Re: ProxyPass and HA Singleton Applications

[Federico Calì <f.cali@xxxxxxxx>]

Hi Roman,
my scenario is similar what you describe, but whit a little difference:
I have one or more contexts served by all of the nodes in the cluster, and another contexts served only by the Master node.
The Master node manages the contexts deployed in HA Singleton.
Only one Master node is active in the cluster, and the cluster manages the Master election.
When the Master node fails, the cluster elects a new Master node and contexts being served by it.

Using mod_cluster each back-end node can advise contexts that are serving to the apache, in this case the switch of Master Node is automatically notified to the front-end.

I would like to implement a similar solution using ProxyPass, why this?

I think that problem "503" may be caused by the JBoss GC (stop the world) activity.
When it occurs on HA Singleton MASTER node it makes the web-applications not availables, and the node is considered "off-line" for 1 minute by the apache.
No problem arises for "standard cluster applications running on all nodes" because another node can responde.

I've seen that "retry=0" parameters using ProxyPass should avoid the 1 minute "black out" that generates the 503 errors.
I'm not able to find the same behavior using mod_cluster.

Federico      

Il 04/07/2014 05:03, Roman Jurkov ha scritto:

Federico,

503 would be due to a node being not responsive, you can configure your cluster to disable a node after one or more failures with mod_cluster. now back to mod_proxy_balancer and mod_proxy, if i understand correctly, you have a scenario where you have one or more contexts served by all of the nodes in the cluster, and another context served only by some nodes, i.e.:

Node1 => /foo,/bar

Node2 => /foo

Node3 => /foo,/bar

you want /foo to be served  by all of the nodes, and /bar just by Node1, and have Node3 as a hot standby (failover), if that is the case, you would probably need multiple load balancers defined with two different location blocks, so something like this:

<proxy balancer://foocluster>

    BalancerMember http://node1:8080

    BalancerMember http://node2:8080

    BalancerMember http://node3:8080

</proxy>

<proxy balancer://barcluster>

    BalancerMember http://node1:8080

    # the hot standby on node3

    BalancerMember http://node3:8080 status=+H

</proxy>

<location /foo>

    ProxyPass           balancer://foocluster

    ProxyPassReverse    http://node1:8080

    ProxyPassReverse    http://node2:8080

    ProxyPassReverse    http://node3:8080

</location>

<location /bar>

    ProxyPass           balancer://barcluster

    ProxyPassReverse    http://node1:8080

    ProxyPassReverse    http://node3:8080

</location>

i haven’t tested above configuration.

-Roman