open-nomad/website/source/docs/job-specification/update.html.md
Alex Dadgar 47379d5eea Document update stanza for system scheduler.
This PR adds a note to make it clear the limits of the system
schedulers update stanza.
2017-08-01 08:51:26 -07:00

7.3 KiB

layout page_title sidebar_current description
docs update Stanza - Job Specification docs-job-specification-update The "update" stanza specifies the group's update strategy. The update strategy is used to control things like rolling upgrades and canary deployments. If omitted, rolling updates and canaries are disabled.

update Stanza

Placement job -> **update** job -> group -> **update**

The update stanza specifies the group's update strategy. The update strategy is used to control things like rolling upgrades and canary deployments. If omitted, rolling updates and canaries are disabled. If specified at the job level, the configuration will apply to all groups within the job. If multiple update stanzas are specified, they are merged with the group stanza taking the highest precedence and then the job.

job "docs" {
  update {
    max_parallel     = 3
    health_check     = "checks"
    min_healthy_time = "10s"
    healthy_deadline = "10m"
    auto_revert      = true
    canary           = 1
    stagger          = "30s"
  }
}

~> For system jobs, only max_parallel and stagger are enforced. The job is updated at a rate of max_parallel, waiting stagger duration before the next set of updates. The system scheduler will be updated to support the new update stanza in a future release.

update Parameters

  • max_parallel (int: 0) - Specifies the number of task groups that can be updated at the same time.

  • health_check (string: "checks") - Specifies the mechanism in which allocations health is determined. The potential values are:

    • "checks" - Specifies that the allocation should be considered healthy when all of its tasks are running and their associated checks are healthy, and unhealthy if any of the tasks fail or not all checks become healthy. This is a superset of "task_states" mode.

    • "task_states" - Specifies that the allocation should be considered healthy when all its tasks are running and unhealthy if tasks fail.

    • "manual" - Specifies that Nomad should not automatically determine health and that the operator will specify allocation health using the HTTP API.

  • min_healthy_time (string: "10s") - Specifies the minimum time the allocation must be in the healthy state before it is marked as healthy and unblocks further allocations from being updated. This is specified using a label suffix like "30s" or "15m".

  • healthy_deadline (string: "5m") - Specifies the deadline in which the allocation must be marked as healthy after which the allocation is automatically transitioned to unhealthy. This is specified using a label suffix like "2m" or "1h".

  • auto_revert (bool: false) - Specifies if the job should auto-revert to the last stable job on deployment failure. A job is marked as stable if all the allocations as part of its deployment were marked healthy.

  • canary (int: 0) - Specifies that changes to the job that would result in destructive updates should create the specified number of canaries without stopping any previous allocations. Once the operator determines the canaries are healthy, they can be promoted which unblocks a rolling update of the remaining allocations at a rate of max_parallel.

  • stagger (string: "30s") - Specifies the delay between migrating allocations off nodes marked for draining. This is specified using a label suffix like "30s" or "1h".

update Examples

The following examples only show the update stanzas. Remember that the update stanza is only valid in the placements listed above.

Parallel Upgrades Based on Checks

This example performs 3 upgrades at a time and requires the allocations be healthy for a minimum of 30 seconds before continuing the rolling upgrade. Each allocation is given at most 2 minutes to determine its health before it is automatically marked unhealthy and the deployment is failed.

update {
  max_parallel     = 3
  min_healthy_time = "30s"
  healthy_deadline = "2m"
}

Parallel Upgrades Based on Task State

This example is the same as the last but only requires the tasks to be healthy and does not require registered service checks to be healthy.

update {
  max_parallel     = 3
  min_healthy_time = "30s"
  healthy_deadline = "2m"
  health_check     = "task_states"
}

Canary Upgrades

This example creates a canary allocation when the job is updated. The canary is created without stopping any previous allocations from the job and allows operators to determine if the new version of the job should be rolled out.

update {
  canary       = 1
  max_parallel = 3
}

Once the operator has determined the new job should be deployed, the deployment can be promoted and a rolling update will occur performing 3 updates at a time until the remainder of the groups allocations have been rolled to the new version.

# Promote the canaries for the job.
$ nomad job promote <job-id>

Blue/Green Upgrades

By setting the canary count equal to that of the task group, blue/green deployments can be achieved. When a new version of the job is submitted, instead of doing a rolling upgrade of the existing allocations, the new version of the group is deployed along side the existing set. While this duplicates the resources required during the upgrade process, it allows very safe deployments as the original version of the group is untouched.

group "api-server" {
    count = 3

    update {
      canary       = 3
      max_parallel = 3
    }
    ...
}

Once the operator is satisfied that the new version of the group is stable, the group can be promoted which will result in all allocations for the old versions of the group to be shutdown. This completes the upgrade from blue to green, or old to new version.

# Promote the canaries for the job.
$ nomad job promote <job-id>

Serial Upgrades

This example uses a serial upgrade strategy, meaning exactly one task group will be updated at a time. The allocation must be healthy for the default min_healthy_time of 10 seconds.

update {
  max_parallel = 1
}

Upgrade Stanza Inheritance

This example shows how inheritance can simplify the job when there are multiple task groups.

job "example" {
  ...

  update {
    max_parallel     = 2
    health_check     = "task_states"
    healthy_deadline = "10m"
  }

  group "one" {
    ...

    update {
      canary = 1      
    }
  }

  group "two" {
    ...

    update {
      min_healthy_time = "3m" 
    }
  }
}

By placing the shared parameters in the job's update stanza, each groups update stanza may be kept to a minimum. The merged update stanzas for each group becomes:

group "one" {
  update {
    canary           = 1
    max_parallel     = 2
    health_check     = "task_states"
    healthy_deadline = "10m"
  }
}

group "two" {
  update {
    min_healthy_time = "3m" 
    max_parallel     = 2
    health_check     = "task_states"
    healthy_deadline = "10m"
  }
}