feat(site-replication): show replication health on the console page

app/(dashboard)/status/page.tsx

@@ -12,6 +12,7 @@ import {
   isScannerCycleActive,
   resolveUsageFreshness,
 } from "@/lib/performance-data"
+import { formatDuration } from "@/lib/functions"
 import { cn } from "@/lib/utils"
 import {
   RiArchiveDrawerFill,
@@ -32,20 +33,6 @@ import { PerformanceStatusSources } from "../_components/performance-status-sour
 import { PerformanceStatusSummary } from "../_components/performance-status-summary"
 import { PerformanceUsageCard } from "../_components/performance-usage-card"
 
-function formatDuration(seconds: number | undefined, t: (key: string) => string) {
-  if (seconds === undefined || !Number.isFinite(seconds) || seconds < 0) return t("Unknown")
-  const totalMinutes = Math.floor(seconds / 60)
-  const days = Math.floor(totalMinutes / 1440)
-  const hours = Math.floor((totalMinutes % 1440) / 60)
-  const minutes = totalMinutes % 60
-  const parts = [
-    days ? `${days} ${t("Days")}` : "",
-    hours ? `${hours} ${t("Hours")}` : "",
-    !days && minutes ? `${minutes} ${t("Minutes")}` : "",
-  ].filter(Boolean)
-  return parts.slice(0, 2).join(" ") || `0 ${t("Minutes")}`
-}
-
 export default function PerformancePage() {
   const { t, i18n } = useTranslation()
   const { canAccessPath } = usePermissions()

hooks/use-site-replication.ts

@@ -5,19 +5,28 @@ import { useApi } from "@/contexts/api-context"
 import type {
   SiteReplicationAddResponse,
   SiteReplicationAddSiteInput,
+  SiteReplicationBucketStats,
   SiteReplicationEditResponse,
+  SiteReplicationEntityStats,
   SiteReplicationInfo,
   SiteReplicationMetric,
   SiteReplicationMetricsSummary,
+  SiteReplicationPeerError,
   SiteReplicationPeerInfo,
+  SiteReplicationPendingOperation,
+  SiteReplicationPolicyStats,
+  SiteReplicationQueueStat,
   SiteReplicationRemoveInput,
   SiteReplicationRemoveResponse,
   SiteReplicationResyncOperation,
   SiteReplicationResyncResponse,
+  SiteReplicationRetryStats,
   SiteReplicationSiteSummary,
   SiteReplicationStateInfo,
   SiteReplicationStatus,
   SiteReplicationSyncState,
+  SiteReplicationTimedErrors,
+  SiteReplicationVolume,
   SiteReplicationWorkerStat,
 } from "@/types/site-replication"
 
@@ -39,6 +48,14 @@ function asBoolean(value: unknown): boolean {
   return typeof value === "boolean" ? value : false
 }
 
+function asStringArray(value: unknown): string[] {
+  return Array.isArray(value) ? value.filter((item): item is string => typeof item === "string") : []
+}
+
+function mapRecord<T>(value: unknown, normalize: (item: unknown) => T): Record<string, T> {
+  return Object.fromEntries(Object.entries(asRecord(value)).map(([key, item]) => [key, normalize(item)]))
+}
+
 function normalizeSyncState(value: unknown): SiteReplicationSyncState {
   const normalized = asString(value).toLowerCase()
   if (normalized === "enable" || normalized === "disable") {
@@ -57,6 +74,102 @@ function normalizeWorkerStat(value: unknown): SiteReplicationWorkerStat {
   }
 }
 
