You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@tomcat.apache.org by Mladen Turk <mt...@apache.org> on 2005/02/16 10:01:15 UTC
JK - need some testings for new lb
Hi,
As said before load balancer does not have any more all
that fuzzy local_worker settings.
For most people (at least for me) that was totally unusable
and hard to understand. I hope the new settings and algo
will be more clear and much usable in conjunction with
shared memory and jk_status.
There are two major concepts:
Sessions and Domains (well, not quite new).
If there is a session that matches the worker name,
and the worker is not in error state, then this worker
will be used. OTOH if it is in error state, and the
sticky_session_force is set, then 500 will be returned.
If there is no session, and the worker is disabled, then
it will be skipped during worker election. This resolves
shutting down the node for maintenance.
If worker has the redirect parameter set, then all its
session requests will be redirected to the 'redirect'
worker. The param is meant to be used from jk_status for
redirecting nodes.
Domain is group of workers in load balancer that has
session replication in place. The jvmRoute in that case
is domain name, not the worker name. During election
phase, only the workers with same domain will be elected.
If all are in error state and the sticky_session_force is
set 500 will be returned. If not set another domain will
be elected.
I hope this resolves all use cases.
So the following directives are deprecated:
worker.xxx.local_worker
worker.xxx.local_worker_only
The following has been added:
worker.xxx.redirect (string, name of the worker or domain)
worker.xxx.sticky_session_force (boolean)
worker.xxx.domain (string, name of cluster domain/group)
Also there is a new lb directive called 'method' that
can be set to either to 0(request, default) or 1(traffic),
meaning that lb can balance depending on number of requests
or actual data transferred/read.
So, worker.lb.method=1 will balance on data transfer.
(Will not work with sticky sessions for now).
Regards,
Mladen.
---------------------------------------------------------------------
To unsubscribe, e-mail: tomcat-dev-unsubscribe@jakarta.apache.org
For additional commands, e-mail: tomcat-dev-help@jakarta.apache.org
Re: JK - need some testings for new lb
Posted by Dominik Drzewiecki <dr...@post.pl>.
As I mentioned during 1.2.8 development phase, building jk for win32 is a
bit more troublesome than on linux - it requires a non-free M$ VC.
Is anybody going to provide windoze binaries, please?
regz,
/dd
---------------------------------------------------------------------
To unsubscribe, e-mail: tomcat-dev-unsubscribe@jakarta.apache.org
For additional commands, e-mail: tomcat-dev-help@jakarta.apache.org
Re: JK - need some testings for new lb
Posted by Henri Gomez <he...@gmail.com>.
Great :)
Build it right now and test on my Linux boxes
On Wed, 16 Feb 2005 10:01:15 +0100, Mladen Turk <mt...@apache.org> wrote:
> Hi,
>
> As said before load balancer does not have any more all
> that fuzzy local_worker settings.
> For most people (at least for me) that was totally unusable
> and hard to understand. I hope the new settings and algo
> will be more clear and much usable in conjunction with
> shared memory and jk_status.
>
> There are two major concepts:
> Sessions and Domains (well, not quite new).
>
> If there is a session that matches the worker name,
> and the worker is not in error state, then this worker
> will be used. OTOH if it is in error state, and the
> sticky_session_force is set, then 500 will be returned.
>
> If there is no session, and the worker is disabled, then
> it will be skipped during worker election. This resolves
> shutting down the node for maintenance.
>
> If worker has the redirect parameter set, then all its
> session requests will be redirected to the 'redirect'
> worker. The param is meant to be used from jk_status for
> redirecting nodes.
>
> Domain is group of workers in load balancer that has
> session replication in place. The jvmRoute in that case
> is domain name, not the worker name. During election
> phase, only the workers with same domain will be elected.
> If all are in error state and the sticky_session_force is
> set 500 will be returned. If not set another domain will
> be elected.
>
> I hope this resolves all use cases.
>
> So the following directives are deprecated:
> worker.xxx.local_worker
> worker.xxx.local_worker_only
>
> The following has been added:
> worker.xxx.redirect (string, name of the worker or domain)
> worker.xxx.sticky_session_force (boolean)
> worker.xxx.domain (string, name of cluster domain/group)
>
> Also there is a new lb directive called 'method' that
> can be set to either to 0(request, default) or 1(traffic),
> meaning that lb can balance depending on number of requests
> or actual data transferred/read.
> So, worker.lb.method=1 will balance on data transfer.
> (Will not work with sticky sessions for now).
>
> Regards,
> Mladen.
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: tomcat-dev-unsubscribe@jakarta.apache.org
> For additional commands, e-mail: tomcat-dev-help@jakarta.apache.org
>
>
---------------------------------------------------------------------
To unsubscribe, e-mail: tomcat-dev-unsubscribe@jakarta.apache.org
For additional commands, e-mail: tomcat-dev-help@jakarta.apache.org