Integrating an Existing Grid Engine Installation
From UniCluster
Contents |
Integrating an Existing Grid Engine Installation
With the release of UniCluster Express version 3.2 it is possible to integrate an existing Grid Engine instance into UniCluster Express during the UniCluster Express installation process. When using an existing Grid Engine instance, UniCluster Express does not attempt to configure Grid Engine or integrate the Monitor Console into the existing Grid Engine install as these steps would vary based on the version and configuration of the existing Grid Engine install. All nodes within a UniCluster Express based cluster *must* be based on existing Grid Engine installs or all *must* be based on the bundled versions of Grid Engine. You cannot mix and match nodes with existing UniCluster Express installations with nodes that use the bundled versions of the Grid Engine modules.
Installer Options
The --sge-existing-path, --sge-existing-cell-name, and --sge-existing-default-gram-queue configuration options provide the new existing Grid Engine installation configuration options
- --sge-existing-path=<path to existing sge root> - This option is used to pass in a path to an existing $SGE_ROOT. The path must exist or the UniCluster Express installer will throw an error.
- --sge-existing-cell-name=<existing cell name> - This is the value of $SGE_CELL in the existing Grid Engine installation. This option defaults to "default" and only takes affect if --sge-existing-path is set. There must also be a valid install of this cell inside of --sge-existing-path or the installer will exit.
- --sge-existing-default-gram-queue configuration - This option is used to configure the default queue used when issuing jobs to Grid Engine through GRAM. The default value is "all.q" and only takes affect if --sge-existing-path is set.
Differences From Standard Install
When using an existing Grid Engine installation all UniCluster Express bundled Grid Engine components are disabled including the core modules, the system init scripts, and the configuration agents. Because the configuration of the existing Grid Engine is not managed by UniCluster Express the Monitor Console cannot be completely integrated with Grid Engine during installation. This can, however, be manually configured after installation. If the manual post install steps are not performed the Monitor Console will still be able to display non-Grid Engine information as well as the realtime Grid Engine data obtainable through "qstat".
In order to maintain a similar installation structure between UniCluster Express instances incorporating an existing Grid Engine install and UniCluster Express instances with a new Grid Engine install, a symbolic link is created at <uce-install-path>/sge that points to the value of --sge-existing-path instead of a Grid Engine instance being installed at <ce-install-path>/sge.
Enabling Monitoring
In order to completely tie the UniCluster Express Console Monitor into an existing Grid Engine installation, a few manual steps need to be executed on the main node.
- Download the appropriate version of ARCo that matches the version of the existing Grid Engine installation.
- Install ARCo against a PostgreSQL database. The UniCluster Express instance installed on the main node can be used, but new database users must be created first.
- Enable reporting in Grid Engine by adding "reporting=true" value to the reporting_params attribute in Grid Engine.
- Update the MonitorService configuration at <install-path>/etc/univa_mgmt_service/jndi-config.xml by setting the "dbURL", "dbUser", and "dbPassword" parameters values to match the configuration of the ARCo PostgreSQL server.
- Restart the container with (as root):
$ /etc/init.d/univa-container restart
At this point the Console Monitor should be able to display data retrieved from your ARCo installation.
Back to Administrative How Tos.