Skip to content

Your first deployment

This page walks one change end to end. The example moves checkout-api in namespace shop on cluster acme-prod from 1.4.2 to 1.4.3. It covers choosing the tag, the approval, and what Seylo reports while Kubernetes rolls the change out.

What you need: releaser for the namespace, a cluster that is connected (see install the agent), and someone who holds approver or admin. That can be you. Self-approval is allowed, as described below.

  1. Open the namespace.

    Operate → Overview shows every cluster and namespace with the image tag each component is running. Click the version for checkout-api in shop to open the namespace page.

    Each row shows the current tag, a rollout status dot and an owner badge. What the dots and owner badges mean, at the bottom of the card, explains both.

  2. Tick the component and choose the tag.

    The field to the right of the current tag takes the tag to deploy. If Seylo can list your registry, you pick from its tags. On an agent-managed cluster it usually cannot, because Seylo holds no cloud or registry credential for that cluster. The page then says so once and you type the tag (1.4.3). Nothing checks the tag against the registry at this point. The rollout is what proves the tag exists.

  3. Optionally, change configuration in the same release.

    Ticking a component opens Also change configuration for checkout-api under its row. It is off by default. Turn it on when the new version needs an environment variable or different CPU/memory requests and limits. Seylo then applies the image and the configuration together, as one apply and one rollout, so the new image never runs against the old configuration.

    • The environment grid is pre-filled from the cluster and is the complete set of inline and ConfigMap-sourced variables. Removing a row removes the variable. Secret-sourced variables are kept and never shown.
    • Resource fields: blank leaves a field untouched, a single - removes it.
    • If the pod has more than one container, for example a service-mesh sidecar, you must choose the container. Seylo does not guess.
  4. Press Deploy selected (1).

    The Confirm release dialog lists every component as old → new, and names any configuration change too. Add release notes if you like, then press Deploy. You can tick several components and release them together as one change.

When approvals are enabled, which is the default once you sign in through an identity provider, nothing touches the cluster yet. The page shows Submitted for approval, with the request’s one-line summary and a link to the queue:

release(shop): checkout-api -> 1.4.3

An approver opens Operate → Approvals and approves or rejects it. Approvals covers what they are shown and what is re-checked.

If your organisation has approvals turned off, the change applies when you press Deploy. It is still recorded in the component’s history and in the audit trail, and the page shows the result straight away.

Once the change is approved, Seylo follows the rollout. Expand the request in the approvals queue (Detail) and look at Live rollout, or open it in Operate → Activity. Each component gets a chip:

Chip Meaning
Queued Nothing has been read for this component yet
Deploying New pods are coming up. The replica count shows progress, such as 2/3 ready
Deployed ✓ The rollout reached a healthy state
Degraded The rollout ended in a degraded state
Failed ✗ The rollout failed for a reason Kubernetes will not fix by retrying

Accepted, verified and rolled out are three different things

Section titled “Accepted, verified and rolled out are three different things”

Seylo does not report success just because the Kubernetes API accepted the write. It checks three things in order:

  1. Applied. The API server accepted the change.
  2. Verified. Seylo read the Deployment back and the container really names 1.4.3, plus any configuration you set. A webhook or a GitOps controller can rewrite a change straight after a successful write, so this is checked every time.
  3. Rolled out. The new pods became ready.

The operation in Activity ends in one of these states:

Status Meaning
Succeeded Applied, verified, and the rollout reached a healthy state
Partial Some components applied and some did not, or the rollout was still in progress when Seylo stopped watching
Failed The cluster rejected the change, or the rollout failed for an actionable reason
Reverted (drift) The API server accepted the write, but reading it back shows a different value. Something else owns that field

If Seylo could not read a component back, it shows verification indeterminate (or unknown). That is never counted as verified.

A failure names the pod, the container and Kubernetes’ own message, not just a red dot. A tag that does not exist in the registry looks like this:

Rollout FAILED: ImagePullBackOff on pod checkout-api-7d9f8-abcde (container checkout-api):
Back-off pulling image "registry.acme.example/shop/checkout-api:1.4.3".
The manifests were applied, but the new pods are not running. Kubernetes will keep
retrying this forever until the underlying cause is fixed.

