auto deactivation disabled

[Bhupesh Goel]

Hi paul,

I have a 2 node setup consisting of 2 separate instances of HA-JDBC part of two different tomcat. Each instance of HA-JDBC is connected to same set of 2 HSQL DBs. In my setup if i bring down one instance of tomcat as well as HSQL DB in same node then it doesn't get deactivated in other node. I am not using local tag in database config. Shouldn't down HSQL be deactivated in other node HA-JDBC instance? If this behavior is intentional then is there a way i can enforce deactivation in two node setup.

[Paul Ferraro]

You need to add <distributable/> to your ha-jdbc configuration file. See the "Distributed capabilities" section of the docs.

[Bhupesh Goel]

I am using distributable tag as well. Even then it doesn't get deactivated.

[Paul Ferraro]

OK - so let me see if I fully understand your scenario...
You have 2 tomcat nodes, each with an HSQLDB instance. When you shutdown one tomcat instance, the HSQLDB for that node is not getting deactivated, correct?

Currently, HA-JDBC will only deactivate a database if:
a. it tries to use it and it is unresponsive (lazy detection)
b. an failure-detect-schedule is configured (eager detection)

So, as you can see, we never deactivate database as part of a lifecycle of a cluster. However, I think you raise a good point. Upon shutdown, it would make sense for HA-JDBC to deactivate all "local" databases - since these will be detected and deactivated by remote nodes eventually.

[Paul Ferraro]

OK - I just added this logic to master. However, there is no universal mechanism for HA-JDBC to detect when your server is shutting down (except in order to leverage this functionality, you'll need to add a lifecycle listener to your Tomcat instances that triggers shutdown of your HA-JDBC cluster when your server or context shuts down.

Let's assume that we want to add a listener at the server level. In this case, you would add the following to your server.xml:
e.g.

<Server>
    <Listener class="org.myorg.DriverDatabaseClusterLifecycleListener" cluster="mycluster"/>
    <Service>...</Service>
</Server>

DriverDatabaseClusterLifecycleListener would look something like:

package org.myorg;

public class DriverDatabaseClusterLifecycleListener implements org.apache.catalina.LifecycleListener
{
    private String cluster;

    public void lifecycleEvent(org.apache.catalina.LifecycleEvent event)
    {
        if (event.getLifecycle().getState() == LifecycleState.STOPPED)
        {
            net.sf.hajdbc.sql.Driver driver = (net.sf.hajdbc.sql.Driver) java.sql.DriverManager.getDriver("jdbc:ha-jdbc:" + this.cluster);
            try
            {
                driver.stop(this.cluster);
            }
            catch (java.sql.SQLException e)
            {
                // Log the exception
            }
        }
    }

    public void setCluster(String cluster)
    {
        this.cluster = cluster;
    }
}

[Bhupesh Goel]

paul,

there is slight change in the scenario you described above. When i shutdown one tomcat instance as well as kill HSQLDB on same node, killed HSQLDB is not getting deactivated in other node HA-JDBC. I am using ha-jdbc v2.0.16.

[Paul Ferraro]

HA-JDBC 2.0.x does not automatically deactivate database when you shutdown your server. I just added this logic to 2.1 (in master).

The code I posted above is for HA-JDBC 2.1. I highly recommend upgrading...

The equivalent listener logic in 2.0.x would involve unregistering the relevant mbeans from the mbean server (there is no public stop() method on the HA-JDBC driver in 2.0.x).

[Bhupesh Goel]

Paul,

i guess you didn't get my scenerio. let us say we have distributed setup of tomcat_1 and tomcat_2 in node 1 and node 2. Same way we have hsql_1, hsql_2. When i kill tomcat_2 and hsql_2 in 2nd node then ha-jdbc instance in tomcat_1 doesnt deactivate hsql_2 and shows both hsql_1 and hsql_2 as active. I am not using local attribute with database. So, ideally ha-jdbc in tomcat_1 should deactivate hsql_2.

I guess it has to do something with 1 node cluster. There is a check in ha-jdbc code where if number of remote nodes in cluster is zero then ha-jdbc skip deactivation step altogether.