Why is this task waiting while that one runs?

How a task gets admitted

Admission is the rule that decides, at each moment, which of the open tasks may start. It runs every time a front frees up, and it answers one question: does this task’s file set overlap a file set that is already running?

The rule, in order

1. Does the task declare files? If it does, that is its set. 2. If not, it inherits its role’s lane. A task with no declared files is assumed to be able to touch anything in its role’s lane. 3. A file with no owner runs alone. If the task touches a path no lane claims, it is admitted only when nothing else is running. 4. Overlap means wait. If the set intersects the set of any running task, the task stays in the queue and the runner looks at the next one.

Nothing about this is a guess. Overlap is computed from paths, not from what the task is about.

Reading the queue while it runs

Every front is a pane, and every pane header carries the role and the task title. A task that is waiting says so, with the reason — the lane it needs is held by the task that has it.

That visibility is the point. A front that does not fit on the screen would be automation you cannot see, which is exactly what Muster refuses to be.

When a task finishes

The agent hands the turn back. Muster waits for the silence to settle — two and a half seconds of no output, then five more seconds to confirm — and then runs the acceptance command outside the session. Exit zero closes the task; anything else blocks it with the output attached; no command at all stops it for human review.

Only then does the runner look for the next task that fits. A closed task frees its files, and something that was waiting on them can start.

Why it can look slower than you expected

Three tasks in the same lane will run one after another, no matter how many fronts you allow. That is the rule working. If you expected parallelism and did not get it, the lanes are the place to look — see When parallel turns serial.

Next

Several fronts at once covers how many can run, and what each one costs your machine.