Operations

The Omnivector Slurm Distribution can be configured to the extent that you can configure Slurm itself. Meaning that all upstream Slurm configurations are supported by OSD.

The operational magic in OSD is encapsulated in ops code called charms. We have curated a suite of automation for each Slurm component: slurmctld, slurmd, slurmdbd, and slurmrestd.

Each Slurm component charm provides a set of options that enable operators/administrators to operate the cluster.

The slurm.conf file can be customized through the slurmctld charm configuration options. There is no need to manually update this file in all compute nodes: OSD combines the power of Juju operations with Slurm in configless mode (see OSD Architecture for details) to configure all the nodes in the cluster with one single command. This also means that any modification to a node's slurm.conf file will be discarded.

Alongside the capability to manage the slurm.conf, the charms also provide a set of operational tools called actions.

Find below a list of operations on how to manage and configure OSD. Refer to Configuration and Administration for a complete list of all configuration options and actions.

• Slurm Installation
  • Changing the repository
• Nodes operations
  • Adding nodes
  • Removing nodes
  • Draining nodes
  • Resuming nodes
  • Rebooting Nodes
• Partitions
  • Adding partitions
  • Changing partition name
  • Setting the default partition
  • Changing partition state
  • Extra partition configuration options
• Munge operations
  • Creating an etcd account
  • Querying the munge key
  • Configuring HTTPS
• Infiniband
• GPU Support
  • Nvidia
• NHC
  • Acquire and Provide NHC
  • Configuration
• Singularity
• Backup and restore the Slurm database
• Upgrade OSD
  • Charm Upgrades
• Accounting And Profiling
  • Slurmdbd
  • ElasticSearch plugin
  • InfluxDB profiling plugin
• Monitoring
  • Prometheus node_exporter
  • Prometheus Slurm exporter
• Logging
  • Fluentbit