Checking the Cluster Manager agent cloud bursting setup

2 minute read

You should launch a build so you can verify that your Cluster Manager cloud setup is properly launching instances and do any troubleshooting if needed.

Launching a build

Launch a build by invoking eMake with a command such as:

emake --emake-cm=<cluster_manager_public_DNS> --emake-resource=<resource_name> --emake-root=<root_dir>

You can limit the number of launched instances. To do so, include the --emake-maxagents=N eMake option, where N is the number of agents. You can also do this by defining a build class. For details, see Creating Build Classes.

When you start the build, the instances are launched. The build starts immediately, but it might not make progress for some time, until agents are deployed and made available to the build. A No agents available message might appear while the instances are launching.

When the first instance is up, the build proceeds. When the build finishes, any agents it used are available for use with other builds using the same resource. If no other build uses an agent, that agent becomes idle. If a cloud burst agent is idle for 15 minutes (by default), the instance hosting that agent will be terminated automatically by the Cluster Manager. You can configure this timeout period by using the Instance Idle Timeout (minutes) setting when you create or edit a resource in the Cluster Manager. For details, see Resources - Create or Edit a Resource.

If agents did not even connect to the Cluster Manager within four minutes after being deployed, they are considered to be stalled. When a subsequent agent request is made, if the instance is stalled for the resource, the instance is terminated, and the agent request will return an error, which kills the build.

Using the Cluster Manager to check that the launched instances are running properly

After at least one build is currently running, navigate to the Cloud > Status subtab in the Cluster Manager web UI to see that your setup works. Launched instances appear in the Current Instances section on this subtab. For details about the information in the Cloud > Status subtab, see the Cloud topics.

Troubleshooting the agent cloud bursting setup

To troubleshoot your agent cloud bursting setup, on the Cluster Manager machine:

  • Examine the <install_dir>/i686_Linux/logs/accelerator.log and <install_dir>/i686_Linux/cloud/cloud.log files, where <install_dir> is /opt/ecloud by default.

  • See the state of the instances in the Cloud > Cloud Burst Status subtab in the Cluster Manager web UI.

    This subtab contains status for Kubernetes, Azure Services Platform, Amazon EC2, or Google Cloud Platform instances that are active (that is, instances that are running builds). If no instances are active, this page is blank. For details, see Cloud Burst Status.

  • See the Cloud > Cloud Messages subtab in the Cluster Manager web UI.

    This subtab contains messages about agent on-demand request processing (actions performed) on the cloud platform. For details, see Cloud Messages.