Seylo marks the rollout failed as soon as it sees one of these reasons on a pod, because Kubernetes retries them forever: ImagePullBackOff, ErrImagePull, ErrImageNeverPull, InvalidImageName, CrashLoopBackOff, CreateContainerConfigError, CreateContainerError and RunContainerError. A Deployment that runs past its progress deadline is reported as ProgressDeadlineExceeded.

To undo a release, see rollback.

What is checked before anything is applied

Section titled “What is checked before anything is applied”
  • Your access. You need releaser in that namespace on that cluster, and the namespace must be inside the cluster’s namespace policy.
  • The tag format. A tag is 1 to 128 letters, numbers, _, . or -.
  • The component. It must be deployed in the namespace, and can appear only once in a release.
  • Who owns the Deployment. See the next section.

An image-tag release is not dry-run against the API server first. The server-side dry run is part of the manifest editor. For an image change, the read-back and the rollout are the checks.

Seylo records the change as a new version in the component’s history before it applies it. If the apply fails, that version stays as the record of what was requested, and the result shows what actually landed.

Releases into the same namespace are applied one at a time, so two people deploying at once cannot interleave their writes.

If Argo CD, Flux or Helm owns the Deployment

Section titled “If Argo CD, Flux or Helm owns the Deployment”

Seylo checks who owns every object it is about to write. If Argo CD, Flux, a Helm release or another controller owns checkout-api, the release is refused and nothing is applied:

This release was REFUSED: another controller owns these Deployments

The card names the owner, the evidence (the label, annotation or field manager that identified it), and what will happen next:

  • Argo CD and Flux revert your change at their next sync.
  • Helm does not re-sync, but the next helm upgrade overwrites your change.

Usually the right fix is to change the tag where that controller reads it, in the Git repository or the chart values. To deploy through Seylo anyway, tick the box that starts “I understand this change will be reverted by the controller above…” (“…overwritten by that owner” for Helm) and press Deploy selected again. Your acknowledgement and your name are stored on the request, shown to the approver, and written to the audit log.

You see What it means What to do
Seylo holds no credential for … Type the tag instead The cluster is agent-managed, so its registry cannot be listed from the portal Type the tag. The rollout verifies it
changes[0].newTag must be a valid OCI tag (1-128 letters, numbers, _, ., or -). The tag contains a character a tag cannot have Check for a stray space, : or @
Refusing to change Deployment shop/checkout-api: … A controller owns the Deployment Change it at the source, or acknowledge as described above
… has 2 containers (…), so "container" must be specified for an env/resources change You attached configuration to a multi-container pod Choose the container in the configuration panel
This needs the releaser role, which … does not hold. You do not hold releaser Ask an administrator
… holds releaser, but not in "shop" on cluster "acme-prod". Your role is granted for other namespaces only Ask an administrator to extend the grant
Cluster "acme-prod" is locked READ-ONLY … Nothing writes to this cluster through Seylo, whatever your role An administrator clears the lock on the cluster
Rollout FAILED: ImagePullBackOff … The tag does not exist, or the nodes cannot pull from the registry Check the tag and the cluster’s pull access
Rollout FAILED: CreateContainerConfigError … The pod references a ConfigMap or Secret key that is missing The message names the missing object
Rollout FAILED: Kubernetes gave up on checkout-api (ProgressDeadlineExceeded) Kubernetes stopped waiting for the rollout Check the pods’ events and readiness probes
Applied, but the rollout was STILL PROGRESSING when the …s watch timeout expired The change is on the cluster, but the pods were not ready when Seylo stopped watching Check the rollout status on the namespace page. Reported as Partial
… was accepted by the API server but reading it back does NOT match what we set Something rewrote the change straight after it was applied Usually a GitOps controller, Helm, or a mutating webhook. Change it at that source