Skip to content

Cron Workflows

v2.5 and after

Introduction

CronWorkflow are workflows that run on a preset schedule. They are designed to be converted from Workflow easily and to mimic the same options as Kubernetes CronJob. In essence, CronWorkflow = Workflow + some specific cron options.

CronWorkflow Spec

An example CronWorkflow spec would look like:

apiVersion: argoproj.io/v1alpha1
kind: CronWorkflow
metadata:
  name: test-cron-wf
spec:
  schedule: "* * * * *"
  concurrencyPolicy: "Replace"
  startingDeadlineSeconds: 0
  workflowSpec:
    entrypoint: whalesay
    templates:
    - name: whalesay
      container:
        image: alpine:3.6
        command: [sh, -c]
        args: ["date; sleep 90"]

workflowSpec and workflowMetadata

CronWorkflow.spec.workflowSpec is the same type as Workflow.spec and serves as a template for Workflow objects that are created from it. Everything under this spec will be converted to a Workflow.

The resulting Workflow name will be a generated name based on the CronWorkflow name. In this example it could be something like test-cron-wf-tj6fe.

CronWorkflow.spec.workflowMetadata can be used to add labels and annotations.

CronWorkflow Options

Option Name Default Value Description
schedule None, must be provided Schedule at which the Workflow will be run. E.g. 5 4 * * *
timezone Machine timezone Timezone during which the Workflow will be run from the IANA timezone standard, e.g. America/Los_Angeles
suspend false If true Workflow scheduling will not occur. Can be set from the CLI, GitOps, or directly
concurrencyPolicy Allow Policy that determines what to do if multiple Workflows are scheduled at the same time. Available options: Allow: allow all, Replace: remove all old before scheduling a new, Forbid: do not allow any new while there are old
startingDeadlineSeconds 0 Number of seconds after the last successful run during which a missed Workflow will be run
successfulJobsHistoryLimit 3 Number of successful Workflows that will be persisted at a time
failedJobsHistoryLimit 1 Number of failed Workflows that will be persisted at a time

Cron Schedule Syntax

The cron scheduler uses the standard cron syntax, as documented on Wikipedia.

More detailed documentation for the specific library used is documented here.

Crash Recovery

If the workflow-controller crashes (and hence the CronWorkflow controller), there are some options you can set to ensure that CronWorkflows that would have been scheduled while the controller was down can still run. Mainly startingDeadlineSeconds can be set to specify the maximum number of seconds past the last successful run of a CronWorkflow during which a missed run will still be executed.

For example, if a CronWorkflow that runs every minute is last run at 12:05:00, and the controller crashes between 12:05:55 and 12:06:05, then the expected execution time of 12:06:00 would be missed. However, if startingDeadlineSeconds is set to a value greater than 65 (the amount of time passing between the last scheduled run time of 12:05:00 and the current controller restart time of 12:06:05), then a single instance of the CronWorkflow will be executed exactly at 12:06:05.

Currently only a single instance will be executed as a result of setting startingDeadlineSeconds.

This setting can also be configured in tandem with concurrencyPolicy to achieve more fine-tuned control.

Daylight Saving

Daylight Saving (DST) is taken into account when using timezone. This means that, depending on the local time of the scheduled job, argo will schedule the workflow once, twice, or not at all when the clock moves forward or back.

For example, with timezone set at America/Los_Angeles, we have daylight saving

  • +1 hour (DST start) at 2020-03-08 02:00:00:

    Note: The schedules between 02:00 a.m. to 02:59 a.m. were skipped on Mar 8th due to the clock being moved forward:

    cron sequence workflow execution time
    59 1 ** * 1 2020-03-08 01:59:00 -0800 PST
    2 2020-03-09 01:59:00 -0700 PDT
    3 2020-03-10 01:59:00 -0700 PDT
    0 2 ** * 1 2020-03-09 02:00:00 -0700 PDT
    2 2020-03-10 02:00:00 -0700 PDT
    3 2020-03-11 02:00:00 -0700 PDT
    1 2 ** * 1 2020-03-09 02:01:00 -0700 PDT
    2 2020-03-10 02:01:00 -0700 PDT
    3 2020-03-11 02:01:00 -0700 PDT
  • -1 hour (DST end) at 2020-11-01 02:00:00:

    Note: the schedules between 01:00 a.m. to 01:59 a.m. were triggered twice on Nov 1st due to the clock being set back:

    cron sequence workflow execution time
    59 1 ** * 1 2020-11-01 01:59:00 -0700 PDT
    2 2020-11-01 01:59:00 -0800 PST
    3 2020-11-02 01:59:00 -0800 PST
    0 2 ** * 1 2020-11-01 02:00:00 -0800 PST
    2 2020-11-02 02:00:00 -0800 PST
    3 2020-11-03 02:00:00 -0800 PST
    1 2 ** * 1 2020-11-01 02:01:00 -0800 PST
    2 2020-11-02 02:01:00 -0800 PST
    3 2020-11-03 02:01:00 -0800 PST

Managing CronWorkflow

CLI

CronWorkflow can be created from the CLI by using basic commands:

$ argo cron create cron.yaml
Name:                          test-cron-wf
Namespace:                     argo
Created:                       Mon Nov 18 10:17:06 -0800 (now)
Schedule:                      * * * * *
Suspended:                     false
StartingDeadlineSeconds:       0
ConcurrencyPolicy:             Forbid

$ argo cron list
NAME           AGE   LAST RUN   SCHEDULE    SUSPENDED
test-cron-wf   49s   N/A        * * * * *   false

# some time passes

$ argo cron list
NAME           AGE   LAST RUN   SCHEDULE    SUSPENDED
test-cron-wf   56s   2s         * * * * *   false

$ argo cron get test-cron-wf
Name:                          test-cron-wf
Namespace:                     argo
Created:                       Wed Oct 28 07:19:02 -0600 (23 hours ago)
Schedule:                      * * * * *
Suspended:                     false
StartingDeadlineSeconds:       0
ConcurrencyPolicy:             Replace
LastScheduledTime:             Thu Oct 29 06:51:00 -0600 (11 minutes ago)
NextScheduledTime:             Thu Oct 29 13:03:00 +0000 (32 seconds from now)
Active Workflows:              test-cron-wf-rt4nf

Note: NextScheduledRun assumes that the workflow-controller uses UTC as its timezone

kubectl

Using kubectl apply -f and kubectl get cwf

Back-Filling Days

See cron backfill.

GitOps via Argo CD

CronWorkflow resources can be managed with GitOps by using Argo CD

UI

CronWorkflow resources can also be managed by the UI

Comments