Configuration:
1) Windows 2003 (same result on Web, Standard, and Enterprise Editions), 2
servers
2) IIS 6
3) ColdFusion 7 Enterprise
4) Microsoft NLB
5) Each server has a single instance of the web site and the remote
instances are
registered (in Instance Manager)
Everything works as expected while both nodes are up. The load is distributed
as expected and session information is replicated correctly (very little
session information is involved). Testing went very well (better than
expected)...until we shutdown a node. Whether we do an actual server shutdown
or disable the nic the result is the same.
When only one node is operating that node still tries to balance the load to
the non-operating server and we get timeout issues. After a while (several
minutes) it will no longer try to send anything to the other server and things
go quite well. Then it will try again after a longer period of time.
This is not ideal. Should a server in the cluster fail our customers will
continue to experience issues for what seems to be a quite extensive time.
This would be acceptable if we could tell the active node to no longer try to
use the failed node such as using the STOP action in CF Administrator's
Instance Manager. However, when a node is down neither CF Administrator nor
JRun Administrator on the active node can 'Stop' the failed node (the action is
not allowed so it keeps trying to talk to it for a while). The only solution
seems to be to delete the failed node from the CF cluster.
I had hoped this would work more like NLB does, where no traffic would be sent
to the down node once it is known to be down...until it joins the cluster
again.
Our monitoring systems can rapidly detect a failed (or failing) node and take
action such as updating the NLB cluster quicker than it will detect it on its
own. I'd like to be able to do the same with CF7 instances as soon as the
failure is detected. But removing a server is a beyond the type of action I
like to take...
Is there a configuration setting that will modify this behavior or do we just
have to deal with it?
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~|
Get involved in the latest ColdFusion discussions, product
development sharing, and articles on the Adobe Labs wiki.
http://labs/adobe.com/wiki/index.php/ColdFusion_8
Archive:
http://www.houseoffusion.com/groups/CF-Server/message.cfm/messageid:6408
Subscription: http://www.houseoffusion.com/groups/CF-Server/subscribe.cfm
Unsubscribe:
http://www.houseoffusion.com/cf_lists/unsubscribe.cfm?user=11502.10531.10
