Skip to content

Workloads

What a deployed workload is, what you can change after it is running, and what is fixed on purpose.

A workload is a container deployment that ALYNREX manages for you inside one of your project’s namespaces. It carries an image, a replica count, a container port, and resource requests and limits.

Templates decide the shape

Most workloads are created from a template rather than from a blank form. The template fixes the image, the replicas, the port, the environment and the resource caps.

That is a deliberate trade. It means deploying is a choice between known-good shapes instead of thirty fields where a wrong resource limit is an outage or an unexpected bill. It also means the things that move cost are decided once, by someone who is accountable for them, rather than per-deploy.

What you can do to a running workload

Stop it. The workload scales to zero and stops consuming compute. Its namespace, configuration and any attached storage stay exactly as they are.

Resume it. It comes back at the replica count it had.

Restart it. The pods are replaced without changing anything about the workload’s definition — the usual fix for a process that has wedged.

Delete it. This removes the deployment. Read the confirmation: deleting is not the same as stopping, and stopping is what you want if you intend to come back.

Any project developer or admin can do all four.

What you cannot change

The image, the replica count and the resource limits are not editable from a project role. See Projects and roles for why.

Storage

A workload can mount object storage from its project. The bucket belongs to the project, not to the workload, so it survives the workload being deleted and is shared by anything else in the same project that mounts it.

Credentials for that mount are managed for you and rotated without interrupting running workloads — a pod that mounted before a rotation keeps working on the credential it holds until it restarts.

When something is wrong

The workload’s page shows pod status and logs. Two failures account for most of what you will see:

  • The image cannot be pulled. Usually a private registry with no credentials configured, or a tag that does not exist.
  • The pod cannot be scheduled. The namespace has a quota, and a workload that does not fit in what remains will wait rather than evict something.

Every change to your project is recorded in its audit log, including failures and who attempted them. That is usually a faster answer than asking.