Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
#!/usr/bin/env bash
#
# dev.sh — day-to-day minikube lifecycle for the Clerum dev loop.
#
# Thin wrapper over the existing minikube-* Make targets that adds the two
# things the raw targets don't: a service→namespace map for a parametrized
# multi-service rebuild, and a health check on the real liveness routes.
#
# Subcommands:
# up Start Docker+minikube (if down), relaunch background
# port-forwards, health check. Boots an ALREADY-provisioned
# cluster — this is NOT a full regen (use `make minikube-setup`).
# down Stop port-forwards, then `minikube stop`. Safe: preserves
# the cluster, its DB and PVCs (nothing is deleted).
# redeploy [svc...] Rebuild image(s) + rollout restart, then relaunch
# port-forwards + health check. Defaults to the trio we
# iterate on most: control-api control-ui mcp-host.
# Override with an explicit list, e.g.
# scripts/minikube/dev.sh redeploy control-api mcp-host
#
# Honors MINIKUBE_PROFILE (default clerum-test). Health probes use the default
# shared ports (3000/8090/8080/...); branch profiles with random port-forwards
# have their own preflight and are out of scope here.
set -euo pipefail
PROFILE="${MINIKUBE_PROFILE:-clerum-test}"
SAFE_PROFILE="${PROFILE//[^A-Za-z0-9_.-]/_}"
REPO_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
cd "$REPO_ROOT"
KC=(kubectl --context="$PROFILE")
DEFAULT_SVCS=(control-api control-ui mcp-host)
# svc → "<namespace> <deployment> <port> <health_path>"
# port/health_path empty ("- -") means: rollout-only, no HTTP probe.
# deployment column encodes the mcp-host→chatllm rename already baked into the
# Makefile (minikube_deployment). Keep this table in sync with the build-images.sh
# --only=<svc> selectors and the actual deployment names in-cluster.
svc_meta() {
case "$1" in
control-api) echo "control-plane control-api 8090 /health" ;;
control-ui) echo "control-plane control-ui 3000 /" ;;
mcp-host) echo "mcp-host chatllm 8080 /v1/runtime/health" ;;
profile-ui) echo "profiles profile-ui 3001 /" ;;
external-rest-api) echo "profiles external-rest-api 8091 /health" ;;
rpc-proxy) echo "rpc-proxy rpc-proxy 8094 /health" ;;
hcc) echo "control-plane host-context-controller - -" ;;
*) return 1 ;;
esac
}
KNOWN_SVCS="control-api control-ui mcp-host profile-ui external-rest-api rpc-proxy hcc"
log() { printf '%s\n' "$*"; }
step() { printf '\n=== %s ===\n' "$*"; }
# Relaunch background port-forwards (nohup, non-holding) via the existing target.
relaunch_pf() {
step "port-forwards ($PROFILE)"
make --no-print-directory minikube-pf-all-bg
}
# HTTP health probe for the given services (skips ones with no port).
health_check() {
step "health"
local svc ns dep port path code line=""
for svc in "$@"; do
svc_meta "$svc" >/dev/null || continue
read -r ns dep port path <<<"$(svc_meta "$svc")"
[[ "$port" == "-" ]] && continue
# On a connection failure curl exits non-zero and -w still prints 000.
code="$(curl -s -o /dev/null -m 5 -w '%{http_code}' "http://127.0.0.1:${port}${path}" 2>/dev/null || true)"
line+="${svc}:${port} ${code:-000} · "
done
log "${line% · }"
step "pods"
# List pods per (namespace, deployment) so we show exactly the services we
# touched — the pod name is "<deployment>-<rs-hash>-<pod-hash>", and the
# anchored pattern excludes sibling deployments like control-api-rpc-gateway.
for svc in "$@"; do
svc_meta "$svc" >/dev/null || continue
read -r ns dep port path <<<"$(svc_meta "$svc")"
"${KC[@]}" get pods -n "$ns" --no-headers 2>/dev/null \
| awk -v d="$dep" '$1 ~ "^"d"-[a-z0-9]+-[a-z0-9]+$"' || true
done
}
cmd_up() {
step "start cluster ($PROFILE)"
make --no-print-directory minikube-start
relaunch_pf
health_check "${DEFAULT_SVCS[@]}"
log ""
log "Control UI → http://127.0.0.1:3000"
}
cmd_down() {
step "stop port-forwards"
local pidfile pid
shopt -s nullglob
for pidfile in /tmp/pf-"${SAFE_PROFILE}"-*.pid; do
pid="$(cat "$pidfile" 2>/dev/null || true)"
if [[ -n "$pid" ]] && kill -0 "$pid" 2>/dev/null; then
kill "$pid" 2>/dev/null || true
fi
rm -f "$pidfile"
done
shopt -u nullglob
# NOTE: we deliberately do NOT `pkill -f pf-all-stack.sh` here. Killing the
# port-forward PIDs above already frees the ports (the background flow). A
# broad pkill would also reach a holding `make minikube-pf-all` from a
# DIFFERENT profile, since the profile isn't on that process's command line.
step "minikube stop ($PROFILE) — preserves cluster, DB, PVCs"
make --no-print-directory minikube-stop
log ""
log "Cluster stopped safely. Bring it back with: make minikube-up"
}
cmd_redeploy() {
local svcs=("$@")
[[ ${#svcs[@]} -eq 0 ]] && svcs=("${DEFAULT_SVCS[@]}")
# Validate up front so a typo fails before any build starts.
local svc ns dep port path
for svc in "${svcs[@]}"; do
if ! svc_meta "$svc" >/dev/null; then
log "ERROR: unknown service '$svc'. Known: $KNOWN_SVCS" >&2
exit 2
fi
done
log "Redeploying: ${svcs[*]}"
local results=()
for svc in "${svcs[@]}"; do
read -r ns dep _ _ <<<"$(svc_meta "$svc")"
step "rebuild $svc → deploy/$dep in $ns"
if make --no-print-directory minikube-deploy-service SVC="$svc" NS="$ns" DEPLOYMENT="$dep"; then
results+=("$svc=ok")
else
results+=("$svc=FAIL")
fi
done
# `|| true`: a flaky port-forward must not abort before the build summary.
relaunch_pf || true
health_check "${svcs[@]}"
step "summary"
log "${results[*]}"
# Non-zero exit if any service failed to build/roll out.
[[ "${results[*]}" == *FAIL* ]] && exit 1 || true
}
usage() {
cat >&2 <<EOF
Usage: scripts/minikube/dev.sh <up|down|redeploy> [svc...]
up Start cluster + port-forwards + health check (not a regen).
down Stop port-forwards + minikube stop (safe; preserves data).
redeploy [svc...] Rebuild + rollout + PF + health. Default: ${DEFAULT_SVCS[*]}.
Known services: $KNOWN_SVCS
EOF
exit 64
}
main() {
local sub="${1:-}"
[[ $# -gt 0 ]] && shift || true
case "$sub" in
up) cmd_up ;;
down) cmd_down ;;
redeploy) cmd_redeploy "$@" ;;
""|-h|--help|help) usage ;;
*) log "ERROR: unknown subcommand '$sub'" >&2; usage ;;
esac
}
main "$@"