+function normalizeVolume(value: unknown): SiteReplicationVolume {
+  const record = asRecord(value)
+
+  return { count: asNumber(record.count), bytes: asNumber(record.bytes) }
+}
+
+function normalizeTimedErrors(value: unknown): SiteReplicationTimedErrors {
+  const record = asRecord(value)
+
+  return {
+    lastMinute: normalizeVolume(record.lastMinute),
+    lastHour: normalizeVolume(record.lastHour),
+    totals: normalizeVolume(record.totals),
+  }
+}
+
+function normalizeQueueStat(value: unknown): SiteReplicationQueueStat {
+  const record = asRecord(value)
+
+  return {
+    curr: normalizeVolume(record.curr),
+    avg: normalizeVolume(record.avg),
+    max: normalizeVolume(record.max),
+  }
+}
+
+// Both fields are omitted by the server while nothing is outstanding, so an
+// absent value means "healthy", not "unknown".
+function normalizePendingOperation(value: unknown): SiteReplicationPendingOperation | undefined {
+  const record = asRecord(value)
+  const operation = asString(record.operation)
+  if (!operation) {
+    return undefined
+  }
+
+  return {
+    operation,
+    id: asString(record.id),
+    pendingPeers: asStringArray(record.pendingPeers),
+    ackedPeers: asStringArray(record.ackedPeers),
+    updatedAt: asString(record.updatedAt) || undefined,
+  }
+}
+
+function normalizeRetryStats(value: unknown): SiteReplicationRetryStats | undefined {
+  if (!value || typeof value !== "object") {
+    return undefined
+  }
+  const record = asRecord(value)
+
+  return {
+    pending: asNumber(record.pending),
+    failed: asNumber(record.failed),
+    lastError: asString(record.lastError),
+  }
+}
+
+function normalizePeerError(value: unknown): SiteReplicationPeerError {
+  const record = asRecord(value)
+
+  return { name: asString(record.name), endpoint: asString(record.endpoint), error: asString(record.error) }
+}
+
+function normalizeBucketStats(value: unknown): SiteReplicationBucketStats {
+  const record = asRecord(value)
+
+  return {
+    hasBucket: asBoolean(record.HasBucket),
+    bucketMarkedDeleted: asBoolean(record.BucketMarkedDeleted),
+    tagMismatch: asBoolean(record.TagMismatch),
+    versioningConfigMismatch: asBoolean(record.VersioningConfigMismatch),
+    objectLockConfigMismatch: asBoolean(record.OLockConfigMismatch),
+    policyMismatch: asBoolean(record.PolicyMismatch),
+    sseConfigMismatch: asBoolean(record.SSEConfigMismatch),
+    replicationCfgMismatch: asBoolean(record.ReplicationCfgMismatch),
+    quotaCfgMismatch: asBoolean(record.QuotaCfgMismatch),
+    corsCfgMismatch: asBoolean(record.CorsCfgMismatch),
+  }
+}
+
+function normalizePolicyStats(value: unknown): SiteReplicationPolicyStats {
+  const record = asRecord(value)
+
+  return { hasPolicy: asBoolean(record.HasPolicy), policyMismatch: asBoolean(record.PolicyMismatch) }
+}
+
+function normalizeEntityStats<T>(value: unknown, normalize: (item: unknown) => T): SiteReplicationEntityStats<T> {
+  return mapRecord(value, (entity) => mapRecord(entity, normalize))
+}
+
+// User, group, and ILM expiry rule entries only exist on the sites that hold
+// the entity, so the inner keys are the holding deployment IDs.
+function normalizeEntitySites(value: unknown): Record<string, string[]> {
+  return mapRecord(value, (entity) => Object.keys(asRecord(entity)))
+}
+
 function normalizePeerInfo(value: unknown): SiteReplicationPeerInfo {
   const record = asRecord(value)
   const bandwidth = asRecord(record.defaultbandwidth)
@@ -174,6 +287,7 @@ function normalizeMetric(value: unknown): SiteReplicationMetric {
     online: asBoolean(record.isOnline),
     replicatedSize: asNumber(record.replicatedSize),
     replicatedCount: asNumber(record.replicatedCount),
+    failed: normalizeTimedErrors(record.failed),
   }
 }
 
