This document is meant to be a guide for new users of how to create and submit Jobs to Armada. Because Armada jobs are run as containers in Kubernetes, a basic understanding of Docker and Kubernetes is required; see, e.g., the following links:
For more information about the design of Armada (e.g., how jobs are prioritised), see:
The Armada workflow is:
- Create a job specification, which is a Kubernetes podspec with additional metadata fields specific to Armada.
- Submit the job specification to one of Armada’s job queues.
To create a job that sleeps for 60 seconds, start by creating a yaml file with the following contents:
queue: test jobSetId: set1 jobs: - priority: 0 podSpecs: - terminationGracePeriodSeconds: 0 restartPolicy: Never containers: - name: sleep imagePullPolicy: IfNotPresent image: busybox:latest args: - sleep - 60s resources: limits: memory: 64Mi cpu: 150m requests: memory: 64Mi cpu: 150m
In the above yaml snippet,
podSpec is a Kubernetes podspec, which consists of one or more containers that contain the user code to be run. In addition, the job specification (jobspec) contains metadata fields specific to Armada:
queue: which of the available job queues the job should be submitted to.
priority: the job priority (lower values indicate higher priority).
jobSetId: jobs with the same
jobSetIdcan be followed and cancelled in a single operation. The
jobSetIdhas no impact on scheduling.
Resource requests and limits must be equal. Armada does not yet support limit > request.
Now, the job can be submitted to Armada using the
armadactl command-line utility (or alternatively via Armada’s gRPC or REST API). In particular, run
armadactl submit <jobspec.yaml>,
<jobspec.yaml> is the path of the file containing the jobspec. Armada automatically handles creating and running the necessary containers.
All containers part of the same podspec will be run on the same node (a physical or virtual machine), i.e., the amount of CPU and memory the above job could consume is limited by what the nodes that make up the clusters are equipped with. To overcome this limitation, jobs need to specify several podspecs, each of which may be scheduled on different nodes within a Kubernetes cluster. For example:
queue: test jobSetId: set1 jobs: - priority: 0 podSpecs: - terminationGracePeriodSeconds: 0 restartPolicy: Never containers: - name: sleep imagePullPolicy: IfNotPresent image: busybox:latest args: - sleep - 60s resources: limits: memory: 64Mi cpu: 150m requests: memory: 64Mi cpu: 150m - terminationGracePeriodSeconds: 0 restartPolicy: Never containers: ...
All pods part of the same job will be run simultaneously and within a single Kubernetes cluster. If any of the pods that make up the job fails to start within a certain timeout (specified by the
stuckPodExpiry parameter), the entire job is cancelled and all pods belonging to it removed. Armada may attempt to re-schedule jobs experiencing transient issues. Events relating to a multi-node job contain a
podNumber identifier, corresponding to the index of pod in the
podSpecs list that the event refers to.
Jobs are submitted using either the
armadactl command-line utility, with
armadactl submit <jobspec.yaml>, or using the gRPC or REST API.
Here, we give a complete example of an Armada jobspec with all available parameters.
queue: example # 1. jobSetId: test # 2. jobs: - priority: 1000 # 3. namespace: example # 4. clientId: 12345 # 5. labels: # 6. example-label: "value" annotations: # 7. example-annotation: "value" ingress: # 8. - type: NodePort ports: - 5050 podSpecs: # 9. - containers: name: app imagePullPolicy: IfNotPresent image: vad1mo/hello-world-rest:latest securityContext: runAsUser: 1000 resources: limits: memory: 1Gi cpu: 1 requests: memory: 1Gi cpu: 1 ports: - containerPort: 5050 protocol: TCP name: http
Using this format, it is possible to submit a job set composed of several jobs. The meaning of each field is:
- The queue this job will be submitted to.
- Name of the job set this job belongs to.
- Relative priority of the job.
- The namespace that the pods part of this job will be created in (the
defaultnamespace if not specified).
- An optional ID that can be set to ensure that jobs are not duplicated, e.g., in case of certain network failures. Armada automatically discards any jobs submitted with a
clientIdequal to that of an existing job.
- List of labels that are added to all pods created as part of this job..
- List annotations that are added to all pods created as part of this job.
- List of ports that are exposed with the specified ingress type. The ingress only exposes ports for pods that also expose the corresponding port via the
- List of podspecs that make up the job; see the Kubernetes documentation for an overview of the available parameters.