Mod cluster Installation: Difference between revisions

From NovaOrdis Knowledge Base
Jump to navigation Jump to search
Line 321: Line 321:
After JBoss nodes have started, the httpd manager console available at http://1.2.3.4:8088/mod_cluster-manager should look similar to:
After JBoss nodes have started, the httpd manager console available at http://1.2.3.4:8088/mod_cluster-manager should look similar to:


[[Image:Mod cluster-manager.png|frame]]
[[Image:Mod cluster-manager.png|center|frame|Successful mod_cluster installation]]

Revision as of 03:35, 10 January 2016

External

Internal

Download

Compiled bundles are available here: http://mod-cluster.jboss.org/downloads

Native Components Installation on Linux

This procedure assumes httpd was already installed, configured, and it works well without mod_cluster. Details on httpd installation and configuration are available here: httpd installation, httpd configuration.

At the time of the writing, the pre-compiled native components to be installed within an existing httpd are published on the download page as "mod_cluster modules for httpd". For example, the mod_cluster binaries for Linux x86 httpd are listed as "linux-x86_64 mod_cluster binaries". The file name is mod_cluster-1.3.1.Final-linux2-x64-so.tar.gz. It contains the pre-compiled httpd dynamic libraries (mod_advertise.so, mod_cluster_slotmem.so, mod_manager.so and mod_proxy_cluster.so).

Place the Dynamic Libraries in the modules Directory

Identify the httpd instance's module directory, by looking at the existing LoadModule directives specified in the configuration. It is usually /etc/httpd/modules.

It is good practice to include the mod_cluster version in the name of the dynamic library files, to make it obvious what version is used.

cp mod_advertise.so /etc/httpd/modules/mod_advertise-1.3.1.Final.so 
cp mod_cluster_slotmem.so /etc/httpd/modules/mod_cluster_slotmem-1.3.1.Final.so 
cp mod_manager.so /etc/httpd/modules/mod_manager-1.3.1.Final.so 
cp mod_proxy_cluster.so /etc/httpd/modules/mod_proxy_cluster-1.3.1.Final.so 

Make sure the newly copied files have the same ownership and permissions as the existing modules.

Load Modules at Startup

A standard RHEL httpd installation has its modules configured in /etc/httpd/conf.modules.d. All files present in that directory are automatically included from httpd.conf:

...
Include conf.modules.d/*.conf
...

Follow the same pattern to configure mod_cluster.

Add a new 99-mod_cluster.conf file in /etc/httpd/conf.modules.d with the following content:

#
# This file configures mod_cluster
#

LoadModule proxy_cluster_module modules/mod_proxy_cluster-1.3.1.Final.so
LoadModule cluster_slotmem_module modules/mod_cluster_slotmem-1.3.1.Final.so
LoadModule manager_module modules/mod_manager-1.3.1.Final.so
LoadModule advertise_module modules/mod_advertise-1.3.1.Final.so

Insure that Module Dependencies are Present and Loaded

mod_cluster needs mod_proxy to work. More details about mod_cluster dependency on mod_proxy are found here.

Make sure mod_proxy.so and mod_proxy_ajp.so are loaded. For a standard httpd installation, those modules are loaded in /etc/httpd/conf.modules.d/00-proxy.conf:

...
LoadModule proxy_module modules/mod_proxy.so
LoadModule proxy_ajp_module modules/mod_proxy_ajp.so
...

Remove Undesired Modules

mod_cluster's mod_proxy_cluster cannot work correctly if mod_proxy's mod_proxy_balancer, so mod_proxy_balancer must be removed from the httpd configuration.

For a standard httpd deployment, mod_proxy_balancer is loaded in /etc/httpd/conf.modules.d/00-proxy.conf. Comment out the following line:

#LoadModule proxy_balancer_module modules/mod_proxy_balancer.so

SELinux Configuration

If SELinux it is being enforced, we will need to install additional policies to allow httpd to write mod_cluster-related files in /var/log/httpd, to perform multicast socket operations and bind on non-standard ports.

Start with a mod_cluster_1.te policy file:


module mod_cluster_1 1.0;

require {
        type httpd_log_t;
        type httpd_t;
        type unreserved_port_t;
        class tcp_socket { name_bind name_connect };
        class dir remove_name;
        class file { write unlink };
        class udp_socket name_bind;
}

#============= httpd_t ==============
allow httpd_t httpd_log_t:dir remove_name;
allow httpd_t httpd_log_t:file { write unlink };
allow httpd_t unreserved_port_t:tcp_socket { name_bind name_connect };
allow httpd_t unreserved_port_t:udp_socket name_bind;

Compile the policy:

checkmodule -M -m -o mod_cluster_1.mod mod_cluster_1.te

Create the module package:

semodule_package -o  mod_cluster_1.pp -m mod_cluster_1.mod 

Install the policy:

semodule -i mod_cluster_1.pp

Verify that the policy was installed:

semodule -l | grep mod_cluster_1

More details about managing SELinux policies are available here: SELinux Operations.

Basic httpd.conf Configuration

This is a simple httpd configuration that can be used to validate the mod_cluster installation.

The configuration declares just one virtual node that gives access to one mod_cluster manager for one balancer. For a discussion on mod_cluster balancers, managers and virtual nodes, see mod_cluster Concepts.

Listen Directive

Note that the manager is exposed on port 8088, so httpd needs to be made to listen on that port:

Listen ...
Listen 1.2.3.4:8088

If the host that runs httpd has iptables enabled, you may want to open access to port 8088. Go here for more iptables configuration details.

Manager Virtual Host

Declare the mod_cluster manager virtual host. It is critical to enable MCPM receive with EnableMCPMReceive otherwise nodes won't be able to register themselves with httpd.

