SurfZone is a cost management feature that solves two difficult challenges facing cloud users.

• Proactively stopping a job before the budget is exceeded
• Saving the state of a job so that the job can resume when more budget is available

With this feature, users are assigned to a SurfZone, which is a policy that defines the quota (also known as limit or budget), the threshold for issuing a warning, the action taken when the quota is exhausted, and the action taken when the budget is replenished. The quota is checked periodically; the interval is configurable (default is one hour and also the minimum).

Nextflow is a framework for assembling a collection of tasks into a coordinated pipeline or workflow. One Nextflow pipeline can generate hundreds of tasks that appear to the OpCenter as hundreds of independent jobs. When a workflow is started, Nextflow assigns it a name, for example, furious-mandelbrot. The OpCenter web interface now supports a Workflows screen that displays all the workflows by name. Clicking a workflow name brings up the Workflow Details screen that displays all the tasks started as part of that workflow as well as aggregate metrics of the entire workflow.

Similar information (without the aggregate metrics) is available using the cli, for example, float list -f "tags=nextflow-io-run-name:furious-mandelbrot".

The workflow view is available for any group of jobs (not just Nextflow pipelines) as long as the user provides a workflow name and a run name (analogous to the pipeline name in Nextflow) for each job in the group.

Multiple Machine Images (also called Quiver) is a feature that enables the OpCenter to differentiate among the virtual machine images (VMIs) used to instantiate worker nodes. In earlier releases, the OpCenter uses a single VMI (based on an x86 CPU) for all worker nodes. With Quiver, the OpCenter can select a VMI that is customized for the intended task. For example, the OpCenter can select a VMI that supports an NVIDIA GPU or a VMI custom-built to support specific customer requirements.

Although originally developed for graphics applications, the architecture of a GPU (graphics processing unit) is ideal for accelerating the type of mathematical calculations involved in machine learning and artificial intelligence. The Half Moon Bay release supports jobs that are submitted with the drivers to use NVIDIA GPUs. The Half Moon Bay release does not support GPU checkpoint and restore functions, so jobs can run on GPUs, but SpotSurfer and WaveRider are not available for these jobs.

The Dana Point 2.1.1 release introduced LDAP and local Linux password file as authentication methods (in addition to the built-in method) to use when logging in to the OpCenter. Half Moon Bay keeps the built-in method, removes the local Linux password file method (due to the lack of customer demand), and enhances the LDAP support. The LDAP enhancements are extensive; for example, security for LDAP bind credentials is improved as follows.

• Removal of LDAP-related parameters from the OpCenter configuration list: The LDAP bind credentials are removed from the visible configuration list to prevent unauthorized access or exposure.
• Dedicated command line and OpenAPI for LDAP configuration: A specialized command line tool and OpenAPI endpoints are introduced (for configuring and testing the LDAP connection).
• Mutual TLS verification for LDAP connection: Users can upload their own certificates to enable anonymous access in a trusted environment.

When OpCenter is used in a high-volume production environment, for example, a company delivering genomic analyses as a service, multiple thousands of jobs may be submitted per day. Each of these jobs generates extensive logs that must be retained. In previous releases, logs are stored in the OpCenter's root volume. To improve scalability, Half Moon Bay allows the use of an NFS server to provide storage volumes for the logs.

For a busy OpCenter supporting hundreds or even thousands of jobs, viewing a complete list of jobs is not practical. The Half Moon Bay release introduces extensive filtering capabilities (both in the CLI and the web interface) so that a user can narrow down the list of jobs in a display (or in a report) to only the jobs of interest.

For each job, the OpCenter compiles a number of logs. In addition, the OpCenter compiles logs related to the OpCenter itself. For troubleshooting, it is useful to download some or all of the logs. The Half Moon Bay release provides the capability (in the web interface and CLI) to download user-selected logs in a single zip file.

Users can now submit jobs with the --output /path/to/dir option so that stdout and stderr are permanently saved (in a persistent EBS volume or on an NFS server).

The status of the --errPolicy flag introduced as a preview feature in the Goa 2.4 release changes to general availability.

Similar to the App Library, the Job Template library now includes a private repository to store job templates created by the user. In this release, the user creates a private template by saving a submitted job. This capability is only available for jobs that were submitted to an OpCenter running Half Moon Bay or a later release. After the template is created, the user can use the template to start additional jobs. Private templates are created using the web interface or the CLI.

This release fixes an issue where charges for jobs run in AWS regions where the local currency is not US dollars were incorrectly reported.