stackgres 14+ changes

Signed-off-by: Jesse Nelson <[email protected]>
hashgraph · Dec 21, 2024 · a56f1fa · a56f1fa
1 parent 67c7afd
commit a56f1fa
Show file tree

Hide file tree

Showing 5 changed files with 62 additions and 14 deletions.
diff --git a/charts/hedera-mirror-common/Chart.yaml b/charts/hedera-mirror-common/Chart.yaml
@@ -27,7 +27,7 @@ dependencies:
     condition: stackgres.enabled
     name: stackgres-operator
     repository: https://stackgres.io/downloads/stackgres-k8s/stackgres/helm/
-    version: 1.13.0
+    version: 1.14.1
   - condition: traefik.enabled
     name: traefik
     repository: https://helm.traefik.io/traefik

diff --git a/charts/hedera-mirror-common/values.yaml b/charts/hedera-mirror-common/values.yaml
@@ -417,10 +417,53 @@ promtail:
       operator: Exists
 
 stackgres:
+  collector:
+    config:
+      exporters:
+        prometheus:
+          metric_expiration: 2m
+      service:
+        pipelines:
+          metrics:
+            receivers:
+              - prometheus/prometheus  # This appears to be corrected to just prometheus in 1.15+ and can be updated (possibly removed) when upgrading
+    resources:
+      limits:
+        memory: "4Gi"
+        cpu: "1500m"
+      requests:
+        memory: "3Gi"
+        cpu: "1000m"
+    prometheusOperator:
+      monitors:
+        - name: mirror-prometheus-prometheus  # Should be able to rename this in 1.15+
+          metadata:
+            namespace: common
+            name: mirror-prometheus-prometheus
+          spec:
+            podMetricsEndpoints:
+              - honorLabels: true
+                honorTimestamps: true
+                interval: 1m
+                metricRelabelings:
+                  - action: drop
+                    regex: ^(pg_stat.*_user.*|pg_table.*|pg_total_relation_size_bytes|pg.*)$
+                    sourceLabels:
+                      - __name__
+                path: /metrics
+                port: prom-http
+                scheme: https
+                tlsConfig:
+                  ca:
+                    secret:
+                      key: tls.crt
+                      name: mirror-prometheus-prometheus
+                  serverName: stackgres-collector
+
   enabled: false
   operator:
     image:
-      name: gcr.io/mirrornode/stackgres-operator
+      name: docker.io/xinatswirlds/stackgres-operator
   prometheusRules:
     enabled: true
     DatabaseStorageFull:

diff --git a/charts/hedera-mirror/templates/stackgres/stackgres-cluster.yaml b/charts/hedera-mirror/templates/stackgres/stackgres-cluster.yaml
@@ -6,14 +6,17 @@ metadata:
   name: {{ include "hedera-mirror.stackgres" . }}
   namespace: {{ include "hedera-mirror.namespace" . }}
 spec:
-  {{- if .Values.stackgres.backup.enabled }}
   configurations:
+    {{- if .Values.stackgres.backup.enabled }}
     backups:
     {{- range $backupConfig := .Values.stackgres.backup.instances }}
     - sgObjectStorage: {{ include "hedera-mirror.stackgres" $ }}-object-storage
       {{$backupConfig | toYaml | nindent 6}}
     {{- end }}
-  {{- end }}
+    {{- end }}
+    observability:
+      disableMetrics: {{ not .Values.stackgres.observability.enableMetricsExporter }}
+      prometheusAutobind: {{ .Values.stackgres.observability.enableMetricsExporter }}
   coordinator:
     configurations:
       sgPoolingConfig: {{ include "hedera-mirror.stackgres" . }}-coordinator
@@ -23,7 +26,6 @@ spec:
       scripts:
         - sgScript: {{ include "hedera-mirror.stackgres" . }}-coordinator
     pods:
-      disableMetricsExporter: {{ not .Values.stackgres.coordinator.enableMetricsExporter }}
       disablePostgresUtil: {{ not .Values.stackgres.coordinator.enablePostgresUtil }}
       persistentVolume: {{ .Values.stackgres.coordinator.persistentVolume | toYaml | nindent 8 }}
       resources:
@@ -51,9 +53,11 @@ spec:
     ssl:
       enabled: false  # Disable SSL to work around https://github.com/hashgraph/hedera-mirror-node/issues/9143
     version: {{ .Values.stackgres.postgresVersion | quote }}
-  prometheusAutobind: {{ or .Values.stackgres.coordinator.enableMetricsExporter .Values.stackgres.worker.enableMetricsExporter }}
+  prometheusAutobind: {{ .Values.stackgres.observability.enableMetricsExporter }}
   replication:
-    mode: {{ .Values.stackgres.coordinator.replication.mode }}
+    mode: {{ .Values.stackgres.replication.mode }}
+    initialization:
+      mode: {{ .Values.stackgres.replication.initialization.mode }}
   shards:
     clusters: {{ .Values.stackgres.worker.instances }}
     configurations:
@@ -65,7 +69,6 @@ spec:
         - sgScript: {{ include "hedera-mirror.stackgres" . }}-worker
     overrides: {{ .Values.stackgres.worker.overrides | toYaml | nindent 6 }}
     pods:
-      disableMetricsExporter: {{ not .Values.stackgres.worker.enableMetricsExporter }}
       disablePostgresUtil: {{ not .Values.stackgres.worker.enablePostgresUtil }}
       persistentVolume: {{ .Values.stackgres.worker.persistentVolume | toYaml | nindent 8 }}
       resources:

diff --git a/charts/hedera-mirror/values-prod.yaml b/charts/hedera-mirror/values-prod.yaml
@@ -145,10 +145,10 @@ stackgres:
       cronSchedule: "45 22 * * *"
       compression: lz4
       useVolumeSnapshot: true
-      fastVolumeSnapshot: false # Can be true once StackGres issue 2887 is resolved
+      fastVolumeSnapshot: true
       volumeSnapshotClass: zfs
       performance:
-        uploadDiskConcurrency: 1 # Can be 4 once StackGres issue 2887 is resolved
+        uploadDiskConcurrency: 1
     objectStorage:
       endpoint: http://mirror-minio.common:9000
       bucket: stackgres-backup

diff --git a/charts/hedera-mirror/values.yaml b/charts/hedera-mirror/values.yaml
@@ -310,7 +310,6 @@ stackgres:
       wal_init_zero: "off"  # Not needed with ZFS
       wal_recycle: "off"  # Not needed with ZFS
       work_mem: "12MB"
-    enableMetricsExporter: false
     enablePostgresUtil: true
     instances: 1
     pgbouncer:
@@ -325,8 +324,6 @@ stackgres:
           pool_mode: session
         mirror_importer:
           pool_mode: session
-    replication:
-      mode: sync-all
     resources:
       cpu: 100m
       memory: 1Gi
@@ -342,9 +339,15 @@ stackgres:
     - name: pg_trgm
       version: "1.6"
   nameOverride: citus
+  observability:
+    enableMetricsExporter: true
   podAntiAffinity: true
   postgresVersion: "16.2"
   priorityClassName: ""
+  replication:
+    initialization:
+      mode: FromReplica
+    mode: sync-all
   worker:
     config:
       autovacuum_max_workers: "2"
@@ -362,7 +365,6 @@ stackgres:
       wal_init_zero: "off"  # Not needed with ZFS
       wal_recycle: "off"  # Not needed with ZFS
       work_mem: "12MB"
-    enableMetricsExporter: false
     enablePostgresUtil: true
     instances: 1
     overrides: []  # Override shard(s) configuration