@@ -190,6 +304,7 @@ function normalizeMetricsSummary(value: unknown): SiteReplicationMetricsSummary
     activeWorkers: normalizeWorkerStat(record.activeWorkers),
     replicaSize: asNumber(record.replicaSize),
     replicaCount: asNumber(record.replicaCount),
+    queued: normalizeQueueStat(record.queued),
     metrics: normalizeMetricMap(record.replMetrics),
   }
 }
@@ -203,6 +318,8 @@ function normalizeInfo(value: unknown): SiteReplicationInfo {
     name: asString(record.name),
     sites,
     serviceAccountAccessKey: asString(record.serviceAccountAccessKey),
+    pendingOperation: normalizePendingOperation(record.pendingOperation),
+    retryStats: normalizeRetryStats(record.retryStats),
     apiVersion: asString(record.apiVersion) || undefined,
   }
 }
@@ -220,7 +337,15 @@ function normalizeStatus(value: unknown): SiteReplicationStatus {
     sites: normalizePeerMap(record.Sites),
     statsSummary: normalizeSiteSummaryMap(record.StatsSummary),
     peerStates: normalizeStateInfoMap(record.PeerStates),
+    peerErrors: mapRecord(record.PeerErrors, normalizePeerError),
+    pendingOperation: normalizePendingOperation(record.PendingOperation),
+    retryStats: normalizeRetryStats(record.RetryStats),
     metrics: normalizeMetricsSummary(record.Metrics),
+    bucketStats: normalizeEntityStats(record.BucketStats, normalizeBucketStats),
+    policyStats: normalizeEntityStats(record.PolicyStats, normalizePolicyStats),
+    userSites: normalizeEntitySites(record.UserStats),
+    groupSites: normalizeEntitySites(record.GroupStats),
+    ilmExpiryRuleSites: normalizeEntitySites(record.ILMExpiryStats),
     apiVersion: asString(record.apiVersion) || undefined,
   }
 }
@@ -275,7 +400,14 @@ function normalizeResyncResponse(value: unknown): SiteReplicationResyncResponse
     opType: asString(record.op),
     resyncId: asString(record.id),
     status: asString(record.status),
+    state: asString(record.state),
     buckets,
+    totalBuckets: asNumber(record.totalBuckets),
+    completedBuckets: asNumber(record.completedBuckets),
+    failedBuckets: asNumber(record.failedBuckets),
+    replicatedObjects: asNumber(record.replicatedObjects),
+    failedObjects: asNumber(record.failedObjects),
+    updatedAt: asString(record.updatedAt) || undefined,
     errorDetail: asString(record.errorDetail),
   }
 }

i18n/locales/ar-MA.json

@@ -2241,5 +2241,45 @@
   "Operations": "العمليات",
   "Resources": "الموارد",
   "Bucket size distribution": "توزيع أحجام الحاويات",
-  "Other buckets": "حاويات أخرى"
+  "Other buckets": "حاويات أخرى",
+  "Active workers": "العمال النشطون",
+  "Configuration deliveries": "تسليم الإعدادات",
+  "Connection": "الاتصال",
+  "Consistency": "الاتساق",
+  "Consistency cannot be checked until every peer site is reachable.": "لا يمكن التحقق من الاتساق حتى يصبح كل موقع نظير قابلاً للوصول.",
+  "Differs: {settings}": "مختلف: {settings}",
+  "Downtime": "مدة الانقطاع",
+  "Every peer site is reachable and nothing is waiting to be replicated or repaired.": "كل المواقع النظيرة قابلة للوصول ولا يوجد ما ينتظر النسخ أو الإصلاح.",
+  "Every site holds the same buckets, IAM entities, and expiry rules.": "تحتوي كل المواقع على نفس الحاويات وكيانات IAM وقواعد الانتهاء.",
+  "Failed in the last hour": "فشل خلال الساعة الأخيرة",
+  "Failed since start": "فشل منذ البدء",
+  "Failures and Backlog": "الإخفاقات والتراكم",
+  "Missing on {sites}": "مفقود في {sites}",
+  "No resync has been started for this site.": "لم تبدأ أي إعادة مزامنة لهذا الموقع.",
+  "Object counters come from the node that answered this request and reset when it restarts.": "تأتي عدادات الكائنات من العقدة التي أجابت على هذا الطلب وتُعاد إلى الصفر عند إعادة تشغيلها.",
+  "Peer reachability is not reported by this server, so replication health cannot be confirmed.": "لا يبلّغ هذا الخادم عن إمكانية الوصول إلى النظراء، لذا لا يمكن تأكيد سلامة النسخ.",
+  "Queued now": "في الانتظار الآن",
+  "Queued peak": "ذروة الانتظار",
+  "Refreshes every 30 seconds": "يتحدث كل 30 ثانية",
+  "Replication Health": "سلامة النسخ",
+  "Resync Status": "حالة إعادة المزامنة",
+  "Site {site} is unreachable, so nothing replicates to it.": "الموقع {site} غير قابل للوصول، لذا لا يتم نسخ أي شيء إليه.",
+  "Sync Setting": "إعداد المزامنة",
+  "and {count} more": "و{count} أخرى",
+  "{count} buckets, IAM entities, or expiry rules differ between sites.": "{count} من الحاويات أو كيانات IAM أو قواعد الانتهاء تختلف بين المواقع.",
+  "{count} object replication failures in the last hour.": "{count} من حالات فشل نسخ الكائنات خلال الساعة الأخيرة.",
+  "{operation} has not finished.": "{operation} لم تكتمل.",
+  "{operation} has not finished; waiting for {sites}.": "{operation} لم تكتمل؛ في انتظار {sites}.",
+  "{pending} configuration changes are waiting to be delivered to peer sites, {failed} failed.": "{pending} من تغييرات الإعدادات بانتظار التسليم إلى المواقع النظيرة، وفشل {failed}.",
+  "{pending} waiting, {failed} failed": "{pending} قيد الانتظار، {failed} فشل",
+  "{state}: {completed} of {total} buckets completed, {failed} failed. {objects} objects replicated, {failedObjects} failed.": "{state}: اكتملت {completed} من {total} حاوية، وفشلت {failed}. تم نسخ {objects} كائن، وفشل {failedObjects}.",
+  "Site removal": "إزالة الموقع",
+  "Endpoint update": "تحديث نقطة النهاية",
+  "Group": "المجموعة",
+  "ILM Expiry Rule": "قاعدة انتهاء ILM",
+  "Marked for deletion": "محدد للحذف",
+  "Quota": "الحصة",
+  "CORS": "CORS",
+  "Policy document": "مستند السياسة",
+  "Service account rotation": "تدوير حساب الخدمة"
 }

