Tasks

How asynchronous work is tracked, and what to do when a task fails.

Anything that takes longer than an instant becomes a task. Creating a VM, migrating one, building a storage cluster, upgrading a node, exporting a LUN — the API accepts the request, creates a task, and returns immediately.

This is why the console never blocks on a slow operation, and why closing the browser does not cancel anything.

Task states

Status Meaning
PENDING Accepted, not started
RUNNING In progress; progress is a percentage
SUCCESS Completed
FAILED Did not complete — the task carries the reason
CANCELLED Stopped by an operator
TIMED_OUT Exceeded its time limit

The Tasks screen

Every task across the fleet, newest first, with its title, status, progress and start time. Progress streams live over a WebSocket, so a running task’s bar moves without reloading.

Opening a task shows its detail, including the failure reason when there is one, and which host it ran against.

Hyperion[vsnode1]> show task
Hyperion[vsnode1]> show task <id>

When a task fails

A failed task is usually the real explanation for whatever else looks wrong — a VM that will not start, a host stuck in BOOTSTRAPPING, a pool that never appeared.

  1. Read the reason. The task carries the error, not just the fact of failure.
  2. Check the host it ran on. A task that failed because its host was unreachable is a host problem, not a task problem.
  3. Retry, once you have fixed the cause. Tasks are retryable:

    Hyperion[vsnode1]> task retry <id>
    
  4. Cancel it if it is stuck and you do not want it retried:

    Hyperion[vsnode1]> task cancel <id>
    

Note — Retrying a task that failed for an unfixed reason will fail again the same way. Tasks are not a queue that drains if you wait.

Tasks you did not start

Not every task comes from a person. The platform creates tasks for its own scheduled work — snapshot policies, metric retention, health checks, reconciliation after a host comes back. Seeing tasks appear on an idle cluster is normal.

A repeating failure from a task nobody started is worth chasing: something the platform expects to be true is not.

Where tasks show up

  • Tasks — all of them.
  • Dashboard — the most recent, as a health signal.
  • The object itself — a VM’s screen shows the tasks that acted on it.

Anything that changed state also lands in the audit log, which records who asked. Tasks record what happened; the audit log records who asked for it.