In addition to Linchi's remarks, you should know that the cluster's initial
response is to restart the service *on the same node*. If successful, you're
fine. If not, it will try again. If this happens more than 3 times (by
default), it will fail to the other node.

--
Tom

----------------------------------------------------
Thomas A. Moreau, BSc, PhD, MCSE, MCDBA
SQL Server MVP
Toronto, ON Canada

[quoted text, click to view]
Out of box and by default, the SQL Server resource in a failover cluster is
configured to do the LooksAlive check once every 5 seconds to see whether
the
service is running. Note that this is a rather lightweight check and even if
this checks out to be okay, the SQL instance may still be unusable. So the
cluster also does what is called the IsAlive check, which actually tries to
connect to the instance and run a simple query. Depending on the failover
threshold setting of the SQL Server resource, failure of this check and its
retries may cause the cluster either to try to restart the SQL instance on
the same node or fail it over to another node, i.e. start it on a different
node.

So CPU pegged 100% in itself will not necessarily cause a failover. However,
if it causes the IsAlive check to fail and the number of failures meets the
configured failover threshold, the SQL Server resource group will failover.

Same applies to running out of worker threads.

Linchi

[quoted text, click to view]