| Author |
Prohibit new sessions on a cluster-member
|
|
|
| I'd like to prohibit new sessions on one of 2 AppServer cluster-members.
The idea behind is that I'd like to stop one AppServer without killing the sessions of those users that are currently using that server. (Distributed Sessions should not be used).
If one would set the loadweight for the "soon dying" AppServer to 0 (while keeping the the others >0 ) in the webserver's plugin, I'd expect to have the above described effect. Currently existing sessions should then be continued on the soon dying AppServ
er while new sessions are only made and forwarded to the AppServer which stays up - until all old sessions on the doomed server end and it can be stopped.
Could you please confirm or counter this idea ?!
Thanks in advance.
| |
| Brian S Paskin 2007-07-24, 7:21 pm |
| Hi, it sounds like a good idea, but I am not sure that the requests for the particular server that the LoadBalanceWeight has been set to 0 will actually go back to that server even with session affinity. Even if it does, how do you know that all the user
s are done their processing?
http://publib.boulder.ibm.com/infoc..._plugincfg.html
Brian
| |
| Paul Ilechko 2007-07-24, 7:21 pm |
| marco_simon@web.de wrote:
> I'd like to prohibit new sessions on one of 2 AppServer
> cluster-members. The idea behind is that I'd like to stop one
> AppServer without killing the sessions of those users that are
> currently using that server. (Distributed Sessions should not be
> used).
>
> If one would set the loadweight for the "soon dying" AppServer to 0
> (while keeping the the others >0 ) in the webserver's plugin, I'd
> expect to have the above described effect. Currently existing
> sessions should then be continued on the soon dying AppServer while
> new sessions are only made and forwarded to the AppServer which stays
> up - until all old sessions on the doomed server end and it can be
> stopped.
>
> Could you please confirm or counter this idea ?!
Setting it to 0 should prevent any more request going to that server,
including existing sessions which will attempt to migrate to another
server in the cluster. If you have no session persistence, you're going
to see errors.
| |
|
|
|
| > Setting it to 0 should prevent any more request going
> to that server,
> including existing sessions which will attempt to
> migrate to another
> server in the cluster.
That seems not to be correct.
The load-balancing mechanism has an higher priority on session-affinity than on equal-balancing. Whenever an internal routing-counter reaches 0, no further _new_ sessions or not-sticky requests are routed to that appserver anyfurther. Existing sessions ar
e still routed to the AppServer where they've been created. This makes it possible that the internal routing-counter of the plugin-module becomes negative.
Source of that information:
http://www-1.ibm.com/support/docvie...uid=swg21219567
|
|
|
|