mirror of
https://github.com/dredx/prole.git
synced 2026-09-23 12:03:59 +00:00
565 lines
21 KiB
Bash
Executable File
565 lines
21 KiB
Bash
Executable File
#!/usr/bin/env bash
|
|
|
|
set -euo pipefail
|
|
|
|
# init_garage_store.sh
|
|
# Purpose:
|
|
# - Deploy Garage (S3-compatible object store) in Kubernetes
|
|
# - Initialize single-node layout for immediate use
|
|
|
|
# Initialize SCRIPT_DIR
|
|
SCRIPT_DIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
|
|
|
|
# Shared option parsing for common core scripts
|
|
# shellcheck disable=SC1090
|
|
source "$SCRIPT_DIR/common_core_lib.sh"
|
|
|
|
common_core_preparse_config "$@"
|
|
|
|
# shellcheck disable=SC1090
|
|
source "$SCRIPT_DIR/knoe_cfg.sh"
|
|
|
|
set -- "${COMMON_CORE_ARGS[@]}"
|
|
common_core_parse_args "$@"
|
|
|
|
if [[ "${COMMON_CORE_HELP:-0}" == 1 ]]; then
|
|
common_core_usage "$0"
|
|
exit 0
|
|
fi
|
|
|
|
if [[ -n "${COMMON_CORE_PARSE_ERROR:-}" ]]; then
|
|
echo "ERROR: ${COMMON_CORE_PARSE_ERROR}" >&2
|
|
common_core_usage "$0"
|
|
exit 2
|
|
fi
|
|
|
|
ACTION="$COMMON_CORE_ACTION"
|
|
|
|
if [[ -n "${GARAGE_INIT_LOG:-}" ]]; then
|
|
mkdir -p "$(dirname "$GARAGE_INIT_LOG")"
|
|
exec > >(tee -a "$GARAGE_INIT_LOG") 2>&1
|
|
fi
|
|
|
|
RESOLVED_NAMESPACE="$(common_core_resolve_namespace "default")"
|
|
common_core_apply_namespace "$RESOLVED_NAMESPACE"
|
|
|
|
GARAGE_NAME=${GARAGE_NAME:-garage}
|
|
GARAGE_SECRET_NAME=${GARAGE_SECRET_NAME:-garage-secrets}
|
|
GARAGE_NODE_CAPACITY=${GARAGE_NODE_CAPACITY:-10GB}
|
|
GARAGE_ZONE=${GARAGE_ZONE:-local}
|
|
GARAGE_NAMESPACE=${GARAGE_NAMESPACE:-$RESOLVED_NAMESPACE}
|
|
NAMESPACE="$GARAGE_NAMESPACE"
|
|
APP_CLUSTER_KUBECONTEXT=${APP_CLUSTER_KUBECONTEXT:-${init_cluster_app_cluster_kubecontext:-}}
|
|
DB_CLUSTER_KUBECONTEXT=${DB_CLUSTER_KUBECONTEXT:-${init_cluster_db_cluster_kubecontext:-}}
|
|
GARAGE_AUTHORITY_ROLE=${GARAGE_AUTHORITY_ROLE:-db}
|
|
GARAGE_AUTHORITY_ROLE="${GARAGE_AUTHORITY_ROLE,,}"
|
|
ACTIVE_KUBE_CONTEXT=${KUBECTL_CONTEXT:-${KUBE_CONTEXT_NAME:-${KUBECONTEXT:-}}}
|
|
|
|
garage_context_role() {
|
|
local ctx="${ACTIVE_KUBE_CONTEXT:-}"
|
|
if [[ -z "$ctx" ]]; then
|
|
ctx=$(kubectl config current-context 2>/dev/null || true)
|
|
fi
|
|
if [[ -n "$APP_CLUSTER_KUBECONTEXT" && "$ctx" == "$APP_CLUSTER_KUBECONTEXT" ]]; then
|
|
printf '%s' "app"
|
|
return 0
|
|
fi
|
|
if [[ -n "$DB_CLUSTER_KUBECONTEXT" && "$ctx" == "$DB_CLUSTER_KUBECONTEXT" ]]; then
|
|
printf '%s' "db"
|
|
return 0
|
|
fi
|
|
printf '%s' "unknown"
|
|
}
|
|
|
|
garage_split_cluster_contexts_enabled() {
|
|
[[ -n "$APP_CLUSTER_KUBECONTEXT" && -n "$DB_CLUSTER_KUBECONTEXT" && "$APP_CLUSTER_KUBECONTEXT" != "$DB_CLUSTER_KUBECONTEXT" ]]
|
|
}
|
|
|
|
garage_is_authoritative_context() {
|
|
local role
|
|
if ! garage_split_cluster_contexts_enabled; then
|
|
return 0
|
|
fi
|
|
role="$(garage_context_role)"
|
|
[[ "$role" == "$GARAGE_AUTHORITY_ROLE" ]]
|
|
}
|
|
|
|
garage_ensure_absent_non_authoritative() {
|
|
local role
|
|
role="$(garage_context_role)"
|
|
echo "[INFO] Garage authority role is '${GARAGE_AUTHORITY_ROLE}'; current role '${role}' is non-authoritative. Ensuring Garage is absent in namespace '$NAMESPACE'."
