Skip to content

Scheduled deployments

A scheduled deploy is the same release as your first deployment, run later and unattended: once at a date and time, or on a recurring maintenance window. Because nobody is there when it fires, approval happens when you create the schedule, not when it runs.

What you need: releaser for the namespace.

  1. Select what to deploy.

    On the namespace page, tick the components and choose their target tags, exactly as for an immediate release.

  2. Press Schedule deploy (N).

    The button is disabled until at least one component has a tag. It is also disabled if any selected component has Also change configuration turned on. A schedule carries image tags only. Deploy a configuration change now, or turn the configuration off to schedule the tag alone.

  3. Choose when.

    • When: Once, at a specific date and time, or Recurring maintenance window.
    • Timezone (IANA): required, chosen from a list. Your browser’s zone comes first, then a short list of common zones, then every zone, each shown with its current offset from UTC. Seylo never assumes the server’s timezone.
    • For a one-off: Date and Time (local to the zone above).
    • For a recurring window: Time (local to the zone above) and Days. No day selected means every day. Stored as shows the stored form, <minute> <hour> * * <day-of-week> with 0 for Sunday. Only minute, hour and day of week can be expressed. There is no “the 31st of the month”.
    • Missed-window catch-up grace (seconds): see missed windows. The dialog starts at 900. The allowed range is 0 to 86400.
  4. Check Next runs.

    The dialog lists the next run times in the zone you chose, worked out by the same code the scheduler uses. This list is your confirmation. If a time falls on a daylight-saving change, a note explains what will happen.

  5. Press Submit schedule for approval.

The schedule appears under Scheduled deploys on the namespace page as pending approval. It will not run until an approver approves it, and if it is rejected it never runs.

Creating a schedule submits a Scheduled deploy request to the approvals queue. The approver sees whether it is one-off or recurring, the time and timezone, the first run, the catch-up grace and the tags.

Approving arms the schedule. It deploys nothing at that moment. The approval covers that exact set of tags at those exact times. The usual rules apply: one approval is sufficient, and a user who holds approver can approve their own schedule.

A one-off schedule whose time passes while it waits for approval is not armed when it is finally approved. It is marked missed, because an immediate unattended deploy is not what the approver agreed to.

  • Edit cancels the schedule and creates a replacement that must be approved again. The original’s approval and run history are kept as they were. An approval is never carried over to different content.
  • Cancel stops it firing. Past runs are kept.
  • Only the person who created a schedule, or an approver or admin, can edit or cancel it.
  • A schedule that is running cannot be edited or cancelled until that run finishes.

If your organisation has approvals turned off, a new schedule is armed immediately. The audit trail still records who created it.

Status Meaning
pending approval Waiting for an approver. It will not run
scheduled Approved and armed
running A run is executing now, or it was applied and is waiting for its rollout to become healthy
succeeded A one-off whose run was applied, read back, and rolled out healthily. The tooltip reads Its run was applied, read back, and rolled out healthily.
failed A one-off whose run failed: before the apply, at the apply, or because the rollout didn’t become healthy. The reason says which
missed Its window passed while Seylo was not running, beyond the catch-up grace. Nothing was deployed
cancelled Cancelled, or replaced by an edit
rejected An approver rejected it. It never runs

A recurring schedule stays scheduled across runs, including failed ones. A failed run does not disarm a window that was already approved, and the next window doesn’t wait for the previous run’s rollout. Each occurrence is recorded separately. Press history on the schedule to see every run, with its outcome and a link to the operation it created.

Each run in the history has its own status:

Run status Meaning
running Executing right now
applied - rollout pending Applied and read back, and waiting for its rollout. It is marked succeeded only once the new pods are healthy
succeeded Applied, read back, and rolled out healthily
failed Didn’t complete. The reason says whether it failed before the apply, at the apply, or in the rollout
missed Its window passed while Seylo was down, beyond the catch-up grace. Nothing was deployed

The schedule stores the wall-clock time and the zone, not a fixed instant, so a window at 02:30 local stays at 02:30 local across a clock change.

  • Clocks go forward and the time does not exist that day (02:30 when 02:00 jumps to 03:00): the run moves forward past the gap, to 03:30. It is never skipped.
  • Clocks go back and the time happens twice (01:30 when 02:00 returns to 01:00): the first occurrence is used. It never runs twice.

A DST badge next to the next run time explains which one applies.

