add migration support to the party allocator

[static]

Signed-off-by: Nicu Reut <nicu.reut@digitalasset.com>

Author: nicu-da

cluster/expected/validator-runbook/expected.json

@@ -564,6 +564,11 @@
           "validatorApiUrl": "http://validator-app:5003"
         },
         "imageRepo": "us-central1-docker.pkg.dev/da-cn-shared/ghcr/digital-asset/decentralized-canton-sync-dev/docker",
+        "pvc": {
+          "name": "party-allocator-keys",
+          "size": 100,
+          "storageClassName": "standard-rwo"
+        },
         "tolerations": [
           {
             "effect": "NoSchedule",

cluster/helm/splice-party-allocator/templates/party-allocator.yaml

@@ -54,7 +54,7 @@ spec:
 apiVersion: v1
 kind: PersistentVolumeClaim
 metadata:
-  name: party-allocator-keys
+  name: {{ .Values.pvc.name }}
   namespace: {{ $.Release.Namespace }}
   annotations:
     helm.sh/resource-policy: keep
@@ -63,8 +63,12 @@ spec:
     - ReadWriteOnce
   resources:
     requests:
-      storage: 20G
+      storage: {{ .Values.pvc.size }}
   storageClassName: {{ .Values.pvc.volumeStorageClass }}
+  {{- with .Values.pvc.dataSource }}
+  dataSource:
+    {{- toYaml . | nindent 4 }}
+  {{- end }}
 ---
 apiVersion: v1
 kind: Service
@@ -77,8 +81,8 @@ spec:
   selector:
     app: {{ .Release.Name }}
   ports:
-  - name: metrics
-    port: 10013
+    - name: metrics
+      port: 10013
 ---
 apiVersion: monitoring.coreos.com/v1
 kind: ServiceMonitor
@@ -89,11 +93,11 @@ metadata:
   namespace: {{ .Release.Namespace }}
 spec:
   endpoints:
-  - port: metrics
-    interval: {{ .Values.metrics.interval }}
+    - port: metrics
+      interval: {{ .Values.metrics.interval }}
   selector:
     matchLabels:
       app: {{ .Release.Name }}
   namespaceSelector:
     matchNames:
-    - {{ .Release.Namespace }}
+      - {{ .Release.Namespace }}

cluster/helm/splice-party-allocator/values-template.yaml

@@ -4,7 +4,9 @@
 imageRepo: "ghcr.io/digital-asset/decentralized-canton-sync/docker"
 
 pvc:
+  name: party-allocator-keys
   volumeStorageClass: standard-rwo
+  size: 20Gi
 
 metrics:
   release: prometheus-grafana-monitoring

cluster/pulumi/common-validator/src/config.ts

@@ -111,6 +111,7 @@ export const PartyAllocatorConfigSchema = z.object({
   maxParties: z.number().default(1000000),
   preapprovalRetries: z.number().default(120),
   preapprovalRetryDelayMs: z.number().default(1000),
+  pvcSize: z.number().default(100),
 });
 export type PartyAllocatorConfig = z.infer<typeof PartyAllocatorConfigSchema>;
 

cluster/pulumi/common/src/storage/storageClass.ts

@@ -6,4 +6,11 @@ export const standardStorageClassName = hyperdiskSupportConfig.hyperdiskSupport.
   ? 'hyperdisk-standard-rwo'
   : 'standard-rwo';
 
+export const infraStandardStorageClassName = hyperdiskSupportConfig.hyperdiskSupport.enabled
+  ? 'hyperdisk-standard-rwo'
+  : 'standard-rwo';
+
+export const infraPremiumStorageClassName = hyperdiskSupportConfig.hyperdiskSupport.enabled
+  ? 'hyperdisk-balanced-rwo'
+  : 'premium-rwo';
 export const pvcSuffix = hyperdiskSupportConfig.hyperdiskSupport.enabled ? 'hd-pvc' : 'pvc';

cluster/pulumi/infra/src/observability.ts

@@ -11,11 +11,14 @@ import {
   CLUSTER_NAME,
   clusterProdLike,
   commandScriptPath,
+  createVolumeSnapshot,
   ExactNamespace,
   GCP_PROJECT,
   GrafanaKeys,
   HELM_MAX_HISTORY_SIZE,
   infraAffinityAndTolerations,
+  infraPremiumStorageClassName,
+  infraStandardStorageClassName,
   isMainNet,
   loadTesterConfig,
   ObservabilityReleaseName,
@@ -32,6 +35,7 @@ import { local } from '@pulumi/command';
 import { getSecretVersionOutput } from '@pulumi/gcp/secretmanager/getSecretVersion';
 import { Input } from '@pulumi/pulumi';
 