|
|
delete_manifests || true
|
|
}
|
|
|
|
# Support both KNOE_HOME/k8s and sibling k8s directory
|
|
if [[ -d "$SCRIPT_DIR/../k8s/knoe" ]]; then
|
|
GARAGE_MANIFEST_DIR="$SCRIPT_DIR/../k8s/knoe"
|
|
elif [[ -n "${KNOE_HOME:-}" && -d "$KNOE_HOME/k8s/knoe" ]]; then
|
|
GARAGE_MANIFEST_DIR="$KNOE_HOME/k8s/knoe"
|
|
else
|
|
GARAGE_MANIFEST_DIR="$SCRIPT_DIR/../k8s/knoe"
|
|
fi
|
|
|
|
GARAGE_FILES=(
|
|
"$GARAGE_MANIFEST_DIR/storageclass-synology-iscsi.yaml"
|
|
"$GARAGE_MANIFEST_DIR/iscsi-pvs.yaml"
|
|
"$GARAGE_MANIFEST_DIR/garage-configmap.yaml"
|
|
"$GARAGE_MANIFEST_DIR/garage-statefulset.yaml"
|
|
"$GARAGE_MANIFEST_DIR/garage-service.yaml"
|
|
)
|
|
if [[ "${KNOE_MODE:-}" == "k3d" ]]; then
|
|
# k3d: no local storage provisioner — skip synology StorageClass and static PVs.
|
|
GARAGE_FILES=(
|
|
"$GARAGE_MANIFEST_DIR/garage-configmap.yaml"
|
|
"$GARAGE_MANIFEST_DIR/garage-statefulset.yaml"
|
|
"$GARAGE_MANIFEST_DIR/garage-service.yaml"
|
|
)
|
|
elif [[ "${KNOE_MODE:-}" == "k8s" ]]; then
|
|
# GKE Autopilot: apply our custom garage-hdd StorageClass (pd-standard HDD, avoids SSD quota).
|
|
# The skip-if-exists guard in apply_manifests handles idempotent re-runs safely.
|
|
# Use GCP-specific statefulset (no synology selectors, explicit resource requests).
|
|
GARAGE_FILES=(
|
|
"$GARAGE_MANIFEST_DIR/storageclass-gcp-hdd.yaml"
|
|
"$GARAGE_MANIFEST_DIR/garage-configmap.yaml"
|
|
"$GARAGE_MANIFEST_DIR/garage-statefulset-gcp.yaml"
|
|
"$GARAGE_MANIFEST_DIR/garage-service.yaml"
|
|
)
|
|
fi
|
|
|
|
GARAGE_APPLY_CHANGED=0
|
|
GARAGE_CONFIG_CHANGED=0
|
|
GARAGE_STATEFULSET_CHANGED=0
|
|
|
|
ensure_tools() {
|
|
for t in kubectl openssl; do
|
|
command -v "$t" >/dev/null || { echo "Missing required tool: $t" >&2; exit 1; }
|
|
done
|
|
}
|
|
|
|
ensure_namespace() {
|
|
if ! kubectl get namespace "$NAMESPACE" >/dev/null 2>&1; then
|
|
echo "Creating namespace '$NAMESPACE' ..."
