Troubleshooting issues in clusters with Amazon Batch integration
This section is relevant to clusters with Amazon Batch scheduler integration.
Head node issues
You can troubleshoot head node setup issues in the same way as a Slurm cluster (except for Slurm specific logs). For more information about these issues, see Head node.
Compute issues
Amazon Batch manages the scaling and compute aspects of your services. If you encounter compute related issues, see the Amazon Batch troubleshooting documentation for help.
Job failures
If a job fails, you can run the awsbout command to retrieve the job output. You can also run the awsbstat command to obtain a link to the job logs stored by Amazon CloudWatch.
Connect timeout on endpoint URL error
If multi-node parallel jobs fail with error: Connect timeout on endpoint URL
:
-
In the
awsbout
output log, check that the job is multi-node parallel from the output:Detected 3/3 compute nodes. Waiting for all compute nodes to start.
-
Verify whether the compute nodes subnet is public.
Multi-node parallel jobs don't support the use of public subnets when using Amazon Batch in Amazon ParallelCluster. Use a private subnet for your compute nodes and jobs. For more information, see Compute environment considerations in the Amazon Batch User Guide. To configure a private subnet for your compute nodes, see Amazon ParallelCluster with Amazon Batch scheduler.