10:41 p.m.
When I disable a worker (or all workers in a given domain) in
mod_cluster_manager, requests to *existing* sessions get rerouted to
other workers.
This is IMO *not* what DISABLE should do (it is what STOP used to do in
/status) ! DISABLE should *not* remove a worker from the worker tables,
but have the following behavior:
* Sticky sessions
o Requests for existing sessions: route them to the existing
workers to which they were tied (also disabled ones !)
o New sessions are *not* created on disabled workers, but only
on enabled ones
* Non-sticky sessions
o Route requests to any enabled worker
o No requests are sent to disabled workers
WDYT ?
--
Bela Ban
Lead JGroups / Clustering Team
JBoss
1:57 a.m.
On 08/13/2009 05:41 AM, Bela Ban wrote:
When I disable a worker (or all workers in a given domain) in
mod_cluster_manager, requests to *existing* sessions get rerouted to
other workers.
This is IMO *not* what DISABLE should do (it is what STOP used to do in
/status) ! DISABLE should *not* remove a worker from the worker tables,
but have the following behavior:
* Sticky sessions
o Requests for existing sessions: route them to the existing
workers to which they were tied (also disabled ones !)
o New sessions are *not* created on disabled workers, but only
on enabled ones
* Non-sticky sessions
o Route requests to any enabled worker
o No requests are sent to disabled workers
WDYT ?
I though it was working... I will add a test for that.
Cheers
Jean-Frederic
2 a.m.
A simple test is to create a sticky session, say to node2, and then
disable node2. The next request will go to node1, which is not correct !
If you can fix this by JBW, that'd be great (I'd like to show this) !
jean-frederic clere wrote:
On 08/13/2009 05:41 AM, Bela Ban wrote:
> When I disable a worker (or all workers in a given domain) in
> mod_cluster_manager, requests to *existing* sessions get rerouted to
> other workers.
>
> This is IMO *not* what DISABLE should do (it is what STOP used to do in
> /status) ! DISABLE should *not* remove a worker from the worker tables,
> but have the following behavior:
>
> * Sticky sessions
> o Requests for existing sessions: route them to the existing
> workers to which they were tied (also disabled ones !)
> o New sessions are *not* created on disabled workers, but
> only
> on enabled ones
> * Non-sticky sessions
> o Route requests to any enabled worker
> o No requests are sent to disabled workers
>
> WDYT ?
I though it was working... I will add a test for that.
Cheers
Jean-Frederic
--
Bela Ban
Lead JGroups / Clustering Team
JBoss
2:05 a.m.
On 08/13/2009 09:00 AM, Bela Ban wrote:
A simple test is to create a sticky session, say to node2, and then
disable node2. The next request will go to node1, which is not correct !
If you can fix this by JBW, that'd be great (I'd like to show this) !
I will write automated tests for the mod_cluster_manager features.
Cheers
Jean-Frederic
5618
days inactive
5618
days old
mod_cluster-dev@lists.jboss.org
4 comments
2 participants
participants (2)
-
Bela Ban -
jean-frederic clere