|
|
kubectl create namespace "$NAMESPACE" >/dev/null 2>&1 || true
|
|
fi
|
|
}
|
|
|
|
ensure_secrets() {
|
|
if kubectl get secret "$GARAGE_SECRET_NAME" -n "$NAMESPACE" >/dev/null 2>&1; then
|
|
return 0
|
|
fi
|
|
|
|
echo "Creating Garage secrets in namespace '$NAMESPACE' ..."
|
|
# Wait for apiserver to be ready
|
|
local count=0
|
|
while ! kubectl get nodes >/dev/null 2>&1; do
|
|
if [ $count -ge 30 ]; then
|
|
echo "ERROR: apiserver not ready after 60s" >&2
|
|
exit 1
|
|
fi
|
|
echo "Waiting for apiserver... ($count/30)"
|
|
sleep 2
|
|
count=$((count + 1))
|
|
done
|
|
|
|
local rpc_secret admin_token metrics_token
|
|
rpc_secret=$(openssl rand -hex 32)
|
|
admin_token=$(openssl rand -base64 32)
|
|
metrics_token=$(openssl rand -base64 32)
|
|
|
|
kubectl create secret generic "$GARAGE_SECRET_NAME" -n "$NAMESPACE" \
|
|
--from-literal=rpc_secret="$rpc_secret" \
|
|
--from-literal=admin_token="$admin_token" \
|
|
--from-literal=metrics_token="$metrics_token"
|
|
}
|
|
|
|
k3d_cleanup_pending_pvc() {
|
|
if [[ "${KNOE_MODE:-}" != "k3d" ]]; then
|
|
return 0
|
|
fi
|
|
if ! kubectl get pvc data-garage-0 -n "$NAMESPACE" >/dev/null 2>&1; then
|
|
return 0
|
|
fi
|
|
local phase selector
|
|
phase=$(kubectl get pvc data-garage-0 -n "$NAMESPACE" -o jsonpath='{.status.phase}' 2>/dev/null || true)
|
|
if [[ "$phase" == "Bound" ]]; then
|
|
return 0
|
|
fi
|
|
selector=$(kubectl get pvc data-garage-0 -n "$NAMESPACE" -o jsonpath='{.spec.selector}' 2>/dev/null || true)
|
|
if printf '%s' "$selector" | grep -q "synology.storage/"; then
|
|
echo "Removing pending Garage PVC with selector for k3d local-path ..."
|
|
kubectl delete statefulset "$GARAGE_NAME" -n "$NAMESPACE" --ignore-not-found >/dev/null 2>&1 || true
|
|
kubectl delete pvc data-garage-0 -n "$NAMESPACE" --ignore-not-found >/dev/null 2>&1 || true
|
|
fi
|
|
}
|
|
|
|
apply_manifests() {
|
|
GARAGE_APPLY_CHANGED=0
|
|
GARAGE_CONFIG_CHANGED=0
|
|
GARAGE_STATEFULSET_CHANGED=0
|
|
for f in "${GARAGE_FILES[@]}"; do
|
|
if [[ -f "$f" ]]; then
|
|
local output=""
|
|
local rendered=""
|
|
local diff_out=""
|
|
local diff_rc=0
|
|
rendered=$(knoe_render_manifest "$f")
|
|
if diff_out=$(printf '%s' "$rendered" | kubectl diff -n "$NAMESPACE" -f - 2>&1); then
|
|
diff_rc=0
|
|
else
|
|
diff_rc=$?
