OpenPBS

Connect PBS with CycleCloud

CycleCloud manages OpenPBS clusters through an installable agent called azpbs. This agent connect to CycleCloud to read cluster and VM configurations and also integrates with OpenPBS to effectively process the job and host information. All azpbs configurations are found in the autoscale.json file, normally /opt/cycle/pbspro/autoscale.json.

  "password": "260D39rWX13X",
  "url": "https://cyclecloud1.contoso.com",
  "username": "cyclecloud_api_user",
  "logging": {
    "config_file": "/opt/cycle/pbspro/logging.conf"
  },
  "cluster_name": "mechanical_grid",

Important Files

The azpbs agent parses the PBS configuration each time it's called - jobs, queues, resources. Information is provided in the stderr and stdout of the command as well as to a log file, both at configurable levels. All PBS management commands (qcmd) with arguments are logged to file as well.

All these files can be found in the /opt/cycle/pbspro/ directory where the agent is installed.

File	Location	Description
Autoscale Config	autoscale.json	Configuration for Autoscale, Resource Map, CycleCloud access information
Autoscale Log	autoscale.log	Agent main thread logging including CycleCloud host management
Demand Log	demand.log	Detailed log for resource matching
qcmd Trace Log	qcmd.log	Logging the agent qcmd calls
Logging Config	logging.conf	Configurations for logging masks and file locations

Defining OpenPBS Resources

This project allows for a generally association of OpenPBS resources with Azure VM resources via the cyclecloud-pbspro (azpbs) project. This resource relationship defined in autoscale.json.

The default resources defined with the cluster template we ship with are

{"default_resources": [
   {
      "select": {},
      "name": "ncpus",
      "value": "node.vcpu_count"
   },
   {
      "select": {},
      "name": "group_id",
      "value": "node.placement_group"
   },
   {
      "select": {},
      "name": "host",
      "value": "node.hostname"
   },
   {
      "select": {},
      "name": "mem",
      "value": "node.memory"
   },
   {
      "select": {},
      "name": "vm_size",
      "value": "node.vm_size"
   },
   {
      "select": {},
      "name": "disk",
      "value": "size::20g"
   }]
}

The OpenPBS resource named mem is equated to a node attribute named node.memory, which is the total memory of any virtual machine. This configuration allows azpbs to process a resource request such as -l mem=4gb by comparing the value of the job resource requirements to node resources.

Note that disk is currently hardcoded to size::20g. Here is an example of handling VM Size specific disk size

   {
      "select": {"node.vm_size": "Standard_F2"},
      "name": "disk",
      "value": "size::20g"
   },
   {
      "select": {"node.vm_size": "Standard_H44rs"},
      "name": "disk",
      "value": "size::2t"
   }

Autoscale and Scalesets

CycleCloud treats spanning and serial jobs differently in OpenPBS clusters. Spanning jobs will land on nodes that are part of the same placement group. The placement group has a particular platform meaning (VirtualMachineScaleSet with SinglePlacementGroup=true) and CC will managed a named placement group for each spanned node set. Use the PBS resource group_id for this placement group name.

The hpc queue appends the equivalent of -l place=scatter:group=group_id by using native queue defaults.

Installing the CycleCloud OpenPBS Agent azpbs

The OpenPBS CycleCloud cluster will manage the installation and configuration of the agent on the server node. The preparation includes setting PBS resources, queues, and hooks. A scripted install can be done outside of CycleCloud as well.

# Prerequisite: python3, 3.6 or newer, must be installed and in the PATH
wget https://github.com/Azure/cyclecloud-pbspro/releases/download/2.0.5/cyclecloud-pbspro-pkg-2.0.5.tar.gz
tar xzf cyclecloud-pbspro-pkg-2.0.5.tar.gz
cd cyclecloud-pbspro

# Optional, but recommended. Adds relevant resources and enables strict placement
./initialize_pbs.sh

# Optional. Sets up workq as a colocated, MPI focused queue and creates htcq for non-MPI workloads.
./initialize_default_queues.sh

# Creates the azpbs autoscaler
./install.sh  --venv /opt/cycle/pbspro/venv

# Otherwise insert your username, password, url, and cluster name here.
./generate_autoscale_json.sh --install-dir /opt/cycle/pbspro \
                             --username user \
                             --password password \
                             --url https://fqdn:port \
                             --cluster-name cluster_name

azpbs validate

CycleCloud supports a standard set of autostop attributes across schedulers:

Attribute	Description
cyclecloud.cluster.autoscale.stop_enabled	Is autostop enabled on this node? [true/false]
cyclecloud.cluster.autoscale.idle_time_after_jobs	The amount of time (in seconds) for a node to sit idle after completing jobs before it is scaled down.
cyclecloud.cluster.autoscale.idle_time_before_jobs	The amount of time (in seconds) for a node to sit idle before completing jobs before it is scaled down.

Note

CycleCloud does not support the bursting configuration with Open PBS.

Note

Even though Windows is an officially supported Open PBS platform, CycleCloud does not support running Open PBS on Windows at this time.