2 Node Cluster - how fast before detection and package enabled on 2nd node ?

[Stephen_23]

All,

I have a 2 node cluster. We are doing some tests to determine how fast the second node will detect a power off of the primary node
i.e. Just press the power off button to simulate a catastrofic failure.

I am after your time frames on this issue
i.e. how long does it take to determine that the primary has failed and SGuard descides to start the first package on the 2nd node.

We are getting around 30 seconds for the first package but I have no point of reference for this time wether it is good or bad.

Any information appretiated,
Cheers,
Stephen

[Artyom Voronchihin]

Hello

Nodes in a MC/SG cluster send heartbeat messages to each other to indicate they are still alive. If heartbeat messages are not recieved within a prescribed amount of time, the cluster will try to reform minus the node not sending heartbeat messages. To change the time interval edit HEARTBEAT_INTERVAL parameter in the cluster ascii configuration file. Also when cluster is reformed, a package needs some time to activate VGs and start his services. I think 30 seconds to reform is OK. In our configuration Oracle package starts about 1 minute.

[melvyn burnard]

This wil all be determined by the settings you have for hte Heartbeat Interval, and Node Timeout interval.
If you have those settings to hand, post them and I can give you an idea of the sort of timing it should take.

[Vincent Farrugia]

Hello,

30 seconds is good. For Service/Guard to be functioning correctly, this number should be less than 1 minute maximum. That is a guarantee from SG.

This, of course, does not take into account, package startup etc.

HTH,
Vince

[Stephen_23]

>>> This wil all be determined by the settings you have for hte Heartbeat Interval, and Node Timeout interval.

Heartbeat - 1sec
Node Timeout = 2sec
Network Poll Interval = 2sec
Auto Start TImeout = 5 minutes

Cheers,
Stephen

[Fred.Wu]

From HP document ,the recommended value of HeartBeat Internal is around 6-10seconds.But I think it's up to your applications.

[melvyn burnard]

Ok, from these values
Heartbeat - 1sec
Node Timeout = 2sec
You should indeed be looking at around hte 30 sec mark, as they are minimal, default settings.
We generally recommend increasing hte node timeout to around 4-8 seconds, but then this will increase the length of time before a failure is detected.

[Stephen_23]

> Ok, from these values
> Heartbeat - 1sec
> Node Timeout = 2sec
> You should indeed be looking at around hte
> 30 sec mark, as they are minimal, default
> settings.

Could you explain to me how the 30 seconds is worked out using these numbers i.e. is there some kind of formula ?

AND

What happens exactly when one node fails - what is the process from the daemons ?

Cheers,
Stephen