fix(monitoring): use a per-second ring buffer for DNS query metrics, improve DNS logging rate limiting and security event aggregation, and bump smartmta dependency

ts/monitoring/classes.metricsmanager.ts

@@ -35,7 +35,9 @@ export class MetricsManager {
     queryTypes: {} as Record<string, number>,
     topDomains: new Map<string, number>(),
     lastResetDate: new Date().toDateString(),
-    queryTimestamps: [] as number[], // Track query timestamps for rate calculation
+    // Per-second query count ring buffer (300 entries = 5 minutes)
+    queryRing: new Int32Array(300),
+    queryRingLastSecond: 0, // last epoch second that was written
     responseTimes: [] as number[], // Track response times in ms
     recentQueries: [] as Array<{ timestamp: number; domain: string; type: string; answered: boolean; responseTimeMs: number }>,
   };
@@ -95,12 +97,13 @@ export class MetricsManager {
         this.dnsMetrics.cacheMisses = 0;
         this.dnsMetrics.queryTypes = {};
         this.dnsMetrics.topDomains.clear();
-        this.dnsMetrics.queryTimestamps = [];
+        this.dnsMetrics.queryRing.fill(0);
+        this.dnsMetrics.queryRingLastSecond = 0;
         this.dnsMetrics.responseTimes = [];
         this.dnsMetrics.recentQueries = [];
         this.dnsMetrics.lastResetDate = currentDate;
       }
-      
+
       if (currentDate !== this.securityMetrics.lastResetDate) {
         this.securityMetrics.blockedIPs = 0;
         this.securityMetrics.authFailures = 0;
@@ -111,15 +114,6 @@ export class MetricsManager {
         this.securityMetrics.lastResetDate = currentDate;
       }
 
-      // Prune old query timestamps (keep last 5 minutes)
-      const fiveMinutesAgo = Date.now() - 300000;
-      const idx = this.dnsMetrics.queryTimestamps.findIndex(ts => ts >= fiveMinutesAgo);
-      if (idx > 0) {
-        this.dnsMetrics.queryTimestamps = this.dnsMetrics.queryTimestamps.slice(idx);
-      } else if (idx === -1) {
-        this.dnsMetrics.queryTimestamps = [];
-      }
-
       // Prune old time-series buckets every minute (don't wait for lazy query)
       this.pruneOldBuckets();
     }, 60000); // Check every minute
@@ -150,16 +144,16 @@ export class MetricsManager {
       const smartMetricsData = await this.smartMetrics.getMetrics();
       const proxyMetrics = this.dcRouter.smartProxy ? this.dcRouter.smartProxy.getMetrics() : null;
       const proxyStats = this.dcRouter.smartProxy ? await this.dcRouter.smartProxy.getStatistics() : null;
+      const { heapUsed, heapTotal, external, rss } = process.memoryUsage();
 
       return {
         uptime: process.uptime(),
         startTime: Date.now() - (process.uptime() * 1000),
         memoryUsage: {
-          heapUsed: process.memoryUsage().heapUsed,
-          heapTotal: process.memoryUsage().heapTotal,
-          external: process.memoryUsage().external,
-          rss: process.memoryUsage().rss,
-          // Add SmartMetrics memory data
+          heapUsed,
+          heapTotal,
+          external,
+          rss,
           maxMemoryMB: this.smartMetrics.maxMemoryMB,
           actualUsageBytes: smartMetricsData.memoryUsageBytes,
           actualUsagePercentage: smartMetricsData.memoryPercentage,
@@ -228,11 +222,8 @@ export class MetricsManager {
         .slice(0, 10)
         .map(([domain, count]) => ({ domain, count }));
       
-      // Calculate queries per second from recent timestamps
-      const now = Date.now();
-      const oneMinuteAgo = now - 60000;
-      const recentQueries = this.dnsMetrics.queryTimestamps.filter(ts => ts >= oneMinuteAgo);
-      const queriesPerSecond = recentQueries.length / 60;
+      // Calculate queries per second from ring buffer (sum last 60 seconds)
+      const queriesPerSecond = this.getQueryRingSum(60) / 60;
       
       // Calculate average response time
       const avgResponseTime = this.dnsMetrics.responseTimes.length > 0
@@ -436,8 +427,8 @@ export class MetricsManager {
       this.dnsMetrics.cacheMisses++;
     }
     
-    // Track query timestamp (pruning moved to resetInterval to avoid O(n) per query)
-    this.dnsMetrics.queryTimestamps.push(Date.now());
+    // Increment per-second query counter in ring buffer
+    this.incrementQueryRing();
     
     // Track response time if provided
     if (responseTimeMs) {
@@ -609,7 +600,7 @@ export class MetricsManager {
         requestsPerSecond,
         requestsTotal,
       };
-    }, 200); // Use 200ms cache for more frequent updates
+    }, 1000); // 1s cache — matches typical dashboard poll interval
   }
 
   // --- Time-series helpers ---
@@ -638,6 +629,63 @@ export class MetricsManager {
     bucket.queries++;
   }
 
+  /**
+   * Increment the per-second query counter in the ring buffer.
+   * Zeros any stale slots between the last write and the current second.
+   */
+  private incrementQueryRing(): void {
+    const currentSecond = Math.floor(Date.now() / 1000);
+    const ring = this.dnsMetrics.queryRing;
+    const last = this.dnsMetrics.queryRingLastSecond;
+
+    if (last === 0) {
+      // First call — zero and anchor
+      ring.fill(0);
+      this.dnsMetrics.queryRingLastSecond = currentSecond;
+      ring[currentSecond % ring.length] = 1;
+      return;
+    }
+
+    const gap = currentSecond - last;
+    if (gap >= ring.length) {
+      // Entire ring is stale — clear all
+      ring.fill(0);
+    } else if (gap > 0) {
+      // Zero slots from (last+1) to currentSecond (inclusive)
+      for (let s = last + 1; s <= currentSecond; s++) {
+        ring[s % ring.length] = 0;
+      }
+    }
+
+    this.dnsMetrics.queryRingLastSecond = currentSecond;
+    ring[currentSecond % ring.length]++;
+  }
+
+  /**
+   * Sum query counts from the ring buffer for the last N seconds.
+   */
+  private getQueryRingSum(seconds: number): number {
+    const currentSecond = Math.floor(Date.now() / 1000);
+    const ring = this.dnsMetrics.queryRing;
+    const last = this.dnsMetrics.queryRingLastSecond;
+
+    if (last === 0) return 0;
+
+    // First, zero stale slots so reads are accurate even without writes
+    const gap = currentSecond - last;
+    if (gap >= ring.length) return 0; // all data is stale
+
+    let sum = 0;
+    const limit = Math.min(seconds, ring.length);
+    for (let i = 0; i < limit; i++) {
+      const sec = currentSecond - i;
+      if (sec < last - (ring.length - 1)) break; // slot is from older cycle
+      if (sec > last) continue; // no writes yet for this second
+      sum += ring[sec % ring.length];
+    }
+    return sum;
+  }
+
   private pruneOldBuckets(): void {
     const cutoff = Date.now() - 86400000; // 24h
     for (const key of this.emailMinuteBuckets.keys()) {