fuse host metrics

Show a host's capacity vs. allocated

fuse host metrics [id]

No flags beyond the globals. [id] defaults to the active host selection if omitted; fails with a clear error if there’s no ID and no active host.

$ fuse host metrics prod-east-1
host               prod-east-1
state              active
cpus               4 / 16 (12 free)
ram mb             8192 / 65536 (57344 free)
storage gb         100 / 500 (400 free)
vms                1 / 20 (19 free)
gpus               2 / 4 (2 free, a100)
mig                1g.10gb 1 / 4 (3 free), 2g.20gb 0 / 2 (2 free)

The gpus and mig rows are conditional, exactly as in fuse host get: gpus appears only when the host advertises GPU capacity, and mig only when it advertises MIG instance pools. A CPU-only host prints neither rather than a row of zeros:

$ fuse host metrics cpu-only-1
host               cpu-only-1
state              active
cpus               4 / 16 (12 free)
ram mb             8192 / 65536 (57344 free)
storage gb         100 / 500 (400 free)
vms                1 / 20 (19 free)

MIG profiles are listed in sorted order, and each pool is counted in MIG instances rather than whole devices. The gpus row counts whole devices, so on a MIG-partitioned host the two rows are not expected to agree.

This is per-host capacity accounting, not Prometheus data, for fleet-wide metrics see fuse metrics. There is no per-VM metrics API, this is the finest granularity available from the CLI.