i18n/locales/de-DE.json

@@ -2241,5 +2241,45 @@
   "Operations": "Betrieb",
   "Resources": "Ressourcen",
   "Bucket size distribution": "Verteilung der Bucket-Größen",
-  "Other buckets": "Weitere Buckets"
+  "Other buckets": "Weitere Buckets",
+  "Active workers": "Aktive Worker",
+  "Configuration deliveries": "Konfigurationszustellungen",
+  "Connection": "Verbindung",
+  "Consistency": "Konsistenz",
+  "Consistency cannot be checked until every peer site is reachable.": "Die Konsistenz kann erst geprüft werden, wenn alle Peer-Standorte erreichbar sind.",
+  "Differs: {settings}": "Abweichend: {settings}",
+  "Downtime": "Ausfallzeit",
+  "Every peer site is reachable and nothing is waiting to be replicated or repaired.": "Alle Peer-Standorte sind erreichbar, und nichts wartet auf Replikation oder Reparatur.",
+  "Every site holds the same buckets, IAM entities, and expiry rules.": "Alle Standorte enthalten dieselben Buckets, IAM-Entitäten und Ablaufregeln.",
+  "Failed in the last hour": "Fehlgeschlagen in der letzten Stunde",
+  "Failed since start": "Fehlgeschlagen seit Start",
+  "Failures and Backlog": "Fehler und Rückstand",
+  "Missing on {sites}": "Fehlt auf {sites}",
+  "No resync has been started for this site.": "Für diesen Standort wurde noch keine Neusynchronisierung gestartet.",
+  "Object counters come from the node that answered this request and reset when it restarts.": "Die Objektzähler stammen vom Knoten, der diese Anfrage beantwortet hat, und werden bei dessen Neustart zurückgesetzt.",
+  "Peer reachability is not reported by this server, so replication health cannot be confirmed.": "Dieser Server meldet die Erreichbarkeit der Peers nicht, daher kann der Replikationszustand nicht bestätigt werden.",
+  "Queued now": "Aktuell in Warteschlange",
+  "Queued peak": "Höchststand Warteschlange",
+  "Refreshes every 30 seconds": "Aktualisierung alle 30 Sekunden",
+  "Replication Health": "Replikationszustand",
+  "Resync Status": "Status der Neusynchronisierung",
+  "Site {site} is unreachable, so nothing replicates to it.": "Standort {site} ist nicht erreichbar, daher wird nichts dorthin repliziert.",
+  "Sync Setting": "Sync-Einstellung",
+  "and {count} more": "und {count} weitere",
+  "{count} buckets, IAM entities, or expiry rules differ between sites.": "{count} Buckets, IAM-Entitäten oder Ablaufregeln unterscheiden sich zwischen den Standorten.",
+  "{count} object replication failures in the last hour.": "{count} fehlgeschlagene Objektreplikationen in der letzten Stunde.",
+  "{operation} has not finished.": "{operation} ist nicht abgeschlossen.",
+  "{operation} has not finished; waiting for {sites}.": "{operation} ist nicht abgeschlossen; wartet auf {sites}.",
+  "{pending} configuration changes are waiting to be delivered to peer sites, {failed} failed.": "{pending} Konfigurationsänderungen warten auf Zustellung an Peer-Standorte, {failed} fehlgeschlagen.",
+  "{pending} waiting, {failed} failed": "{pending} wartend, {failed} fehlgeschlagen",
+  "{state}: {completed} of {total} buckets completed, {failed} failed. {objects} objects replicated, {failedObjects} failed.": "{state}: {completed} von {total} Buckets abgeschlossen, {failed} fehlgeschlagen. {objects} Objekte repliziert, {failedObjects} fehlgeschlagen.",
+  "Site removal": "Standortentfernung",
+  "Endpoint update": "Endpunkt-Aktualisierung",
+  "Group": "Gruppe",
+  "ILM Expiry Rule": "ILM-Ablaufregel",
+  "Marked for deletion": "Zum Löschen markiert",
+  "Quota": "Kontingent",
+  "CORS": "CORS",
+  "Policy document": "Richtliniendokument",
+  "Service account rotation": "Rotation des Dienstkontos"
 }