|
|
fi
|
|
if [[ $diff_rc -eq 0 ]]; then
|
|
continue
|
|
fi
|
|
if [[ $diff_rc -ne 1 ]]; then
|
|
# For immutable StatefulSet VolumeClaimTemplates, kubectl diff itself returns an
|
|
# error (diff_rc != 1 but contains the immutable spec error). Handle it here.
|
|
if [[ "$(basename "$f")" == garage-statefulset*.yaml ]] \
|
|
&& echo "$diff_out" | grep -q "updates to statefulset spec"; then
|
|
echo "WARN: Garage StatefulSet VolumeClaimTemplates changed (diff-stage); deleting and recreating ..."
|
|
kubectl delete statefulset "$GARAGE_NAME" -n "$NAMESPACE" --ignore-not-found >/dev/null 2>&1 || true
|
|
local _pvc_sc_d
|
|
_pvc_sc_d=$(kubectl get pvc "data-garage-0" -n "$NAMESPACE" \
|
|
-o jsonpath='{.spec.storageClassName}' 2>/dev/null || true)
|
|
if [[ -n "$_pvc_sc_d" ]]; then
|
|
echo " Removing stale PVC 'data-garage-0' (storageClass: $_pvc_sc_d) ..."
|
|
kubectl delete pvc "data-garage-0" -n "$NAMESPACE" --ignore-not-found >/dev/null 2>&1 || true
|
|
fi
|
|
printf '%s' "$rendered" | kubectl apply --validate=false -n "$NAMESPACE" -f -
|
|
GARAGE_APPLY_CHANGED=1
|
|
GARAGE_STATEFULSET_CHANGED=1
|
|
continue
|
|
fi
|
|
echo "$diff_out" >&2
|
|
exit 1
|
|
fi
|
|
# StorageClass resources have immutable parameters/reclaimPolicy.
|
|
# If the StorageClass already exists in the cluster, skip re-applying it.
|
|
if echo "$rendered" | grep -q "^kind: StorageClass"; then
|
|
local _sc_name
|
|
_sc_name=$(echo "$rendered" | grep "^ name:" | head -1 | awk '{print $2}')
|
|
if [[ -n "$_sc_name" ]] && kubectl get storageclass "$_sc_name" >/dev/null 2>&1; then
|
|
echo "[SKIP] StorageClass '$_sc_name' already exists; skipping apply (immutable fields)."
|
|
continue
|
|
fi
|
|
fi
|
|
if output=$(printf '%s' "$rendered" | kubectl apply --validate=false -n "$NAMESPACE" -f - 2>&1); then
|
|
printf '%s\n' "$output"
|
|
GARAGE_APPLY_CHANGED=1
|
|
case "$(basename "$f")" in
|
|
garage-configmap.yaml) GARAGE_CONFIG_CHANGED=1 ;;
|
|
garage-statefulset.yaml) GARAGE_STATEFULSET_CHANGED=1 ;;
|
|
esac
|
|
else
|
|
if [[ "${KNOE_MODE:-}" == "k3d" && "$(basename "$f")" == "garage-statefulset.yaml" ]] \
|
|
&& echo "$output" | grep -q "updates to statefulset spec"; then
|
|
echo "WARN: Garage StatefulSet immutable in k3d; skipping apply."
|
|
continue
|
|
elif [[ "$(basename "$f")" == garage-statefulset*.yaml ]] \
|
|
&& echo "$output" | grep -q "updates to statefulset spec"; then
|
|
# VolumeClaimTemplates are immutable; delete the StatefulSet (PVCs are orphaned/preserved)
|
|
# and recreate so the new storageClass name takes effect.
|
|
echo "WARN: Garage StatefulSet VolumeClaimTemplates changed; deleting and recreating ..."
|
|
kubectl delete statefulset "$GARAGE_NAME" -n "$NAMESPACE" --ignore-not-found >/dev/null 2>&1 || true
|
|
local _pvc_sc
|
|
_pvc_sc=$(kubectl get pvc "data-garage-0" -n "$NAMESPACE" \
|
|
-o jsonpath='{.spec.storageClassName}' 2>/dev/null || true)
|
|
if [[ -n "$_pvc_sc" ]]; then
|
|
echo " Removing stale PVC 'data-garage-0' (storageClass: $_pvc_sc) ..."