Note that multicast advertising is turned off and the application server nodes will have to be explicitly configured to connect to this manager.

Also note that access is allowed to all, you may want to configure stricter access rules in production.

In case you need to troubleshoot mod_cluster problems, you may want to uncomment LogLevel debug.

<VirtualHost 1.2.3.4:8088>

    # LogLevel debug

    <Directory />
        Allow from all
    </Directory>

    KeepAliveTimeout 60
    MaxKeepAliveRequests 0

    ManagerBalancerName mycluster
    EnableMCPMReceive
    ServerAdvertise Off

    <Location /mod_cluster-manager>
        SetHandler mod_cluster-manager
        Allow from all
    </Location>

</VirtualHost>

<Directory /> Permissions

httpd:

[Sat Jan 09 19:12:10.564490 2016] [:debug] [pid 1715] mod_manager.c(2271): manager_trans INFO (/)
[Sat Jan 09 19:12:10.564590 2016] [authz_core:debug] [pid 1715] mod_authz_core.c(809): [client 172.20.2.43:34545] AH01626: authorization result of Require all denied: denied
[Sat Jan 09 19:12:10.564599 2016] [authz_core:debug] [pid 1715] mod_authz_core.c(809): [client 172.20.2.43:34545] AH01626: authorization result of <RequireAny>: denied
[Sat Jan 09 19:12:10.564621 2016] [authz_core:error] [pid 1715] [client 172.20.2.43:34545] AH01630: client denied by server configuration: /

client:

19:12:20,564 ERROR [org.jboss.modcluster] (UndertowEventHandlerAdapter - 1) MODCLUSTER000042: Error null sending INFO command to 172.20.2.41/172.20.2.41:8088, configuration will be reset: null

Temporarily circumvented by enabling access on the "/" directory as shown below. However, this is not a viable solution. Research the proper solution and move the above failures in a "Troubleshooting" section.

#<Directory />
#    AllowOverride none
#    Require all denied
#</Directory>
<Directory />
    AllowOverride none
    Require all granted
</Directory>

Further Configuration

More complex configurations are of course possible. For more details on how to configure mod_cluster, go to mod_cluster Configuration.

Native Component Installation on Windows

TODO, manual page 6, section 2.2.4 Install on Windows.

iptables Configuration

If iptables is enabled, make sure the multicast, MCPM, HTTP and AJP traffic are allowed on various mod_cluster hosts that need to talk to each other:

MCPM

On the httpd node, make sure MCPM inbound traffic is allowed:

-A INPUT -i eth0 -p tcp -m tcp --dport 8088 -m state --state NEW -j ACCEPT

mod_cluster Manager Port

If you chose to expose the mod_cluster manager on non-standard port, make sure the nodes have access to that port. You could open the port to everyone (non-recommended) or restrict access to that port only to the hosts that are supposed to be able to access it (recommended):

-A INPUT -i eth0 -p tcp -m tcp --dport 8088 -m state --state NEW -s 172.20.0.0/16 -j ACCEPT

HTTP and AJP Traffic

On the JBoss nodes, add this to /etc/sysconfig/iptables:

-A INPUT -i eth0 -p tcp -m tcp --dport 8080 -m state --state NEW -j ACCEPT
-A INPUT -i eth0 -p tcp -m tcp --dport 8009 -m state --state NEW -j ACCEPT

Multicast Traffic

Multicast traffic should propagate from the httpd nodes to the JBoss node and maybe back? Are JBoss nodes multicasting to httpd? Research that.. Normally iptables is configured to leave all traffic out by default, so the httpd nodes should be fine. However, JBoss nodes must be enabled to allow the multicast traffic in, so add this to /etc/sysconfig/iptables:

-A INPUT -m pkttype --pkt-type multicast -j ACCEPT

More details on configuring iptables are available here: iptables configuration examples.

JBoss Component Installation

standard-ha and standard-full-ha WildFly 9 profiles come with mod_cluster enabled.

Static Proxy List

If the operational environment does not support multicast traffic, the JBoss nodes can be configured with a static httpd server list, as shown below. Unlike mod_jk, where the httpd configuration needs to be changed every time a new JBoss node is brought on-line, in mod_cluster's case JBoss nodes register themselves dynamically with httpd. This is a considerable advantage in environments with a large number of JBoss nodes.

WildFly 9

    ...
    <subsystem xmlns="urn:jboss:domain:modcluster:2.0">
        <mod-cluster-config proxies="mod-cluster-server-1" advertise="false" connector="ajp">
           <dynamic-load-provider>
              <load-metric type="cpu"/>
           </dynamic-load-provider>
       </mod-cluster-config>
    </subsystem>
     ...
    <socket-binding-group name="standard-sockets"...>
        ...
        <outbound-socket-binding name="mod-cluster-server-1">
            <remote-destination host="172.20.2.41" port="8088"/>
        </outbound-socket-binding>
        ...
    </socket-binding-group>
    ...

WildFly 8

        ...
        <subsystem xmlns="urn:jboss:domain:modcluster:1.2">
            <mod-cluster-config proxy-list="1.2.3.4:8088"  advertise="false" connector="ajp">
                <dynamic-load-provider>
                    <load-metric type="busyness"/>
                </dynamic-load-provider>
            </mod-cluster-config>
        </subsystem>
        ...

Multicast Advertise

See mod_cluster JBoss Configuration#Multicast_Advertise

Test Initial Installation

Start httpd, it should start without warnings or errors.

mod_cluster will create a set of new files in (by default) /var/log/httpd: manager.*.slotmem and manager.*.lock.

JBoss nodes should start without warning or errors.

They should register with httpd servers they were configured with.

After JBoss nodes have started, the httpd manager console available at http://1.2.3.4:8088/mod_cluster-manager should look similar to:

Successful mod_cluster installation