7 Replies Latest reply on Apr 11, 2013 2:39 AM by michal_rakoczy

mod_jk redirect for worker

michal_rakoczy Apr 9, 2013 8:11 AM

I would like to ask what is the proper configuration for loadbalancing worker and non-balanced worker referring to retries parameter.

My current configuration :

worker.list=loadbalancer,node3,node4

worker.loadbalancer.type=lb

worker.loadbalancer.balance_workers=node3,node4

worker.loadbalancer.sticky_session=1

worker.loadbalancer.method=R

worker.template.type=ajp13

worker.template.socket_connect_timeout=5000

worker.template.socket_keepalive=true

worker.template.ping_mode=A

worker.template.ping_timeout=10000

worker.template.connection_pool_minsize=0

worker.template.connection_pool_timeout=60

worker.template.reply_timeout=300000

worker.template.recovery_options=3

worker.node3.reference=worker.template

worker.node3.port=8309

worker.node3.host=127.0.0.1

worker.node3.type=ajp13

worker.node3.lbfactor=1

worker.node3.redirect=node4

worker.node4.reference=worker.template

worker.node4.port=8409

worker.node4.host=127.0.0.1

worker.node4.type=ajp13

worker.node4.lbfactor=1

worker.node4.redirect=node3

Before balancing node3 and node4 i used retries parameter for both workers, to avoid double requests to tomcat --> http://www.itpile.com/r2089/apache-mod-jk-repeats-request-to-tomcat-why :

worker.node3.port=8309

worker.node3.host=127.0.0.1

worker.node3.type=ajp13

worker.node3.lbfactor=1

worker.node3.retries=1

worker.node3.connect_timeout=15000

worker.node3.socket_timeout=60

worker.node3.connection_pool_timeout=60

worker.node3.socket_keepalive=True

worker.node4.port=8409

worker.node4.host=127.0.0.1

worker.node4.type=ajp13

worker.node4.lbfactor=1

worker.node4.retries=1

worker.node4.connect_timeout=15000

worker.node4.socket_timeout=60

worker.node4.connection_pool_timeout=60

worker.node4.socket_keepalive=True

I had skipped retries parameter between non-balanced and balanced configuration as i'm confused where to set it up - on loadbalancer or directly on node, i would like to also avoid infinite loops (is it possible ?) because of misconfiguration.

The doubts are :

- if retries parameter for loadbalancer is set up to 0 is it equal to node parameter equal 1 ?

- how the request is handled when retries is set up on loadbalancer and balanced worker simultaneously - which of them has higher "priority" ?

1. Re: mod_jk redirect for worker

mladen.turk Apr 10, 2013 7:34 AM (in response to michal_rakoczy)

Hi,

You should not mix load balancer and standard workers.
I suppose you are trying to have failover worker or something like that. Am I right?
In that case those workers should be part of the same load balancer.
Actions
2. Re: mod_jk redirect for worker

mladen.turk Apr 10, 2013 7:38 AM (in response to michal_rakoczy)

... and also, the more important thing:
members of load balancer must not appear in worker.list directive
Actions
3. Re: mod_jk redirect for worker

michal_rakoczy Apr 10, 2013 8:54 AM (in response to mladen.turk)

Thank you Mleden for you answer. Could you explain it more as in my configuration i MUST mix balanced and non-balanced workers.

I have few applications running on each JBoss intance, some of them are replicated on balanced workers (application is prepared for that) some of them work on the one instance only. That's why i need to mix both types of workers, and also this is why it's appear in worker.list directive.

Apache configuration :

<VirtualHost>

JkMount /App3 node3
JkMount /App4 node4
JkMount /App5 loadbalancer

</VirtualHost>

As you may see App3 and App5 is deployed on node3 and App4 and App5 is deployed on node4. It means that only App5 is balanced - in case of node failure only requests to /App5 are redirected to spare node. It's also the reason why i node3 and node4 appear in worker.list - as i need to redirect requests to App3 and App4 directly to proper node, not to loadbalancer.

You also asked : I suppose you are trying to have failover worker or something like that. Am I right? - No, i'm not trying, i have failover workers but only for App5 (at least this is how it works for me) - correct me if i'm wrong please.
Actions
4. Re: mod_jk redirect for worker

mladen.turk Apr 10, 2013 9:11 AM (in response to michal_rakoczy)

You should create two additional workers for non load balancer nodes.
That way you will have 3 workers in worker.list eg loadbalancer,app3worker,app4worker

They can point to the same appserver instance (even have the same ip:port) as loadbalancer members.
However they will be separated inside internal mod_jk shared memory making them separate.
Actions
5. Re: mod_jk redirect for worker

michal_rakoczy Apr 10, 2013 9:24 AM (in response to mladen.turk)

So if i understand it properly i should ammend the configuration to :

worker.list=loadbalancer,node3worker,node4worker
worker.loadbalancer.type=lb
worker.loadbalancer.balance_workers=node3,node4
worker.loadbalancer.sticky_session=1
worker.loadbalancer.method=R

worker.template.type=ajp13
worker.template.socket_connect_timeout=5000
worker.template.socket_keepalive=true
worker.template.ping_mode=A
worker.template.ping_timeout=10000
worker.template.connection_pool_minsize=0
worker.template.connection_pool_timeout=60
worker.template.reply_timeout=300000
worker.template.recovery_options=3

worker.node3.reference=worker.template
worker.node3.port=8309
worker.node3.host=127.0.0.1
worker.node3.type=ajp13
worker.node3.lbfactor=1
worker.node3.redirect=node4

worker.node3worker.reference=worker.template
worker.node3worker.port=8309
worker.node3worker.host=127.0.0.1
worker.node3worker.type=ajp13

worker.node4.reference=worker.template
worker.node4.port=8409
worker.node4.host=127.0.0.1
worker.node4.type=ajp13
worker.node4.lbfactor=1
worker.node4.redirect=node3

worker.node4worker.reference=worker.template
worker.node4worker.port=8409
worker.node4worker.host=127.0.0.1
worker.node4worker.type=ajp13

and

<VirtualHost>

JkMount /App3 node3worker
JkMount /App4 node4worker
JkMount /App5 loadbalancer

</VirtualHost>

Is that correct ?
Actions
6. Re: mod_jk redirect for worker

mladen.turk Apr 10, 2013 11:35 AM (in response to michal_rakoczy)

Yes.

Just note that the potential number of connections between httpd and AS will double. Same as having two httpd's in front.
So perhaps you can choose to have two vhosts on AS and two connectors on different ports. But that's optional.
Actions
7. Re: mod_jk redirect for worker

michal_rakoczy Apr 11, 2013 2:39 AM (in response to mladen.turk)

What about retries parameter ?
The doubts are :

- if retries parameter for loadbalancer is set up to 0 is it equal to node parameter equal 1 ?

- how the request is handled when retries is set up on loadbalancer and balanced worker simultaneously - which of them has higher "priority" ?
Actions

Go to original post