|
|
kubectl delete pvc "data-garage-0" -n "$NAMESPACE" --ignore-not-found >/dev/null 2>&1 || true
|
|
fi
|
|
printf '%s' "$rendered" | kubectl apply --validate=false -n "$NAMESPACE" -f -
|
|
GARAGE_APPLY_CHANGED=1
|
|
GARAGE_STATEFULSET_CHANGED=1
|
|
continue
|
|
fi
|
|
echo "$output" >&2
|
|
exit 1
|
|
fi
|
|
else
|
|
echo "ERROR: Missing manifest: $f" >&2
|
|
exit 1
|
|
fi
|
|
done
|
|
}
|
|
|
|
ensure_container_command() {
|
|
if ! kubectl get statefulset "$GARAGE_NAME" -n "$NAMESPACE" >/dev/null 2>&1; then
|
|
return 0
|
|
fi
|
|
|
|
local cmd args
|
|
cmd=$(kubectl get statefulset "$GARAGE_NAME" -n "$NAMESPACE" -o jsonpath='{.spec.template.spec.containers[?(@.name=="garage")].command}' 2>/dev/null || true)
|
|
args=$(kubectl get statefulset "$GARAGE_NAME" -n "$NAMESPACE" -o jsonpath='{.spec.template.spec.containers[?(@.name=="garage")].args}' 2>/dev/null || true)
|
|
|
|
if [[ -z "$cmd" || "$cmd" == "[]" || "$cmd" != *"/garage"* || -z "$args" || "$args" == "[]" || "$args" != *"server"* ]]; then
|
|
echo "Patching Garage container command/args ..."
|
|
kubectl patch statefulset "$GARAGE_NAME" -n "$NAMESPACE" --type merge -p '{
|
|
"spec": {
|
|
"template": {
|
|
"spec": {
|
|
"containers": [
|
|
{
|
|
"name": "garage",
|
|
"command": ["/garage"],
|
|
"args": ["server"]
|
|
}
|
|
]
|
|
}
|
|
}
|
|
}
|
|
}' >/dev/null || true
|
|
fi
|
|
}
|
|
|
|
pin_garage_node_k3d() {
|
|
if [[ "${KNOE_MODE:-}" != "k3d" ]]; then
|
|
return 0
|
|
fi
|
|
if [[ -z "${GARAGE_PIN_NODE:-}" ]]; then
|
|
return 0
|
|
fi
|
|
if ! kubectl get statefulset "$GARAGE_NAME" -n "$NAMESPACE" >/dev/null 2>&1; then
|
|
return 0
|
|
fi
|
|
local target=""
|
|
target=$(kubectl get nodes -l node-role.kubernetes.io/master -o jsonpath='{.items[0].metadata.name}' 2>/dev/null || true)
|
|
if [[ -z "$target" ]]; then
|
|
target=$(kubectl get nodes -l node-role.kubernetes.io/control-plane -o jsonpath='{.items[0].metadata.name}' 2>/dev/null || true)
|
|
fi
|
|
if [[ -z "$target" ]]; then
|
|
return 0
|
|
fi
|
|
if [[ "$target" != k3d-* ]]; then
|
|
return 0
|
|
fi
|
|
echo "Pinning Garage StatefulSet to node '$target' for k3d exec access ..."
|
|
kubectl patch statefulset "$GARAGE_NAME" -n "$NAMESPACE" --type merge \
|
|
-p "{\"spec\":{\"template\":{\"spec\":{\"nodeSelector\":{\"kubernetes.io/hostname\":\"$target\"}}}}}" >/dev/null || true
|
|
}
|
|
|
|
restart_statefulset() {
|
|
if [[ "${GARAGE_CONFIG_CHANGED:-0}" -eq 0 ]]; then
|
|
return 0
|
|
fi
|
|
if [[ "${GARAGE_STATEFULSET_CHANGED:-0}" -eq 1 ]]; then
|
|
return 0
|
|
fi
|
|
if kubectl get statefulset "$GARAGE_NAME" -n "$NAMESPACE" >/dev/null 2>&1; then
|
|
echo "Restarting Garage StatefulSet to pick up config changes ..."