+import { hyperdiskSupportConfig } from '../../common/src/config/hyperdiskSupportConfig';
 import {
   clusterIsResetPeriodically,
   enableAlertEmailToSupportTeam,
@@ -91,9 +95,9 @@ const grafanaExternalUrl = `https://grafana.${CLUSTER_HOSTNAME}`;
 const alertManagerExternalUrl = `https://alertmanager.${CLUSTER_HOSTNAME}`;
 const prometheusExternalUrl = `https://prometheus.${CLUSTER_HOSTNAME}`;
 const shouldIgnoreNoDataOrDataSourceError = clusterIsResetPeriodically;
+const namespaceName = 'observability';
 
 export function configureObservability(dependsOn: pulumi.Resource[] = []): pulumi.Resource {
-  const namespaceName = 'observability';
   const namespace = new k8s.core.v1.Namespace(
     namespaceName,
     {
@@ -116,6 +120,7 @@ export function configureObservability(dependsOn: pulumi.Resource[] = []): pulum
   const prometheusStackCrdVersion = '0.85.0';
   const postgres = installPostgres({ ns: namespace, logicalName: namespaceName });
   const adminPassword = grafanaKeysFromSecret().adminPassword;
+  const migrationSnapshots = getVolumeSnapshotsForHyperdiskMigration();
   const prometheusStack = new k8s.helm.v3.Release(
     'observability-metrics',
     {
@@ -205,13 +210,14 @@ export function configureObservability(dependsOn: pulumi.Resource[] = []): pulum
             storage: {
               volumeClaimTemplate: {
                 spec: {
-                  storageClassName: 'standard-rwo',
+                  storageClassName: infraStandardStorageClassName,
                   accessModes: ['ReadWriteOnce'],
                   resources: {
                     requests: {
                       storage: '5Gi',
                     },
                   },
+                  ...(migrationSnapshots.alertManager ? migrationSnapshots.alertManager : {}),
                 },
               },
             },
@@ -266,13 +272,14 @@ export function configureObservability(dependsOn: pulumi.Resource[] = []): pulum
             storageSpec: {
               volumeClaimTemplate: {
                 spec: {
-                  storageClassName: 'premium-rwo',
+                  storageClassName: infraPremiumStorageClassName,
                   accessModes: ['ReadWriteOnce'],
                   resources: {
                     requests: {
                       storage: infraConfig.prometheus.storageSize,
                     },
                   },
+                  ...(migrationSnapshots.prometheus ? migrationSnapshots.prometheus : {}),
                 },
               },
             },
@@ -398,7 +405,7 @@ export function configureObservability(dependsOn: pulumi.Resource[] = []): pulum
             type: 'Recreate',
           },
           persistence: {
-            enabled: true,
+            enabled: !hyperdiskSupportConfig.hyperdiskSupport.migratingInfra,
             type: 'pvc',
             accessModes: ['ReadWriteOnce'],
             size: '5Gi',
@@ -1065,3 +1072,30 @@ function installPostgres(namespace: ExactNamespace): SplicePostgres {
     true // useInfraAffinityAndTolerations
   );
 }
+
+function getVolumeSnapshotsForHyperdiskMigration() {
+  if (hyperdiskSupportConfig.hyperdiskSupport.migratingInfra) {
+    const { dataSource: prometheusDataSource } = createVolumeSnapshot({
+      resourceName: `prometheus-hd-migration-snapshot`,
+      snapshotName: `prometheus-migration-snapshot`,
+      namespace: namespaceName,
+      pvcName: `prometheus-prometheus-prometheus-db-prometheus-prometheus-prometheus-0`,
+    });
+    const { dataSource: alertManagerDataSource } = createVolumeSnapshot({
+      resourceName: `alertmanager-hd-migration-snapshot`,
+      snapshotName: `alertmanager-migration-snapshot`,
+      namespace: namespaceName,
+      pvcName: `alertmanager-prometheus-alertmanager-db-alertmanager-prometheus-alertmanager-0`,
+    });
+    return {
+      prometheus: {
+        dataSource: prometheusDataSource,
+      },
+      alertManager: {
+        dataSource: alertManagerDataSource,
+      },
+    };
+  } else {
+    return {};
+  }
+}

cluster/pulumi/validator-runbook/src/partyAllocator.ts

@@ -4,18 +4,34 @@ import * as pulumi from '@pulumi/pulumi';
 import {
   activeVersion,
   CnInput,
+  createVolumeSnapshot,
   DecentralizedSynchronizerUpgradeConfig,
   ExactNamespace,
   InstalledHelmChart,
   installSpliceHelmChart,
+  standardStorageClassName,
 } from '@lfdecentralizedtrust/splice-pulumi-common';
 import { PartyAllocatorConfig } from '@lfdecentralizedtrust/splice-pulumi-common-validator';
 
+import { hyperdiskSupportConfig } from '../../common/src/config/hyperdiskSupportConfig';
+
 export function installPartyAllocator(
   xns: ExactNamespace,
   config: PartyAllocatorConfig,
   dependsOn: CnInput<pulumi.Resource>[]
 ): InstalledHelmChart {
+  const dataSource =
+    hyperdiskSupportConfig.hyperdiskSupport.enabled &&
+    hyperdiskSupportConfig.hyperdiskSupport.migrating
+      ? {
+          dataSource: createVolumeSnapshot({
+            resourceName: `party-allocator-keys-migration-snapshot`,
+            snapshotName: `party-allocator-keys-snapshot`,
+            namespace: xns.logicalName,
+            pvcName: `party-allocator-keys`,
+          }).dataSource,
+        }
+      : {};
   return installSpliceHelmChart(
     xns,
     'party-allocator',
@@ -33,6 +49,14 @@ export function installPartyAllocator(
         preapprovalRetries: config.preapprovalRetries,
         preapprovalRetryDelayMs: config.preapprovalRetryDelayMs,
       },
+      pvc: {
+        size: config.pvcSize,
+        storageClassName: standardStorageClassName,
+        name: hyperdiskSupportConfig.hyperdiskSupport.enabled
+          ? 'party-allocator-keys-hd-pvc'
+          : 'party-allocator-keys',
+        ...dataSource,
+      },
     },
     activeVersion,
     { dependsOn }