Before you start
- Access to the cluster in the Ankra portal, or the Ankra CLI logged in to your organisation.
executions.operateon the cluster to retry or cancel. Without it, the operation page says “Retry and cancellation require executions.operate for this cluster.” See Roles and access.
Find the failed operation
1
Open the cluster's operations
Go to Infrastructure → Clusters → <your cluster> → Technical details → Operations (
/next/clusters/<cluster-id>/operations). Technical details is a group in the cluster menu that folds away; open it if it is closed.The page lists everything the platform has run against the cluster, newest first, with how far each got and how it ended. Tick Needs attention only to keep the failures that still need someone, or pick a status in the status filter (Running, Stopping, Success, Cleanup, Failed, Critical or Cancelled). A Live badge means a run is in progress and the list refreshes while it goes.2
Open the operation
Click the failed operation. Its page shows the status, a count of jobs by state, when it started and its last update, then the Jobs list in run order.Type in Search jobs, targets or errors… to find a job, or pick a lane in the job status filter. Switch between List and Kanban: the kanban sorts jobs into Waiting, In progress, Needs attention and Finished (and Other statuses when a job reports one of those). A failed job shows its error under its name.
3
Read the failed step
Under Needs you at the top of the page, click Inspect failed step (or click any job). The step page shows:
- Recorded result - the step’s status, when it started and ended, and its duration.
- Step output - the output each task of the step recorded, and Live output while the step runs. Output Ankra treats as protected is withheld.
- Target and source - the resource, kind, namespace, Stack, chart, registry and source path the step targeted.
- Status history - every status the step went through, with its time.
4
Ask Ankra AI if the error is not obvious
Click Investigate with Ankra on the operation page. It opens Ankra AI with a question about this operation filled in, ready for you to send.
Retry or cancel
- Portal
- CLI
- Review retry appears on a failed, critical or cancelled operation. The dialog asks Start a new attempt?: the platform replays the operation as a new execution, and completed steps may run again. It is not a resume from the failed step. Click Start new attempt. The page opens the new operation, which says it is “A new attempt of the previous operation. The previous record is unchanged.”
- Review cancellation appears while an operation is running. The dialog asks Cancel this operation?: cancellation stops eligible work and cannot undo completed changes. Click Request cancellation. A step that is still running finishes first, and the page reads “Cancelling: a step is still running and finishes first. Retry becomes available once the run has stopped.”
- Re-apply stack (force) appears instead of Review retry when Kubernetes did not apply a resource because it was changed outside the Stack. Retrying without force meets the same conflict. The re-apply applies every manifest in the Stack again and takes back the fields changed outside it, as a new operation. It needs the
stacks.deployandclusters.writepermissions.
Remote-command timeouts
A remote SSH command retains its timeout during startup. If the runner gives up or its SSH session closes, a watchdog asks processes in that session to terminate and force-stops those that remain after the grace period. Delayed watchdog startup does not reset the timeout or make a disconnected session appear live. Processes that detach into another session, including independently started services, can continue. Cleanup does not reverse completed changes. Operation cancellation still follows the current-step behaviour described above.Verify
Open Operations again. The retry appears as a new operation at the top; wait for it to read Completed. The original failed row keeps its failed status, and once the resources it failed on are healthy it carries Resolved by a later run. On the failed operation’s page, Open the resolving operation links to the run that fixed it. An operation’s status says what ran, not whether the workloads are healthy now. To check them, click Inspect workloads under Operation details.What can go wrong
The retry dialog shows the platform’s refusal when a retry cannot start:
If the page says “Ankra started a new run but could not confirm it yet.”, look for the new run on Operations before you retry again.
Failed history versus open problems
The operations list is an audit log, not a to-do list. A failed operation keeps its failed status for as long as the record is kept, and nothing acknowledges or deletes it, so a rollout that needed a few attempts leaves its failed attempts in the history for good. What tells you whether a failure still needs someone is its attention state:
The state is worked out each time you read the list, so a successful retry resolves the operation it retried. In the portal a resolved failure carries Resolved by a later run, and Needs attention only hides it; in the CLI filter with
--attention open or --attention resolved.
Operations with no cluster
An application deploy, a managed service or a security fix can run for the organisation rather than on one cluster. Its operation opens at/next/operations/<operation-id>, the address notifications and check runs link to. That page shows the status, the jobs and each step’s output, read only: Inspect failed step opens the failed step, and there is no retry or cancel. An operation that ran on a cluster opens on that cluster’s operation page from the same address.