refactor(k8s): Reorganize into 5-layer structure with production kustomizations

2026-08-18 15:08:01 -07:00
parent 563f720d09
commit 0f30d77288
214 changed files with 405 additions and 885 deletions
@@ -0,0 +1,68 @@
+# minio.descriptor.yaml
+# Build input for generating svc-minio.yaml dashboard ConfigMap
+# and svc-minio-rules.yaml PrometheusRule.
+
+service: minio
+display_name: "MinIO"
+namespace: storage
+service_type: stateful-store
+folder: "MinIO"
+
+jobs:
+  - minio
+
+metrics:
+  prefix: minio
+  up_selector: 'job="minio"'
+
+  rate:
+    metric: minio_s3_requests_total
+    labels: [method, bucket]
+
+  duration:
+    metric: minio_s3_requests_duration_seconds
+    type: histogram
+    unit: s
+
+  error:
+    metric: minio_s3_requests_total
+    filter: 'error="true"'
+
+  domain:
+    - metric: minio_cluster_capacity_usable_bytes
+      kind: gauge
+    - metric: minio_cluster_capacity_raw_total_bytes
+      kind: gauge
+    - metric: minio_replication_metrics_failed_byte_count
+      kind: gauge
+    - metric: minio_replication_metrics_replicating_byte_count
+      kind: gauge
+    - metric: minio_cluster_health_drives_online
+      kind: gauge
+    - metric: minio_cluster_health_drives_offline
+      kind: gauge
+
+correlation_ids: []
+
+logs:
+  loki_namespace_selector: 'namespace="storage"'
+
+alerts:
+  - name: HighErrorRate
+    expr_template: rate_error_ratio
+    threshold: 0.05
+    for: 10m
+    severity: warning
+  - name: DiskSpaceLow
+    expr_template: gauge_below_percentage
+    metric: minio_cluster_capacity_usable_bytes
+    threshold: 0.1
+    base_metric: minio_cluster_capacity_raw_total_bytes
+    for: 5m
+    severity: critical
+  - name: ReplicationLag
+    expr_template: gauge_above_threshold
+    metric: minio_replication_metrics_replicating_byte_count
+    threshold: 1073741824
+    for: 15m
+    severity: warning