By name: serve-restart
description: Cleanly stop, free the port, and restart hipfire serve. Use when serve "Failed to start (port in use)", a stale daemon holds VRAM, an os-error-2/JSON-parse pre-warm crash left a zombie singleton, or you just want a guaranteed-fresh daemon. Kills both bun CLI serve and the spawned target/release/examples/daemon, reaps stale ~/.hipfire/daemon.pid + serve.pid, fuser-frees the port, then relaunches.
Cleanly restart hipfire serve
Serve thrash burned hours on 2026-05-25: stale daemon.pid singletons
("FATAL already running"), zombie daemons holding 30 GB VRAM (next load
OOMs), and pid-by-name kills missing the actual port owner. This skill
does the whole cycle atomically.
One-shot
scripts/serve-restart.sh [port] [-- <extra hipfire serve args>]
Default port 11435. Env honored: HIPFIRE_MODELS_DIR, HIPFIRE_VERIFY_GRAPH.
It: kills cli/index.ts serve + examples/daemon, fuser -k <port>/tcp,
removes ~/.hipfire/{daemon,serve}.pid, waits the port free, relaunches
detached, tails to warm-up complete. (It no longer touches
/tmp/hipfire-gpu.lock — that is an flock'd file the kernel auto-releases
on holder death; deleting it would break GPU mutual exclusion.)
Just kill, don't restart
scripts/serve-restart.sh --kill-only 11435
Why pid-by-name isn't enough
pkill -f examples/daemon matched a bash wrapper while the real owner of
:11435 survived. Always also fuser -k 11435/tcp. A daemon can show 39 GB
RSS — kill it before the next load or you OOM with "0 MB free". Verify
VRAM frees: rocm-smi --showmeminfo vram | grep Used → ~10 MB idle.