97 lines
3.7 KiB
Bash
97 lines
3.7 KiB
Bash
|
#!/bin/bash
|
||
|
|
# Tier-1 dispatch for prometheus component operations.
|
||
|
|
# Delegates to generated run-{op}.sh scripts when present (manifest-plan workflow),
|
||
|
|
# or falls back to inline _do_{op} functions for direct invocation.
|
||
|
|
set -euo pipefail
|
||
|
|
|
||
|
|
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
|
||
|
|
cd "$SCRIPT_DIR"
|
||
|
|
|
||
|
|
[ -f "${SCRIPT_DIR}/env-prometheus" ] && source "${SCRIPT_DIR}/env-prometheus"
|
||
|
|
|
||
|
|
PROMETHEUS_NAME="${PROMETHEUS_NAME:-prometheus}"
|
||
|
|
PROMETHEUS_NAMESPACE="${PROMETHEUS_NAMESPACE:-observability}"
|
||
|
|
PROMETHEUS_PORT="${PROMETHEUS_PORT:-9090}"
|
||
|
|
PROMETHEUS_STORAGE_CLASS="${PROMETHEUS_STORAGE_CLASS:-longhorn-retain}"
|
||
|
|
PROMETHEUS_STORAGE_SIZE="${PROMETHEUS_STORAGE_SIZE:-20Gi}"
|
||
|
|
PROMETHEUS_RETENTION="${PROMETHEUS_RETENTION:-30d}"
|
||
|
|
PROMETHEUS_SCRAPE_INTERVAL="${PROMETHEUS_SCRAPE_INTERVAL:-30s}"
|
||
|
|
|
||
|
|
CMD_TSK="${CMD_TSK:-${1:-install}}"
|
||
|
|
|
||
|
|
source "${SCRIPT_DIR}/prometheus-lib.sh"
|
||
|
|
|
||
|
|
_resolve_kubeconfig() {
|
||
|
|
if [ -n "${KUBECONFIG:-}" ] && [ -f "$KUBECONFIG" ]; then
|
||
|
|
return
|
||
|
|
fi
|
||
|
|
for candidate in /var/lib/k0s/pki/admin.conf /etc/kubernetes/admin.conf /root/.kube/config; do
|
||
|
|
if [ -f "$candidate" ]; then
|
||
|
|
export KUBECONFIG="$candidate"
|
||
|
|
return
|
||
|
|
fi
|
||
|
|
done
|
||
|
|
echo "ERROR: kubeconfig not found — set KUBECONFIG explicitly" >&2; exit 1
|
||
|
|
}
|
||
|
|
|
||
|
|
_do_install() {
|
||
|
|
for manifest in namespace.yaml serviceaccount.yaml clusterrole.yaml clusterrolebinding.yaml \
|
||
|
|
configmap.yaml alert-rules.yaml pvc.yaml deployment.yaml service.yaml; do
|
||
|
|
local path="$SCRIPT_DIR/manifests/${manifest}"
|
||
|
|
[ -f "$path" ] && _kubectl apply -f "$path"
|
||
|
|
done
|
||
|
|
echo "Waiting for ${PROMETHEUS_NAME} pod to be ready..."
|
||
|
|
_wait_for_pod 180
|
||
|
|
echo "${PROMETHEUS_NAME} installed in namespace ${PROMETHEUS_NAMESPACE}"
|
||
|
|
}
|
||
|
|
|
||
|
|
_do_update() {
|
||
|
|
for manifest in configmap.yaml alert-rules.yaml deployment.yaml; do
|
||
|
|
local path="$SCRIPT_DIR/manifests/${manifest}"
|
||
|
|
[ -f "$path" ] && _kubectl apply -f "$path"
|
||
|
|
done
|
||
|
|
_kubectl rollout restart "deployment/${PROMETHEUS_NAME}" --namespace "$PROMETHEUS_NAMESPACE"
|
||
|
|
echo "Waiting for ${PROMETHEUS_NAME} to be ready..."
|
||
|
|
_wait_for_pod 180
|
||
|
|
echo "${PROMETHEUS_NAME} updated in namespace ${PROMETHEUS_NAMESPACE}"
|
||
|
|
}
|
||
|
|
|
||
|
|
_do_delete() {
|
||
|
|
for manifest in deployment.yaml service.yaml clusterrolebinding.yaml clusterrole.yaml; do
|
||
|
|
local path="$SCRIPT_DIR/manifests/${manifest}"
|
||
|
|
[ -f "$path" ] && _kubectl delete -f "$path" --ignore-not-found
|
||
|
|
done
|
||
|
|
echo "${PROMETHEUS_NAME} deleted from namespace ${PROMETHEUS_NAMESPACE}"
|
||
|
|
}
|
||
|
|
|
||
|
|
_do_restart() {
|
||
|
|
_kubectl rollout restart "deployment/${PROMETHEUS_NAME}" --namespace "$PROMETHEUS_NAMESPACE"
|
||
|
|
_wait_for_pod 180
|
||
|
|
echo "${PROMETHEUS_NAME} restarted"
|
||
|
|
}
|
||
|
|
|
||
|
|
_do_health() {
|
||
|
|
local pod
|
||
|
|
pod=$(_pod_name 2>/dev/null || true)
|
||
|
|
if [ -z "$pod" ]; then
|
||
|
|
echo "UNHEALTHY: no prometheus pod found in namespace ${PROMETHEUS_NAMESPACE}" >&2; exit 1
|
||
|
|
fi
|
||
|
|
if _kubectl exec --namespace "$PROMETHEUS_NAMESPACE" "$pod" -- \
|
||
|
|
wget -qO- "http://localhost:${PROMETHEUS_PORT}/-/healthy" >/dev/null 2>&1; then
|
||
|
|
echo "HEALTHY: ${PROMETHEUS_NAME} responding on /-/healthy"
|
||
|
|
else
|
||
|
|
echo "UNHEALTHY: ${PROMETHEUS_NAME} did not respond on /-/healthy" >&2; exit 1
|
||
|
|
fi
|
||
|
|
}
|
||
|
|
|
||
|
|
_resolve_kubeconfig
|
||
|
|
|
||
|
|
case "$CMD_TSK" in
|
||
|
|
install) [ -f "$SCRIPT_DIR/run-init.sh" ] && exec bash "$SCRIPT_DIR/run-init.sh"; _do_install ;;
|
||
|
|
update) [ -f "$SCRIPT_DIR/run-update.sh" ] && exec bash "$SCRIPT_DIR/run-update.sh"; _do_update ;;
|
||
|
|
delete) [ -f "$SCRIPT_DIR/run-delete.sh" ] && exec bash "$SCRIPT_DIR/run-delete.sh"; _do_delete ;;
|
||
|
|
restart) [ -f "$SCRIPT_DIR/run-restart.sh" ] && exec bash "$SCRIPT_DIR/run-restart.sh"; _do_restart ;;
|
||
|
|
health) _do_health ;;
|
||
|
|
*) echo "ERROR: unknown CMD_TSK='${CMD_TSK}'. Valid: install|update|delete|restart|health" >&2; exit 1 ;;
|
||
|
|
esac
|