When a run is due, Seylo:

  1. Claims the occurrence, so the same window can never be deployed twice.
  2. Opens an ordinary operation in Operate → Activity, attributed to the person who created the schedule and labelled [scheduled].
  3. Checks the tags exist in the registry, against the repository the live Deployment actually pulls from:
    • Tag missing: the run fails before anything touches the cluster.
    • Registry unreachable: the run fails. An unattended deploy does not proceed on a tag it could not check.
    • Registry that Seylo cannot list: the check is skipped and the run records a warning, not a pass. This is the usual case for an agent-managed cluster, because Seylo holds no registry credential for it. A missing tag then fails at rollout as ImagePullBackOff.
  4. Releases through the same path as a release you click yourself. The ownership check runs again: if Argo CD, Flux or Helm has taken over a Deployment since the schedule was approved, the run fails rather than writing a change that would be reverted at 2 a.m. An acknowledgement given when the schedule was submitted is honoured.
  5. Records the run as applied and works out the next occurrence of a recurring schedule. The run shows applied - rollout pending.
  6. Waits for the rollout’s verdict on the run’s operation, and settles the run from it:
    • The rollout became healthy: the run is succeeded.
    • The rollout failed, ended partial or drifted, or was cancelled: the run is failed, with the operation’s reason.
    • The verdict couldn’t be confirmed: the run is failed. With no operation recorded for the run, the reason begins Applied and read back, but rollout health could not be confirmed:. If Seylo restarted while the rollout was being watched, the run is failed with the Interrupted reason, never succeeded, because nobody saw the pods become healthy.

If a controller owns a Deployment when you schedule

Section titled “If a controller owns a Deployment when you schedule”

If Argo CD, Flux or Helm already owns one of the Deployments, submitting the schedule is refused and nothing is scheduled. The dialog shows This schedule was REFUSED: another controller owns the target, naming the owner and the evidence.

The durable fix is to schedule the change where that controller reads it. To schedule it 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 Submit schedule for approval again (Submit edited schedule for approval when editing). The acknowledgement is stored with the schedule, shown to the approver, and honoured when the schedule fires.

The scheduler runs inside the Seylo service. If Seylo is restarting or unavailable when a window passes, nothing is skipped silently:

How late What happens
Within the catch-up grace The run happens as soon as Seylo is back, recorded against its original time with a CATCH-UP: note
Beyond the catch-up grace The run is recorded missed, nothing is deployed, and an operation closed as Cancelled tells you so

The grace is copied onto the schedule when you create it, so a later change to the default cannot change how an existing schedule behaves. 0 turns catch-up off, so any lateness counts as missed. A run that was executing when Seylo restarted is closed failed with an Interrupted: message, because Seylo cannot know how far it got.

You see What it means What to do
Schedule deploy is disabled No component has a tag, a selected component carries a configuration change, or you lack releaser Choose a tag, turn the configuration change off, or ask for releaser
A timezone is required: "22:00" is not a time without one. … The request had no timezone Choose one
That time is in the past (…). A one-off schedule must be in the future - check the date and the timezone. The date or zone is wrong Check both. The time is local to the chosen zone
catchUpGraceSeconds must be between 0 (never catch up) and 86400 (24h). The grace is out of range Use 0 to 86400
This schedule was REFUSED: another controller owns the target A controller owns one of the Deployments. Nothing was scheduled Schedule it at the source, or acknowledge the owner. See if a controller owns a Deployment
Refusing to change Deployment shop/checkout-api: … A controller owns the Deployment As above
TAG NOT FOUND: … does not exist in the registry. … NOTHING was deployed. The tag was deleted, retagged or garbage-collected after approval Push the tag, or schedule a tag that exists
Could not list tags of "…" to verify "…": … An unattended deploy will not proceed on an unverifiable tag. The registry could not be reached when the run fired Check the registry, then re-schedule
… could NOT be pre-verified - … (a warning on the run) Seylo cannot list this registry, so the tag was not checked Expected on agent-managed clusters. Watch the rollout
Component "checkout-api" is not currently deployed in acme-prod/shop, … The component was removed after the schedule was approved Nothing was deployed. Re-create the schedule if still wanted
MISSED: this window passed …s ago while the application was not running, … The window passed beyond the catch-up grace Deploy now, or re-schedule
Interrupted: the application restarted while this scheduled run was still executing. … Seylo restarted mid-run Check the linked operation and the rollout before re-running
Applied and read back, but rollout health could not be confirmed: The change was applied, but no operation was recorded for the run, so nothing watched the pods. The run is failed Check the component’s rollout status before relying on this release
Applied, but its operation ended …: the rollout was not confirmed healthy. The rollout didn’t become healthy, and the operation gave no other reason. The run is failed Open the operation for the per-component detail. See what a failure looks like
This schedule's only run time (…) passed while it was waiting for approval, so it was NOT armed … Approved too late Create a new schedule with a future time
Only the operator who created this schedule (or an approver/admin) can cancel it. You did not create it Ask its creator or an approver
This schedule is executing right now. Cancelling would leave the deploy half-applied; … A run is in progress Let it finish, then cancel to stop future runs