Skip to content

Environments and drift

In Seylo an environment is a cluster and namespace pair, such as shop-qa on acme-staging or shop on acme-prod. Two environments can live in one cluster or in two. This page covers three questions: what runs where, how two environments differ, and how to move a build from one to the next.

Operate → Overview is a matrix of every cluster and namespace, with the image tag each component is running right now. It answers “is 1.4.3 in prod yet?” at a glance. Click a version to open that namespace and change it.

The matrix compares image tags only. For configuration, use the comparison below.

What you need: viewer. Copying keys needs releaser on the destination.

  1. Open the comparison.

    Configure → Environment, then Compare across namespaces / clusters.

  2. Choose the component and the targets.

    Type the component name (checkout-api). Each target is a cluster and a namespace. You start with two. Add column adds more, up to 8.

  3. Press Compare.

Each environment variable is a row, with one column per target and a status:

Status Meaning
same Every target has the same value
differs The values differ
missing At least one target does not declare the variable at all
not compared There is no value Seylo can compare. The row says why

A row is not compared for one of three reasons:

  • The value comes from a Secret, through secretKeyRef or envFrom.secretRef. That includes values synced from a key vault by the Secrets Store CSI driver. The row is listed so you can see the variable exists, but Seylo never reads, compares or copies a Secret value.
  • A ConfigMap reference did not resolve. The variable is declared, but the ConfigMap or key it points at is missing, so there is no value. This is reported separately from missing. Two unresolved references are never treated as equal. If the reference is not marked optional, the row says that container would fail to start.
  • The value comes from the downward API or another field reference, which has no literal value.

Show only differences hides the rows that are the same. A target where the component is not deployed is named at the top of the result (Component "checkout-api" is not deployed in acme-staging/shop-qa.) rather than shown as a column of blanks.

The comparison reads the live environment of the component’s first container: inline values, values resolved from ConfigMap keys, and variables expanded from envFrom ConfigMaps.

Tick the rows you want, then use Copy selected keys at the bottom: choose From and To, and press Copy N keys. This creates an ordinary Env edit request against the destination, which goes through approval and is recorded like any other change.

How each key is copied:

Source of the value What is written to the destination
Inline value The same inline value
ConfigMap key that exists in the destination namespace The same ConfigMap reference
ConfigMap key that does not exist in the destination The resolved value, written inline
Variable expanded from envFrom An explicit env entry, which takes precedence over envFrom in the destination
Secret, downward API, or an unresolved reference Not copied. The result lists it as excluded, with the reason

This copies environment variables of one component into an environment that already runs it. To clone a whole namespace into a new one, with its ConfigMaps, Secrets and everything its workloads reference, use copy a namespace instead.

Variables you did not select are left as they are in the destination. The outcome is shown under the comparison: Submitted for approval: … when it went to the queue, or Copied … to … when approvals are off.

If Argo CD, Flux or Helm owns the destination Deployment, the copy is refused and nothing is applied. The page shows This copy was REFUSED: another controller owns the target, naming the owner and the evidence. The durable fix is to change the variables where that controller reads them. To copy through Seylo anyway, tick the box under the refusal that starts “I understand this change will be reverted by the controller above…” (“…overwritten by that owner” for Helm) and press Copy N keys again. The acknowledgement and your name are recorded on the request. Choosing a different From or To clears the refusal and the acknowledgement.

Seylo uses “drift” for two different things, and keeps them apart:

  • Between environments. shop-qa and shop disagree. That is often deliberate, such as different replica counts or log levels. The comparison above shows it. It is never “fixed” for you.
  • Between Seylo and the cluster. Seylo wrote a value and the live object now holds a different one. It is caught on every write by the read-back (drifted), and on demand for ConfigMaps with Check drift. See how drift is detected.

Promotion moves what one environment is running into the next one, and records where it came from. Open it with the Promote button on Overview or on a namespace’s page; it needs releaser on the target and viewer on the source.

An administrator first defines a path: your environments in order, such as dev → qa → prod. Press Set up: define a path, name it, and add each stage with a name, a cluster and a namespace. A stage is identified by its cluster and namespace, not by its name, so two stages called prod on different clusters are two different places. A path needs at least two stages.