|
|
kubectl rollout restart statefulset/$GARAGE_NAME -n "$NAMESPACE" || true
|
|
fi
|
|
}
|
|
|
|
delete_manifests() {
|
|
for f in "${GARAGE_FILES[@]}"; do
|
|
if [[ -f "$f" ]]; then
|
|
kubectl delete -n "$NAMESPACE" -f "$f" --ignore-not-found
|
|
fi
|
|
done
|
|
}
|
|
|
|
dump_debug() {
|
|
echo "---- Garage debug ----"
|
|
kubectl get statefulset "$GARAGE_NAME" -n "$NAMESPACE" -o wide || true
|
|
kubectl get pods -n "$NAMESPACE" -l "app=$GARAGE_NAME" -o wide || true
|
|
kubectl get svc "$GARAGE_NAME" -n "$NAMESPACE" -o wide || true
|
|
kubectl get events -n "$NAMESPACE" --sort-by=.metadata.creationTimestamp | tail -n 50 || true
|
|
|
|
local pod
|
|
pod=$(kubectl get pods -n "$NAMESPACE" -l "app=$GARAGE_NAME" -o jsonpath='{.items[0].metadata.name}' 2>/dev/null || true)
|
|
if [[ -n "$pod" ]]; then
|
|
echo "--- pod: $pod (describe) ---"
|
|
kubectl describe pod "$pod" -n "$NAMESPACE" || true
|
|
echo "--- logs (current) ---"
|
|
kubectl logs -n "$NAMESPACE" "$pod" --tail=200 || true
|
|
echo "--- logs (previous) ---"
|
|
kubectl logs -n "$NAMESPACE" "$pod" --previous --tail=200 || true
|
|
fi
|
|
echo "---- Garage debug end ----"
|
|
}
|
|
|
|
recycle_released_knoe_iscsi_pv_for_pvc() {
|
|
# StorageClass `synology-iscsi` uses `Retain` PV reclaim policy. On restart, PVs can remain
|
|
# in `Released` with a stale `claimRef`, which prevents a same-named PVC from binding.
|
|
local ns="$1"
|
|
local pvc="$2"
|
|
|
|
if ! kubectl get pvc "$pvc" -n "$ns" >/dev/null 2>&1; then
|
|
return 0
|
|
fi
|
|
|
|
local pvc_phase pvc_uid
|
|
pvc_phase=$(kubectl get pvc "$pvc" -n "$ns" -o jsonpath='{.status.phase}' 2>/dev/null || true)
|
|
if [[ "$pvc_phase" != "Pending" ]]; then
|
|
return 0
|
|
fi
|
|
pvc_uid=$(kubectl get pvc "$pvc" -n "$ns" -o jsonpath='{.metadata.uid}' 2>/dev/null || true)
|
|
|
|
local pv sc rp phase claim_ns claim_name claim_uid
|
|
while IFS=$'\t' read -r pv sc rp phase claim_ns claim_name claim_uid; do
|
|
[[ "$sc" == "synology-iscsi" ]] || continue
|
|
[[ "$rp" == "Retain" ]] || continue
|
|
[[ "$phase" == "Released" ]] || continue
|
|
[[ "$claim_ns" == "$ns" ]] || continue
|
|
[[ "$claim_name" == "$pvc" ]] || continue
|
|
|
|
# If the PV claimRef UID matches the current PVC UID, don't touch it.
|
|
if [[ -n "${claim_uid:-}" && -n "${pvc_uid:-}" && "${claim_uid:-}" == "${pvc_uid:-}" ]]; then
|
|
continue
|
|
fi
|
|
|
|
echo "Recycling Released PV '$pv' for PVC '$ns/$pvc' (clearing stale claimRef) ..."
