predict which node that will be.

Regards,

Gideon
RAC_1
Honored Contributor

Re: SG lan disconnection tests

A node will TOC (if you have service fail fast and node/package fail fast enabled.
Package will be switched to another node and primary node will TOC.

Anil
There is no substitute to HARDWORK
Roberto Volsa
Frequent Advisor

Re: SG lan disconnection tests

Hi Gideon,
i know that the node tocked is the last in taking the cluster lock disk.

Is this true also on the lasts SG releases?

thanks
Roberto
Kent Ostby
Honored Contributor

Re: SG lan disconnection tests

And to clarify G. Vrijhoeven 's comment, there is no way to assign one of the machines to be the one that survives.

It is simply a matter of who can get to the lock disk the "fastest".

This document discusses lock disks in detail:

http://www6.itrc.hp.com/service/cki/search.do?category=c0&mode=id&searchString=UXSGKBAN00000282&searchCrit=allwords&docType=EngineerNotes&search.x=15&search.y=6


You can also use a quorem server.

Differences between the two can be found here:
http://www6.itrc.hp.com/service/cki/search.do?category=c0&mode=id&searchString=UMCSGKBRC00012642&searchCrit=allwords&docType=EngineerNotes&search.x=17&search.y=10

This document gives details on migrating from lock disks to quorem servers:

http://www6.itrc.hp.com/service/cki/search.do?category=c0&mode=id&searchString=UMCSGKBRC00014227&searchCrit=allwords&docType=EngineerNotes&search.x=23&search.y=8

Best regards,

Kent M. Ostby
"Well, actually, she is a rocket scientist" -- Steve Martin in "Roxanne"
Sanjay_6
Honored Contributor

Re: SG lan disconnection tests

Whichever node is able to take control of the lock disk will stay active as a cluster. The other node will go down with a TOC.

The cluster lock disk will act as a tie breaker over here (a two node cluster with both nodes loosing HB connection to each other).

Hope this helps.

Regds
Sundar_7
Honored Contributor

Re: SG lan disconnection tests

Hi Roberto,

I happened to test the exactly same thing couple of weeks ago.

If you disconnect all the lan cards from one node, now the nodes have no means of communicating with each other and each will assume the other one is having problems.

So,the node that can acquire the cluster lock disk the fastest will form the cluster. In the other node, the safety timer will expire and the node will be TOCed.

I was under the impression, SG will take in to account that the nodeB has got all of its network subnets up. But I learned from the testing it is not the case.

-- Sundar.
Learn What to do ,How to do and more importantly When to do ?
melvyn burnard
Honored Contributor

Re: SG lan disconnection tests

simply put, the node getting the cluster lock disk first will stay up.
This may not be the correct node though.
The serial heartbeat link was developed to try to ensure the node that had network connectivity actually did stay up, but this is for two node clusters only, and to be honest is more trouble than it is worth.
It is far better to migrate to the Quorum Server product, as the node has to be able to conatact the Qs via the net.

It may be worth mentioning here that what you are actually testing is not a SPOF but a MPOF, or Multiple Point of Failure, which was not what SG was designed to cater for.

My house is the bank's, my money the wife's, But my opinions belong to me, not HP!
Claudio Gutiérrez
New Member

Re: SG lan disconnection tests

Kent
in some countries we have support contracts with HP but we aren't entitled to have access to support documents like those you are referencing.

Is it possible that you or other forum's member post the docs related to lock disks and quorum server that you referenced.

Regards
melvyn burnard
Honored Contributor

Re: SG lan disconnection tests

Claudio

Unfortunately, posting of the relevant ITRC documents is unfortunately not allowed, hence the links supplied by Kent

My house is the bank's, my money the wife's, But my opinions belong to me, not HP!