Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
#!/usr/bin/env bash
# Runtime orchestration helpers for pre-gate-sync.sh. The caller provides
# strict shell mode, KC, PROFILE, PROJECT_DIR, GATE_NAME, and log().
DEFAULT_HOST_CRD_SCHEMA_PROBE_REF="mcp-host/chatllm"
HOST_CRD_SCHEMA_PROBE_REF="${HOST_CRD_SCHEMA_PROBE_REF:-${DEFAULT_HOST_CRD_SCHEMA_PROBE_REF}}"
export HOST_CRD_SCHEMA_PROBE_REF
preflight_host_lifecycle_probe() {
if ! ${KC} get deployment host-context-controller -n control-plane >/dev/null 2>&1; then
log "Skipping Host lifecycle probe preflight (fresh bootstrap: HCC is absent)"
return 0
fi
if [[ ! "${HOST_CRD_SCHEMA_PROBE_REF}" =~ ^([a-z0-9]([-a-z0-9]*[a-z0-9])?)/([a-z0-9]([-a-z0-9.]*[a-z0-9])?)$ ]]; then
log "ERROR: HOST_CRD_SCHEMA_PROBE_REF must be namespace/name, got '${HOST_CRD_SCHEMA_PROBE_REF}'"
return 1
fi
local namespace="${HOST_CRD_SCHEMA_PROBE_REF%%/*}"
local name="${HOST_CRD_SCHEMA_PROBE_REF#*/}"
if ! ${KC} get host "${name}" -n "${namespace}" >/dev/null 2>&1; then
log "ERROR: manifest-backed Host lifecycle probe ${HOST_CRD_SCHEMA_PROBE_REF} is absent; refusing to rebuild or mutate the profile"
return 1
fi
log "Host lifecycle probe preflight passed: ${HOST_CRD_SCHEMA_PROBE_REF}"
}
rollout_if_present() {
local namespace="$1" deployment="$2"
if ${KC} get deployment "${deployment}" -n "${namespace}" >/dev/null 2>&1; then
log "Waiting for rollout: ${namespace}/${deployment}"
${KC} rollout status "deployment/${deployment}" -n "${namespace}" --timeout=120s >/dev/null
fi
}
rollout_restart_with_retry() {
local namespace="$1" deployment="$2" attempt output
for attempt in 1 2 3; do
if output="$(${KC} rollout restart "deployment/${deployment}" -n "${namespace}" 2>&1)"; then
[[ -n "${output}" ]] && printf '%s\n' "${output}"
return 0
fi
if [[ "${output}" == *"within the past second"* && "${attempt}" != 3 ]]; then
log "Retrying rollout restart for ${namespace}/${deployment} after recent restart"
sleep 2
continue
fi
printf '%s\n' "${output}" >&2
return 1
done
}
rollout_namespace_deployments() {
local namespace="$1" names
names="$(${KC} get deployment -n "${namespace}" -o jsonpath='{range .items[*]}{.metadata.name}{"\n"}{end}' 2>/dev/null || true)"
[[ -n "${names}" ]] || return 0
while IFS= read -r deployment; do
[[ -z "${deployment}" ]] || rollout_if_present "${namespace}" "${deployment}"
done <<<"${names}"
}
# The workflow gateway mounts nginx.conf through a subPath, so a ConfigMap can
# be updated while a running pod continues serving the previous route table.
# The manifest-level test prevents omission in source; this runtime assertion
# prevents a stale or partially applied profile from being used for an SDK gate.
assert_workflow_gateway_prompt_bridge_finalization_route() {
local deployment="nginx-workflow-approval-gateway"
local namespace="control-plane"
local expected_route='location ~ ^/api/v1/mcp-host/plugin-workload-sdk/invocations/[^/]+/finalize$'
local candidate deletion ready nginx_config inspect_rc pod_count=0
if ! ${KC} get deployment "${deployment}" -n "${namespace}" >/dev/null 2>&1; then
log "ERROR: ${namespace}/${deployment} is absent; refusing Plugin Workload SDK gate"
return 1
fi
# A rollout can leave an old pod in Running/Terminating briefly. Do not
# inspect the first pod returned by the API: that made the previous guard
# race a healthy new ReplicaSet and reject a valid deployment. Only inspect
# Ready, non-terminating pods, and require the route in every active pod.
while IFS= read -r candidate; do
[[ -z "${candidate}" ]] && continue
deletion="$(${KC} get pod "${candidate}" -n "${namespace}" \
-o jsonpath='{.metadata.deletionTimestamp}' 2>/dev/null || true)"
[[ -n "${deletion}" ]] && continue
ready="$(${KC} get pod "${candidate}" -n "${namespace}" \
-o jsonpath='{.status.containerStatuses[?(@.name=="nginx")].ready}' 2>/dev/null || true)"
[[ "${ready}" == "true" ]] || continue
pod_count=$((pod_count + 1))
# Consume the complete command before searching it. With the caller's
# `set -o pipefail`, piping `kubectl exec ... nginx -T` into `grep -q`
# lets grep close the pipe as soon as it finds the route. kubectl can then
# exit on SIGPIPE (141), turning a present route into a false failure.
# Keeping execution and content validation separate also preserves the
# distinction between an unreachable pod and a genuinely stale config.
if nginx_config="$(${KC} exec -n "${namespace}" "${candidate}" -c nginx -- nginx -T 2>&1)"; then
:
else
inspect_rc=$?
