Re: NLB Cluster, service restart and persistant socket connection

I would say a .Net error then, because the code is bullet proof, it's just
the callback that never returns (no exception thrown, nothing). And it is the
same for the DB connection (we are using the Microsoft library).
I guess we're going to have to open a ticket with Microsoft.
Thanks for the help.

"Rodney R. Fournier [MVP]" wrote:

Sounds like an application error then.

Cheers,

Rodney R. Fournier

MVP - Windows Server - Clustering
http://www.nw-america.com - Clustering Website
http://www.msmvps.com/clustering - Blog
http://www.clusterhelp.com - Cluster Training
ClusterHelp.com is a Microsoft Certified Gold Partner


"lafeedesbois" <lafeedesbois@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message
news:CC7A3908-BD81-4F2A-825F-8CA3B5BAF6F7@xxxxxxxxxxxxxxxx
Hello,

The NIC are gigabits and aren't overloaded. The problem only arises when
the
application is restarted (after an upgrade or a MS update). It acts as if
something is stuck at the network level that prevents the application from
opening a persistent socket client to another server. And the only way to
"remove" what's "stuck" is to first exclude the server from NLB and then
reboot, anything else doesn't work. And we're not experiencing the problem
when the server is not part of the NLB cluster.On the other hand, the
incoming socket connections (which are managed by the NLB) are coming in
fine.
Thanks for your help

"Rodney R. Fournier [MVP]" wrote:

Is the NIC overloaded? Are you sure the application can handle the load?
What speed are the NICs?

Cheers,

Rodney R. Fournier

MVP - Windows Server - Clustering
http://www.nw-america.com - Clustering Website
http://www.msmvps.com/clustering - Blog
http://www.clusterhelp.com - Cluster Training
ClusterHelp.com is a Microsoft Certified Gold Partner


"lafeedesbois" <lafeedesbois@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message
news:3E238140-B9EC-4EED-BD85-C3940B37DA8E@xxxxxxxxxxxxxxxx
Hello,
We're having problems with a very specific configuration and I can't
find
anything to help us:
We're using NLB between two 2003 servers that run a service. The
service
starts an application which opens a socket server for incoming socket
connections. Those connections are regulated by NLB and work fine.
However
the application also start a couple of persistant socket connections
and
of
course a DB connection (to MS Sql Server cluster). Now that we are
running
at
high volume (Over 100 TPS), when we restart the service, the persistant
socket connections are failing randomly (both socket client and DB
connection). The client is stuck on the connect callback and never
returns.
In order to solve the problem, we have to exclude the node from NLB and
then
reboot the server. Doing one without the other doesn't work.
If anyone has any idea as to why it fails, or anything we could try to
prevent having to exclude and then reboot, it would be great.
Thanks.






.

Relevant Pages

  • Establishing a p2p connection in python
    ... programming PHP on a webdeveloper basis. ... I thought that one approach could be to write a small server for my ... users and establishing p2p connections on demand. ... It would seem like the socket ...
    (comp.lang.python)
  • Re: receiving data over socket
    ... I want to print the data the server received. ... anyone can recommend a good VB2005 book that deals with socket programming I ... if your server needs to support multiple connections ... You aren't guarenteed to recieve all your data ...
    (microsoft.public.dotnet.languages.vb)
  • Re: MSDN Using Asynchronous Server Socket ?
    ... I properly shutdown the socket after receiving the message. ... The problem is in your server code that handles the active connection. ... framework doesn't handle a lot of "idle" connections very well. ... client disconnects, do you shutdown and then close the client socket? ...
    (microsoft.public.dotnet.framework)
  • Re: When do client connections disconnect
    ... Server) it describes the issue that I am having with my application. ... "Rodney R. Fournier" wrote: ... MVP - Windows Server - Clustering ... to completly shutdown and clean up socket connections. ...
    (microsoft.public.windows.server.clustering)
  • Socket connections in TIME_WAIT and apache MaxClients setting
    ... of concurrent connections to the web server? ... BigIP were not responding to connections. ... I noted that a wc -l of socket ...
    (SunManagers)