Approval Queue
Review and decide on manual-mode update candidates from a single dashboard view.
Overview
Under update mode manual, an eligible update sits on the container row waiting for someone to click Update. The approval queue is the same candidate set surfaced as a work list instead: every container whose update the Update button would currently accept gets a row at Approvals in the sidebar, with the release notes and any soft blockers already loaded, and one of three actions to close it out.
The queue is a durable ledger of decisions, not a second source of truth about what's updateable. Whether a candidate belongs in the queue at all — and whether it's still the one it was when the row was created — is answered live against the container on every read, the same eligibility check the container list and the Update button already use.
What produces a row
A row is created for a container when, on any watch cycle:
- it has a raw update candidate (a soft-gated candidate still counts — see below);
- no hard blocker applies (a locked Update button never reaches the queue);
- and it would not be auto-dispatched. Under
updateMode: auto, a container fully resolved to automatic execution by a matching action trigger skips the queue entirely and updates on its own.
| Update mode | Produces queue rows? |
|---|---|
notify | No — manual updates are refused server-wide, so nothing is admissible to queue |
manual | Yes, for every eligible candidate |
auto | Only for candidates a configured action trigger will not auto-dispatch (no matching trigger, AUTO=oninclude without the include label, agent mismatch, and so on) |
A group policy with actions.updateMode: manual lowers the ceiling for its members the same way. Under global auto, a member of that group is never auto-dispatched, so its candidate queues here even when a trigger would otherwise have applied it, which makes a manual group a way to require approval for one group. A group notify is a hard blocker (group-notify-only), so nothing queues for its members. A member's dd.action.auto or dd.action.include label doesn't lift either ceiling.
snoozed, maturity-not-reached, skip-tag, skip-digest, threshold-not-reached, maintenance-window-closed) does not keep a candidate out of the queue — it becomes one of the row's hold reasons, shown inline next to the release notes. These are exactly the conditions the manual Update button already overrides with its "Update anyway" confirmation, so approving a row with hold reasons carries the same override, spelled out before you click. A hard blocker (security-scan-blocked, trigger-excluded, active-operation, and the rest) behaves the same as everywhere else in Drydock: the row is left alone if one already exists, but nothing new queues while it's locked.The three actions
Each action on a row is a thin wrapper over an update primitive that already exists elsewhere in Drydock — approving from the queue and clicking Update on the container panel go through the identical admission path and produce the identical rejection for the identical container state.
- Approve dispatches the update exactly as the container panel's Update button does. Any reason that path would refuse — including the
notify-mode 409 — is returned here unchanged. - Reject writes the container's
skipTags/skipDigestspolicy, the same effect as the container row's Skip action. This is undoable from the Update policy menu, and the confirmation dialog says so before you commit to it. A scope picker next to the button widens a tag rejection: This version only (the default), Whole minor line (everyX.Y.*tag) or Whole major line (everyX.*tag). Rejecting major 3 while you run 2.x keeps offering 2.x patches and a later 4.x, and only hides the 3.x line. The line scopes only appear for tag updates; a digest update, or a tag that isn't a semver version, can only be rejected by exact tag or digest. - Defer snoozes the container until the same instant the row is deferred to (7 days by default; see the API for a custom expiry). A deferred row returns to
pendingthe moment the snooze expires — there is no sweep job, it's a live check on every read.
Select several rows in the queue to approve or reject them in one go; bulk approve asks for confirmation first, lists the hold reasons it will override, and skips rows with a hard blocker. Each row is still decided on its own, so a row that can't be approved is reported and the rest go through.
Every decision writes one audit entry (update-approved, update-rejected, or update-deferred) carrying the approval id, the decided-by identity, and an optional note, and pushes a dd:approval-decided SSE event to any open client.
Where it appears
- Sidebar — the Approvals nav item carries a count badge for pending rows, live over SSE.
- Dashboard — a stat tile mirrors the same pending count and links straight to the queue.
/approvals— the queue itself: tabs for Pending / Deferred / Decided, inline expand for release notes and hold reasons, and keyboard shortcuts (j/kto move,a/r/dto act,Enterto expand).
Retention
Rows leave the queue the moment they're decided or their candidate is superseded or withdrawn, but they aren't deleted — a decided or resolved row is pruned automatically 30 days after it left the queue. A row still pending or under an active deferral is never pruned. The audit entry is the permanent record either way; the row itself is a working-set convenience.
Notifications
update-pending-approval is a notification rule, shipped enabled: false. When you enable it and assign triggers (an empty trigger list means every notification trigger), it fires once when a candidate enters the queue. It's driven by the new row, not by the watch cycle, so the same pending candidate never notifies twice, even across a restart. A superseding candidate is a new row and notifies again. It isn't sent under update mode notify or when container actions are disabled, because approving would be refused; the row still appears in the queue. The rule follows the same THRESHOLD, include/exclude and AUTO gating as update-available, and works with simple, batch and digest trigger modes.
The default title is Update awaiting approval for container <name>, and the body names the version change (1.0.0 to 1.1.0) followed by a link to the queue row. Templates can read event.approvalId, event.fromRef, event.toRef and event.approvalUrl.
DD_PUBLIC_URL. Without it the notification is still sent, but event.approvalUrl is empty and the link is left out of the default body.It ships disabled because manual is the default mode, so enabling it for everyone would send two notifications per candidate (update-available plus this one). Treat it as the richer replacement for update-available when you review updates through the queue.
Related
- Update Eligibility & Blockers — the hard/soft blocker distinction the queue reuses verbatim.
- Triggers — Notification Rules —
update-pending-approval. - Approvals API — list, summary, detail, the three decision endpoints, and bulk.