i18n/locales/en-US.json

@@ -2241,5 +2241,45 @@
   "Operations": "Operations",
   "Resources": "Resources",
   "Bucket size distribution": "Bucket size distribution",
-  "Other buckets": "Other buckets"
+  "Other buckets": "Other buckets",
+  "Active workers": "Active workers",
+  "Configuration deliveries": "Configuration deliveries",
+  "Connection": "Connection",
+  "Consistency": "Consistency",
+  "Consistency cannot be checked until every peer site is reachable.": "Consistency cannot be checked until every peer site is reachable.",
+  "Differs: {settings}": "Differs: {settings}",
+  "Downtime": "Downtime",
+  "Every peer site is reachable and nothing is waiting to be replicated or repaired.": "Every peer site is reachable and nothing is waiting to be replicated or repaired.",
+  "Every site holds the same buckets, IAM entities, and expiry rules.": "Every site holds the same buckets, IAM entities, and expiry rules.",
+  "Failed in the last hour": "Failed in the last hour",
+  "Failed since start": "Failed since start",
+  "Failures and Backlog": "Failures and Backlog",
+  "Missing on {sites}": "Missing on {sites}",
+  "No resync has been started for this site.": "No resync has been started for this site.",
+  "Object counters come from the node that answered this request and reset when it restarts.": "Object counters come from the node that answered this request and reset when it restarts.",
+  "Peer reachability is not reported by this server, so replication health cannot be confirmed.": "Peer reachability is not reported by this server, so replication health cannot be confirmed.",
+  "Queued now": "Queued now",
+  "Queued peak": "Queued peak",
+  "Refreshes every 30 seconds": "Refreshes every 30 seconds",
+  "Replication Health": "Replication Health",
+  "Resync Status": "Resync Status",
+  "Site {site} is unreachable, so nothing replicates to it.": "Site {site} is unreachable, so nothing replicates to it.",
+  "Sync Setting": "Sync Setting",
+  "and {count} more": "and {count} more",
+  "{count} buckets, IAM entities, or expiry rules differ between sites.": "{count} buckets, IAM entities, or expiry rules differ between sites.",
+  "{count} object replication failures in the last hour.": "{count} object replication failures in the last hour.",
+  "{operation} has not finished.": "{operation} has not finished.",
+  "{operation} has not finished; waiting for {sites}.": "{operation} has not finished; waiting for {sites}.",
+  "{pending} configuration changes are waiting to be delivered to peer sites, {failed} failed.": "{pending} configuration changes are waiting to be delivered to peer sites, {failed} failed.",
+  "{pending} waiting, {failed} failed": "{pending} waiting, {failed} failed",
+  "{state}: {completed} of {total} buckets completed, {failed} failed. {objects} objects replicated, {failedObjects} failed.": "{state}: {completed} of {total} buckets completed, {failed} failed. {objects} objects replicated, {failedObjects} failed.",
+  "Site removal": "Site removal",
+  "Endpoint update": "Endpoint update",
+  "Group": "Group",
+  "ILM Expiry Rule": "ILM Expiry Rule",
+  "Marked for deletion": "Marked for deletion",
+  "Quota": "Quota",
+  "CORS": "CORS",
+  "Policy document": "Policy document",
+  "Service account rotation": "Service account rotation"
 }