|
|
kubectl patch pv "$pv" --type json -p '[{"op":"remove","path":"/spec/claimRef"}]' >/dev/null 2>&1 || true
|
|
done < <(kubectl get pv -o jsonpath='{range .items[*]}{.metadata.name}{"\t"}{.spec.storageClassName}{"\t"}{.spec.persistentVolumeReclaimPolicy}{"\t"}{.status.phase}{"\t"}{.spec.claimRef.namespace}{"\t"}{.spec.claimRef.name}{"\t"}{.spec.claimRef.uid}{"\n"}{end}' 2>/dev/null || true)
|
|
}
|
|
|
|
wait_ready() {
|
|
echo "Waiting for Garage StatefulSet to become ready ..."
|
|
|
|
local initial_timeout="${PVC_REPAIR_WAIT_TIMEOUT:-30s}"
|
|
if ! kubectl rollout status statefulset/$GARAGE_NAME -n "$NAMESPACE" --timeout="$initial_timeout"; then
|
|
if [[ "${KNOE_MODE:-}" == "k8s" ]]; then
|
|
echo "WARN: Garage not ready after $initial_timeout; PVC provisioning may still be in progress (GKE CSI)." >&2
|
|
else
|
|
echo "WARN: Garage not ready after $initial_timeout; checking for Released synology-iscsi PVs with stale claimRefs ..." >&2
|
|
fi
|
|
recycle_released_knoe_iscsi_pv_for_pvc "$NAMESPACE" "data-garage-0" || true
|
|
fi
|
|
|
|
if ! kubectl rollout status statefulset/$GARAGE_NAME -n "$NAMESPACE" --timeout=${ROLLOUT_TIMEOUT:-300s}; then
|
|
echo "ERROR: Garage StatefulSet did not become ready in time." >&2
|
|
dump_debug
|
|
return 1
|
|
fi
|
|
}
|
|
|
|
init_layout() {
|
|
local pod
|
|
pod=$(kubectl get pods -n "$NAMESPACE" -l "app=$GARAGE_NAME" -o jsonpath='{.items[0].metadata.name}' 2>/dev/null || true)
|
|
if [[ -z "$pod" ]]; then
|
|
echo "ERROR: Garage pod not found in namespace '$NAMESPACE'." >&2
|
|
exit 1
|
|
fi
|
|
|
|
local node_id=""
|
|
echo "Determining Garage node ID..."
|
|
for i in {1..30}; do
|
|
node_id=$(kubectl exec -n "$NAMESPACE" "$pod" -- /garage node id -q 2>/dev/null | awk '{print $1}' || true)
|
|
if [[ -n "$node_id" ]]; then
|
|
break
|
|
fi
|
|
echo "Waiting for Garage node ID to be available... ($i/30)"
|
|
sleep 2
|
|
done
|
|
|
|
if [[ -z "$node_id" ]]; then
|
|
echo "ERROR: Unable to determine Garage node ID after 30 attempts." >&2
|
|
dump_debug
|
|
exit 1
|
|
fi
|
|
|
|
# Extract the short ID for better matching
|
|
local short_id
|
|
short_id=$(echo "$node_id" | cut -d'@' -f1)
|
|
echo "Garage node ID: $short_id"
|
|
|
|
local layout
|
|
layout=$(kubectl exec -n "$NAMESPACE" "$pod" -- /garage layout show 2>/dev/null || true)
|
|
if ! echo "$layout" | grep -q "$short_id"; then
|
|
echo "Assigning Garage node role (capacity: $GARAGE_NODE_CAPACITY, zone: $GARAGE_ZONE) ..."
|
|
# Retry assigning role as it might fail if node is not yet fully ready in the cluster logic
|
|
local max_assign_retries=10
|
|
local assign_count=0
|
|
while ! kubectl exec -n "$NAMESPACE" "$pod" -- /garage layout assign -z "$GARAGE_ZONE" -c "$GARAGE_NODE_CAPACITY" "$short_id"; do
|
|
if [[ $assign_count -ge $max_assign_retries ]]; then
|
|
echo "ERROR: Failed to assign Garage node role after $max_assign_retries retries." >&2
|
|
exit 1
|
|
fi
|
|
echo "Retrying Garage layout assign... ($((assign_count + 1))/$max_assign_retries)"
|
|
sleep 2
|
|
assign_count=$((assign_count + 1))
|
|
done
|
|
|
|
# Re-fetch layout and apply staged changes using the NEXT layout version
|
|
local layout_after version next_version apply_ok
|
|
layout_after=$(kubectl exec -n "$NAMESPACE" "$pod" -- /garage layout show 2>/dev/null | grep -v "INFO" || true)
|
|
# Extract "Current cluster layout version: X"
|
|
version=$(echo "$layout_after" | awk -F: '/Current cluster layout version/ {gsub(/[[:space:]]/,"",$2); print $2; exit}' || true)
|
|
|
|
if echo "$layout_after" | grep -qi "staged"; then
|
|
if [[ -n "$version" ]]; then
|
|
next_version=$((version + 1))
|
|
echo "Applying Garage layout (current version $version, next version $next_version) ..."
