Free disk and hold the queue

A fine external-aerodynamics sweep writes about 18 GB, most of it field time directories. When the jobs disk fills, a solver dies mid-write, so the server guards against it.

The disk floor

A queued job waits while the jobs disk has less free space than MCOPENFOAM_MIN_FREE_GB (5 by default). Its summary says so:

queued on cfd-host, waiting for disk (3.8 GB free, need 5)

It starts on its own when space comes back. A submit below the floor still queues and carries a disk_low warning.

Reclaim space from finished jobs

prune_job removes what a report and a restart do not need:

prune_job({"job_id": "flight-sweep-fine-087a44e8"})

It keeps time level 0, every sweep point’s end level (what extending a sweep starts from), a plain run’s latest level, postProcessing and the figures. It removes every other field time directory, in the case root and in each processor*.

With samples: true it also removes the VTP surface and plane samples. Rendered figures stay, but render_figures can no longer redraw them, so copy the samples off first if you might.

delete_job removes a job entirely. Both tools refuse on a job that is queued or running, and both report what they freed and the free space left.

Hold the queue

For maintenance, such as moving the jobs directory, hold the queue so nothing new starts while the running job finishes:

make -C deploy hold REASON="moving jobs to a bigger volume"
make -C deploy release

The hold is a .maintenance file in the jobs directory. Queued jobs show held for maintenance (<reason>), server_info reports it, and a running job is unaffected.