Re: Meanings of thresholds in the cluster administration snap-in

Please, can you say something more about the "RetryPeriodOnFailure".
How vcan I change enable it in 2003 EE ? Is it a registry key ?

As far as I know, there is no way to delay these restart attempts in
cluster.

I asked about it and threshold because I'm plagued by a problem with a
service. In our server architecture under particular circumstances it cannot
go online for few minutes for acceptable reasons. Unfortunately the cluster
tries to restart it at the speed of light (or at least it is what I deduced
from watching event viewer) so the threshold is reached no matter how high I
can set it. If only I could delay these restart attempts I could avoid the
few minutes backout of our architecture.
By the way the problem is originated by SQL server running under the domain
account 'cluster' so if the dc cannot authenticate him the service cannot go
online. The unavailability of DC is only temporarily but the cluster is
faster to reach that damn threshold. After that the cluster may be healty but
it will not try to restart or failover longer.

Thanks
Filippo

"John Toner [MVP]" wrote:

If your SQL server resource fails to start on a node, MSCS will immediately
attempt to restart that resource x number of times until it hits the
threshold values. Once the threshold value is reached, the resource will
either cause the whole group to move to another node (default behavior) or
it will fail if all thresholds are hit. There is a Group level threshold
also in place so that a failing resources in a cluster does not cause the
group to "ping pong" eternally between the nodes.

Once all thresholds are hit (resource and group level thresholds) the
resource will fail and remain in a failed state until you try to restart the
resource again. You can have cluster periodically attempt to restart the
resource by setting a "RetryPeriodOnFailure" value...by default, this is
disabled in 2003 (set to 60 minutes in 2008).

As far as I know, there is no way to delay these restart attempts in
cluster. Service control manager settings do not apply to the cluster...it
will attempt to restart immediately.

I hope this helps.

Regards,
John

Visit my blog: http://msmvps.com/blogs/jtoner


"setecastronomy" <setecastronomy@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message
news:A8EB6359-F6D8-4191-B422-B35680A4FEE8@xxxxxxxxxxxxxxxx
I have SQL server 2000 running on cluster realized using 2003 EE. The
cluster
is active/passive, SQL is installed on both nodes while its data file
resides
on a NAS.
It works.
Using cluster administration snap-in you have the possibility, for each
group of resources, to set a threshold and a period for failover.
As far as I know it means the cluster will try to bring online the group
on
the other node only if the the number of failures is less than the
threshold. If, for some reasons, it reaches the threshold, it gives up and
doesn't try any more. It may be a smart choice becuase it is useless to
try
indefinitely to bring online resources that fail permanently due to
serious
reasons, with the risk to compromise and overload the entire cluster.
For each resource inside a group, in the advance tab, there is the
possibility to set "Restart" behaviour with an associated threshold and
period.
I think the meaning is that the node will try to restart the failed
resource
till it reaches the threshold.
What it is not clear to me is how these two threshold are related and when
they cause the switch on the other node.
Focusing on my SQL server, what happens if it fails to start on a node ? I
suppose the node try to restart it till it reaches the threshold fixed for
that resource and only then a failover is initiated on that group. Is it
correct ?
The most important point I'm concerned with is that I would like to fix
the
interval of time between a failure and a restart attempt. Windows let me
configure it using the service.msc but I doubt it is the right way to do
operate on clusters, where you should use the cluster administration
snap-in.
I searched on Microsoft's site a lot but I found only a brief description
of
all these parameters and no one complex scenario where they all interact.
Most documents only say that in case of failure the resource is moved to
the
other node, but how this fact is influenced by threshold and restart
policies
is not analized in depth.
Can you give me a hand ?



.

Relevant Pages

  • Re: ressources unavailables after switch maintenance
    ... Unfortunately this is a common issue in Exchange 2003 clusters. ... This results in each resource executing a separate timeout period in series. ... service be able to successfully stop Exchange running on the cluster? ... This threshold is set at the group level. ...
    (microsoft.public.windows.server.clustering)
  • Re: Cluster says that service is up but it isnt
    ... cluster failed the resource and then restarted it. ... cluster did recognized that the service was died and restart it. ... OpNotes>: Failed the IsAlive test. ...
    (microsoft.public.windows.server.clustering)
  • Re: Automatically Restart Failed Cluster Resource
    ... Probably because the network problems are severe enough that you have exceeded the Restart Threshold for the resource and the threshold does not get reset until the cluster service is stopped and restarted. ...
    (microsoft.public.windows.server.clustering)
  • Re: what happens when the public connection is lost?
    ... What happened is an underlying resource failed so SQL tried to find a node where the resource existed. ... IF SQL fails to start or restart a given number of times within a certain time period (both adjustable using the cluster tool), the clustered resource will be considered unrecoverable and will require manual restarting. ...
    (microsoft.public.sqlserver.clustering)
  • Re: Meanings of thresholds in the cluster administration snap-in
    ... If your SQL server resource fails to start on a node, ... Once the threshold value is reached, ... resource will fail and remain in a failed state until you try to restart the ... You can have cluster periodically attempt to restart the ...
    (microsoft.public.windows.server.clustering)