Immediate eviction of a dead worker

[thys.dewet]

Hi

Our setup is as follows :

Apache 2.2.22

Mod_Cluster 1.2.6 Final

This connects to two Tomcat 7.0.40 instances.

If I do an orderly shutdown of the tomcat instance, the next request is sent to the live node, as expected.

However, if I kill the tomcat process abruptly, the requests are still attempted to the now dead host. I see the ajp connection failing, and after a while the connection fails over.

Is there a way of forcing an immediate eviction of a dead worker ?

Config :

<VirtualHost *:6666>

        MaxKeepAliveRequests 0

        AdvertiseFrequency 5

        AdvertiseSecurityKey mySecurityKey

        <Location />

            Order deny,allow

            Allow from 127.0.0

        </Location>

        ServerAdvertise On

        EnableMCPMReceive

  </VirtualHost>

Further info :

When the hosts goes away, mod_cluster manager reports the host as NOTOK, and the apache logs show :

[Fri Sep 27 10:20:51 2013] [error] proxy: ajp: disabled connection for (196.xx.xx.xx)

[Fri Sep 27 10:20:56 2013] [error] proxy: CLUSTER: (balancer://liferay). All workers are in error state for route (quack)

[jfclere]

A request is needed to find that the tomcat has been killed... It seems you don't do failover, is that correct?

[thys.dewet]

I do want to do failover.

My SSL vhosts contains the following :

<VirtualHost xxxxxxxxxx:443>

    ProxyPass / balancer://liferay stickysession=JSESSIONID|jsessionid nofailover=Off

    ProxyPassReverse / balancer://liferay

    ProxyPreserveHost On

......

</VirtualHost>

[rhusar]

However, if I kill the tomcat process abruptly, the requests are still attempted to the now dead host. I see the ajp connection failing, and after a while the connection fails over.

How long is that while?

[thys.dewet]

Radoslav Husar wrote:

However, if I kill the tomcat process abruptly, the requests are still attempted to the now dead host. I see the ajp connection failing, and after a while the connection fails over.

How long is that while?

5 Minutes

[mbabacek]

5 minutes? Sounds like a familiar number: Re: mod_cluster not dropping workers after abrupt stop.