log "ERROR: could not inspect the active nginx configuration in ${namespace}/${candidate} (kubectl exec rc=${inspect_rc}); refusing gate"
return 1
fi
if [[ "${nginx_config}" != *"${expected_route}"* ]]; then
log "ERROR: active ${namespace}/${candidate} does not serve the SDK finalization route; refusing gate"
return 1
fi
done < <(${KC} get pods -n "${namespace}" -l app="${deployment}" \
--field-selector=status.phase=Running -o jsonpath='{range .items[*]}{.metadata.name}{"\n"}{end}' 2>/dev/null || true)
if [[ "${pod_count}" == "0" ]]; then
log "ERROR: ${namespace}/${deployment} has no Ready non-terminating pod; refusing Plugin Workload SDK gate"
return 1
fi
log "Workflow gateway serves the SDK promptBridge finalization route"
}
# The HCC gateway also mounts nginx.conf through a subPath. NP-08 gates must
# prove every active pod serves only the caller-bound v2 Host routes while the
# two caller-selected legacy Host routes are local tombstones. The global v1
# inventory remains temporarily for the retained mcp-proxy poller (PR 2).
assert_hcc_gateway_np08_routes() {
local deployment="host-context-controller-api-gateway"
local namespace="control-plane"
local candidate deletion ready rendered pod_count=0
if ! ${KC} get deployment "${deployment}" -n "${namespace}" >/dev/null 2>&1; then
log "ERROR: ${namespace}/${deployment} is absent; refusing NP-08 gate"
return 1
fi
while IFS= read -r candidate; do
[[ -z "${candidate}" ]] && continue
deletion="$(${KC} get pod "${candidate}" -n "${namespace}" \
-o jsonpath='{.metadata.deletionTimestamp}' 2>/dev/null || true)"
[[ -n "${deletion}" ]] && continue
ready="$(${KC} get pod "${candidate}" -n "${namespace}" \
-o jsonpath='{.status.containerStatuses[?(@.name=="nginx")].ready}' 2>/dev/null || true)"
[[ "${ready}" == "true" ]] || continue
pod_count=$((pod_count + 1))
rendered="$(${KC} exec -n "${namespace}" "${candidate}" -c nginx -- nginx -T 2>&1)" || {
log "ERROR: could not inspect active ${namespace}/${candidate}; refusing NP-08 gate"
return 1
}
if ! grep -Fq 'location = /api/v2/hosts/self/mcpservers {' <<<"${rendered}" ||
! grep -Fq 'location = /api/v2/hosts/self/mcpservers/credential {' <<<"${rendered}" ||
! grep -Fq 'location ~ ^/api/v1/mcpservers/context/[^/]+$ {' <<<"${rendered}" ||
! grep -Fq 'location ~ ^/api/v1/mcpservers/[^/]+/auth$ {' <<<"${rendered}" ||
! grep -Fq 'proxy_set_header Authorization $http_authorization;' <<<"${rendered}" ||
! grep -Fq 'access_log /dev/stdout hcc_gateway_json;' <<<"${rendered}" ||
! grep -Fq 'add_header Cache-Control "no-store, private" always;' <<<"${rendered}" ||
! grep -Fq 'add_header Pragma "no-cache" always;' <<<"${rendered}" ||
! grep -Fq 'add_header X-Content-Type-Options "nosniff" always;' <<<"${rendered}" ||
[[ "$(grep -Fc "return 410 '{\"error\":\"gone\"}';" <<<"${rendered}")" -lt 2 ]]; then
log "ERROR: active ${namespace}/${candidate} lacks the NP-08 gateway contract"
return 1
fi
done < <(${KC} get pods -n "${namespace}" -l app="${deployment}" \
--field-selector=status.phase=Running -o jsonpath='{range .items[*]}{.metadata.name}{"\n"}{end}' 2>/dev/null || true)
if [[ "${pod_count}" == "0" ]]; then
log "ERROR: ${namespace}/${deployment} has no Ready non-terminating pod; refusing NP-08 gate"
return 1
fi
log "HCC gateway serves caller-bound NP-08 routes"
}
gate_needs_registry() {
[[ "${GATE_NAME}" == *registry* || "${GATE_NAME}" == *marketplace* || "${GATE_NAME}" == *plugin-workload-sdk* ]]
}
ensure_evenfire_registry() {
if ! gate_needs_registry; then
log "Skipping evenfire-registry before ${GATE_NAME}; this gate does not require the sibling service"
return 0
fi
log "Ensuring evenfire-registry and minikube registry egress policies before ${GATE_NAME}"
(
cd "${PROJECT_DIR}"
if ${KC} -n registry get deployment registry-api >/dev/null 2>&1; then
MINIKUBE_PROFILE="${PROFILE}" SKIP_BUILD=1 make minikube-deploy-evenfire-registry
else
MINIKUBE_PROFILE="${PROFILE}" make minikube-deploy-evenfire-registry
fi
)
}