Files
nomad/website/content/docs/concepts/stateful-deployments.mdx

86 lines
3.6 KiB
Plaintext

---
layout: docs
page_title: Stateful deployments
description: |-
Learn how Nomad handles stateful deployments. Use a dynamic host volume for your stateful workload and bind an allocation to a volume ID with the sticky parameter. Learn how Nomad scales jobs with sticky volumes.
---
# Stateful deployments
<Note>
Stateful deployments support only dynamic host volumes. For CSI volumes, use the
[`per_alloc` property](/nomad/docs/job-specification/volume#per_alloc),
which serves a similar purpose.
</Note>
Nomad stateful deployments let you effectively bind your jobs to host [volumes].
On a lower level, this means taking Nomad's basic unit of workload placement,
the [allocation], and making it possible for you to have allocations that are
[sticky] to volume IDs. Since stateful deployments only support dynamic host
volumes, using them effectively means binding a job to a particular node.
Stateful deployments work on a per-task-group basis. Use the `volume.sticky`
parameter in your job specification to indicate that each task group requires a
sticky host volume.
This example sets the volume's `sticky` parameter to `true`.
```hcl
job "app" {
group "example" {
# ...
volume "example" {
type = "host"
source = "ca-certificates"
read_only = true
sticky = true
}
# ...
}
}
```
If you set the volume to sticky, during the deployment the scheduler associates
its ID with the task group that uses it. Stateful deployments require that any
allocations belonging to that task group must be places or replaced on nodes
that have this volume ID available.
Scaling jobs with sticky volumes up results in more volume IDs claimed by the
task group, but scaling jobs down does not delete unused volumes, nor does it
touch the data present on them.
If the scheduler cannot find a node that has the right volume ID present,
perhaps because the node is down or disconnected, Nomad creates a blocked
evaluation.
## Task group host volume claims
Nomad uses its state store to save associations of host volume IDs and their
task groups, as well as job IDs, volume names, and namespaces. Use the API to
[list] and [delete] these associations. If you delete a task group host volume
claim, Nomad assigns a different volume ID to the task group, assuming the old
volume ID is unavailable and another feasible host volume with matching name is.
However, after placement, Nomad once again records the claim in its state store.
You should only use claim deletions in operational emergency situations or while
debugging problems with nodes. For example, use claim deletions in situations
where you want to drain a node with a stateful deployment on it, but do not want
to stop the job.
## Related resources
Refer to the following Nomad pages for more information about stateful workloads
and volumes:
- [Considerations for Stateful Workloads](/nomad/docs/operations/stateful-workloads) explores the options for persistent storage of workloads running in Nomad.
- The [Nomad volume specification][volumes] defines the schema for creating and registering volumes.
- The [job specification `volume` block](/nomad/docs/job-specification/volume) lets you configure a group that requires a specific volume from the cluster.
- The [Stateful Workloads](/nomad/tutorials/stateful-workloads) tutorials explore techniques to run jobs that require access to persistent storage.
[allocation]: /nomad/docs/glossary#allocation
[delete]: /nomad/api-docs/volumes#delete-task-group-host-volume-claims
[list]: /nomad/api-docs/volumes#list-task-group-host-volume-claims
[sticky]: /nomad/docs/job-specification/volume#sticky
[volumes]: /nomad/docs/other-specifications/volume