The order is the point. It is what lets Seylo tell a promotion from a rollback, and a skipped stage from an orderly one.

  1. Choose the path, From and To.

    Choosing changes nothing. Seylo works out what a promotion would do and shows it.

  2. Read What moves.

    Each component the target already runs is listed with a verdict and a sentence explaining it:

    Verdict Meaning
    promotes It will move to the tag the source is running
    already The target already runs the same image reference
    not-in-source The source is not running it, so there is nothing to promote
    not-in-target Only the source has it. Promotion does not create components
    source-unhealthy The source has it but could not run it
  3. Read What will still be different.

    A promotion moves the image tag only. The target keeps its own replicas, resources, environment, ConfigMaps and ingress. This card lists the differences that stay. It compares Seylo’s recorded Deployment in each environment: replica count, CPU and memory requests and limits, and which environment variables exist. Variable values are not read. ConfigMaps, Secrets, Services and Ingresses are not compared here, and the card says so. If a Deployment could not be read, the answer is reported as unknown, not as “no differences”.

  4. Press Promote into prod.

    The promotion is an ordinary release into the target, so it goes through approval, the ownership check and the rollout watch like any other. The plan is worked out again at the moment you press the button, so a source that was redeployed after you opened the page is not promoted unseen.

What Seylo refuses, or makes you acknowledge:

Situation What happens
The target comes before the source on the path Refused: … Moving prod's image into an earlier environment is a rollback of …, not a promotion …
The two places are not stages of one path, or are the same stage Refused
The promotion skips a stage, such as dev → prod Allowed after you tick Promote anyway, and record that the stages above were skipped. The skip is recorded
The source has the image but could not run it (degraded, no ready replica, or scaled to zero) Allowed after you tick Promote anyway, and record that qa could not run it. Recorded
Nothing would change The button stays disabled and says why
Argo CD, Flux or Helm owns a target Deployment Refused: This promotion was REFUSED: another controller owns the target, naming the owner. Allowed after you tick the box that starts “I understand this change will be reverted by the controller above…” and press Promote into prod again. Recorded. Choosing another path, From or To clears it

How offers two more modes besides One step:

  • Along the path (one verified hop at a time). Each hop starts only after the previous one is verified: approved, rolled out, and the target healthy on the promoted tags. The next hop then promotes from the stage just verified.
  • To several stages at once. One source to several later stages, for example two production regions. Each step stands alone. One failing does not stop the others.

Every hop is the same promotion you could have clicked yourself, so with approvals on, each hop waits for an approver.

These two modes offer no acknowledgement for a reconciler-owned target. If Argo CD, Flux or Helm owns a Deployment in a stage, that hop shows the refusal as plain text. Change it at the source, or promote that stage with One step, which offers the acknowledgement.

The card What arrived in prod by promotion lists past promotions into the selected target.

You see What it means What to do
At most 8 targets can be compared at once. You added more than 8 columns Remove some
Component "checkout-api" is not deployed in acme-staging/shop-qa. The component has no Deployment in that target Check the namespace, or the component’s name there
Nothing to copy: every requested key was excluded (…) Every selected key was Secret-sourced, a field reference, or unresolved Secret values must be changed where they live, such as the key vault
The source and destination targets must differ. From and To are the same target Choose another target
No promotion path yet No path is defined An administrator defines one
… is not running in prod. A promotion updates what is already there; … The component does not exist in the target Add it there first, as a separate change with its own approval
… comes BEFORE … on this path. … Use rollback instead. You chose a target earlier on the path See rollback
This promotion was REFUSED: another controller owns the target or This copy was REFUSED: another controller owns the target A controller owns the target Deployment. Nothing was applied Change it at that controller’s source, or tick the acknowledgement and press Promote into … or Copy N keys again
Refusing to change Deployment shop/checkout-api: … A controller owns the target Deployment. On a path or several-stages run this is the only text shown Change it at that controller’s source, or promote that stage with One step and acknowledge the owner
  • Copy a namespace: clone a whole environment into a new namespace or onto another cluster