feat(ingest): batch event writes and harden transport

cmd/event-processor/main.go

@@ -7,6 +7,7 @@ import (
 	"log"
 	"os"
 	"os/signal"
+	"sync"
 	"syscall"
 	"time"
 
@@ -69,15 +70,60 @@ func main() {
 
 	log.Printf("subscribed to %s (%s)", natsTopic, natsURL)
 
+	const (
+		maxBatchSize  = 100
+		flushInterval = 200 * time.Millisecond
+	)
+
+	eventCh := make(chan postgres.InsertEvent, 1000)
+
+	// Flush goroutine
+	var wg sync.WaitGroup
+	wg.Add(1)
+	go func() {
+		defer wg.Done()
+		batch := make([]postgres.InsertEvent, 0, maxBatchSize)
+		ticker := time.NewTicker(flushInterval)
+		defer ticker.Stop()
+
+		flush := func() {
+			if len(batch) == 0 {
+				return
+			}
+			if err := db.InsertEventBatch(ctx, batch); err != nil {
+				log.Printf("batch insert error (%d events): %v", len(batch), err)
+			}
+			batch = batch[:0]
+		}
+
+		for {
+			select {
+			case evt, ok := <-eventCh:
+				if !ok {
+					flush()
+					return
+				}
+				batch = append(batch, evt)
+				if len(batch) >= maxBatchSize {
+					flush()
+				}
+			case <-ticker.C:
+				flush()
+			}
+		}
+	}()
+
 	err = sub.Subscribe(ctx, func(msg []byte) error {
 		var env envelope
 		if err := json.Unmarshal(msg, &env); err != nil {
-			return err
+			log.Printf("unmarshal error: %v", err)
+			return nil // don't fail the subscriber
 		}
 
 		ts, err := parseTS(env.Event.TS)
 		if err != nil {
-			return err
+			log.Printf("timestamp parse error: %v", err)
+			return nil
 		}
 
 		e := postgres.InsertEvent{
@@ -97,9 +143,17 @@ func main() {
 			e.ParentSpanID = sqlNull(env.Correlation.ParentSpanID)
 		}
 
-		return db.InsertEvent(ctx, e)
+		select {
+		case eventCh <- e:
+		default:
+			log.Printf("event channel full, dropping event %s", env.Event.ID)
+		}
+		return nil
 	})
 
+	close(eventCh)
+	wg.Wait()
+
 	if err != nil && err != context.Canceled {
 		log.Printf("processor stopped: %v", err)
 	}