Skip to content
EmbraDocs

Operations and states

Every change (build, deploy, rollback, restore, import…) is an operation with an operation_id and a state. See it with embra status <op_id> or under “Lần deploy” (deploys) in the console.

States

Section titled “States”
state Console Meaning Exit code
requested Đã tiếp nhận (received) Embra has it, not started yet —
planned Đã chuẩn bị (prepared) The deploy has a plan —
rehearsing Đang chạy thử migration (rehearsing migration) Rehearsal on a data copy —
awaiting_approval Chờ duyệt (awaiting approval) Waiting for an Owner or Admin 20
ready Sẵn sàng deploy (ready to deploy) Approved (or no approval needed), waiting to run —
migrating Đang chạy migration (running migration) Applying migrations to production —
starting Đang khởi động (starting) Starting the new container, waiting for the health check —
switching Đang chuyển traffic (switching traffic) Moving traffic to the new release (or to the new database on restore) —
observing Đã nhận traffic, đang theo dõi (live, observing) New release takes traffic, Embra keeps watching —
running Đang thực hiện (in progress) A non-deploy operation is running —
healthy Deploy thành công (deployed; rollback: Rollback xong) Deploy or rollback done, new release serving 0
succeeded Hoàn tất (done) A non-deploy operation finished 0
failed Thất bại (failed) Could not be done; see reason_code 10
rejected Đã từ chối (rejected) The approver rejected it, or Embra refused it before running 10
cancelled Đã huỷ (cancelled) Cancelled before touching production 10
recovery_required Cần đội Embra hỗ trợ (needs the Embra team) Outcome unknown, environment blocked 30
migration_unknown Cần xác minh dữ liệu (data needs verification) Lost contact while running a migration, environment blocked 30
  • Exit code is what embra wait / embra status return when the operation is in that state. “—” means not final yet; the exit code then follows reason_code (Exit codes and JSON).
  • Final states (never change again): healthy, succeeded, failed, rejected, cancelled, recovery_required, migration_unknown. ready is not final.
  • embra wait stops at a final state or at awaiting_approval.

Deploy

Section titled “Deploy”
planned → rehearsing → awaiting_approval → ready → migrating → starting → switching → observing → healthy
From Can move to
planned rehearsing, awaiting_approval, ready, cancelled
rehearsing awaiting_approval, ready, failed, cancelled
awaiting_approval ready, rejected, cancelled
ready migrating, starting, failed, cancelled
migrating starting, failed, migration_unknown
starting switching, failed
switching observing, failed
observing healthy, failed
  • No migrations: rehearsing and migrating are skipped. No approval needed (staging without data-loss migrations): awaiting_approval is skipped.
  • ready → failed: an error known before touching production (for example the pre-migration snapshot fails).
  • From migrating only starting, failed or migration_unknown. Once in migrating it can no longer be cancelled.
  • Failing after migrating (for example the health check): the migration is applied, old code runs on the new schema.

Other operations

Section titled “Other operations”
Operation Path
Rollback requested → switching → observing → healthy; refused early → rejected
Build requested → running → succeeded; a newer push → cancelled
Restore (PITR) requested → running → awaiting_approval → switching → succeeded; not switched → rejected / cancelled
Data import requested → (awaiting_approval on production) → running → succeeded
Fork requested → awaiting_approval → succeeded
Everything else (Postgres setup, backup, export, secret, domain…) requested → running → succeeded

On top of these tables, any non-final state can move to recovery_required (except migrating, which uses migration_unknown). Export and the comparison runs of embra db undo only read, so they never become recovery_required.

recovery_required and migration_unknown

Section titled “recovery_required and migration_unknown”

Both mean Embra does not know whether the last step finished. Embra deliberately does not retry: running a migration or a database switch twice when the first outcome is unknown can corrupt data.