|
|
if kubectl exec -n "$NAMESPACE" "$pod" -- /garage layout apply --version "$next_version"; then
|
|
apply_ok=1
|
|
else
|
|
apply_ok=0
|
|
fi
|
|
else
|
|
apply_ok=0
|
|
fi
|
|
if [[ "${apply_ok:-0}" -eq 0 ]]; then
|
|
echo "WARN: layout apply with explicit version failed; retrying without version..."
|
|
kubectl exec -n "$NAMESPACE" "$pod" -- /garage layout apply || true
|
|
fi
|
|
else
|
|
echo "No staged layout changes detected; skipping layout apply."
|
|
fi
|
|
else
|
|
echo "Garage layout already assigned for node $short_id."
|
|
fi
|
|
}
|
|
|
|
status() {
|
|
ensure_tools
|
|
echo "Garage status in namespace '$NAMESPACE':"
|
|
kubectl get statefulset "$GARAGE_NAME" -n "$NAMESPACE" || true
|
|
kubectl get pods -n "$NAMESPACE" -l "app=$GARAGE_NAME" || true
|
|
kubectl get svc "$GARAGE_NAME" -n "$NAMESPACE" || true
|
|
}
|
|
|
|
case "$ACTION" in
|
|
start|initialize|update|reload|restart)
|
|
ensure_tools
|
|
if [[ "$GARAGE_AUTHORITY_ROLE" != "db" && "$GARAGE_AUTHORITY_ROLE" != "app" ]]; then
|
|
echo "ERROR: GARAGE_AUTHORITY_ROLE must be 'db' or 'app' (got '$GARAGE_AUTHORITY_ROLE')." >&2
|
|
exit 1
|
|
fi
|
|
if garage_split_cluster_contexts_enabled; then
|
|
current_role="$(garage_context_role)"
|
|
if [[ "$current_role" == "unknown" ]]; then
|
|
echo "ERROR: Unable to resolve Garage deployment cluster role from kubecontext '${ACTIVE_KUBE_CONTEXT:-<unset>}' (APP='${APP_CLUSTER_KUBECONTEXT:-<unset>}', DB='${DB_CLUSTER_KUBECONTEXT:-<unset>}')." >&2
|
|
exit 1
|
|
fi
|
|
if ! garage_is_authoritative_context; then
|
|
garage_ensure_absent_non_authoritative
|
|
exit 0
|
|
fi
|
|
fi
|
|
ensure_namespace
|
|
ensure_secrets
|
|
k3d_cleanup_pending_pvc
|
|
echo "Applying Garage manifests in namespace '$NAMESPACE'..."
|
|
apply_manifests
|
|
ensure_container_command
|
|
pin_garage_node_k3d
|
|
restart_statefulset
|
|
wait_ready
|
|
init_layout
|
|
knoe_register_port_forward "garage" "${NAMESPACE:-default}" "svc/garage" "3900" "3900" "0.0.0.0" "TCP" "Garage S3 API"
|
|
;;
|
|
stop)
|
|
ensure_tools
|
|
echo "Deleting Garage manifests from namespace '$NAMESPACE'..."
|
|
delete_manifests
|
|
;;
|
|
status)
|
|
status
|
|
;;
|
|
*)
|
|
common_core_usage "$0"
|
|
exit 1
|
|
;;
|
|
esac
|