private_key: { filename: "/etc/envoy/certs/server.key" }
        tls_params:
          tls_minimum_protocol_version: TLSv1_2
          cipher_suites: "ECDHE-ECDSA-AES128-GCM-SHA256:ECDHE-RSA-AES128-GCM-SHA256"
        tls_session_ticket_keys:
        - filename: "/etc/envoy/tickets/key.bin"
  filters:
  - name: envoy.filters.network.http_connection_manager
    typed_config:
      "@type": type.googleapis.com/envoy.extensions.filters.network.http_connection_manager.v3.HttpConnectionManager
      stat_prefix: ingress_http
      http2_protocol_options:
        max_concurrent_streams: 100
      codec_type: AUTO
      route_config:
        name: local_route
        virtual_hosts:
        - name: backend
          domains: ["*"]
          routes:
          - match: { prefix: "/api/" }
            route:
              cluster: backend_cluster
              timeout: 10s
              retry_policy:
                retry_on: "5xx,reset,connect-failure"
                num_retries: 2
                per_try_timeout: 3s
      http_filters:
      - name: envoy.filters.http.router
        typed_config:
          "@type": type.googleapis.com/envoy.extensions.filters.http.router.v3.Router
      - name: envoy.filters.http.local_ratelimit
        typed_config:
          "@type": type.googleapis.com/envoy.extensions.filters.http.local_ratelimit.v3.LocalRateLimit
          stat_prefix: http_local_rate_limiter
          token_bucket:
            max_tokens: 1000
            tokens_per_fill: 500
            fill_interval: 1s
      - name: envoy.filters.http.circuit_breakers
        typed_config:
          "@type": type.googleapis.com/envoy.extensions.filters.http.circuit_breakers.v3.CircuitBreakers
          thresholds:
          - priority: DEFAULT
            max_connections: 1024
            max_pending_requests: 512
            max_requests: 2048
            max_retries: 3

clusters:

• name: backend_cluster connect_timeout: 2s type: STRICT_DNS lb_policy: LEAST_REQUEST outlier_detection: consecutive_5xx: 5 interval: 10s base_ejection_time: 30s max_ejection_percent: 50 health_checks:
  • timeout: 3s interval: 5s unhealthy_threshold: 2 healthy_threshold: 2 http_health_check: path: "/healthz" expected_statuses: range: start: 200 end: 299 load_assignment: cluster_name: backend_cluster endpoints:
    • lb_endpoints:
      • endpoint: address: socket_address: address: backend-svc.default.svc.cluster.local port_value: 8080 circuit_breakers: thresholds:
    • priority: DEFAULT max_connections: 2048 max_pending_requests: 1024 max_requests: 4096 max_retries: 2 transport_socket: name: envoy.transport_sockets.tls typed_config: "@type": type.googleapis.com/envoy.extensions.transport_sockets.tls.v3.UpstreamTlsContext sni: backend.internal

### Backend Health Check Endpoint (Go)
Load balancers require accurate backend state. A naive `/healthz` returning `200 OK` is insufficient. Implement a composite health check that verifies dependencies, connection pools, and queue depth.

```go
package main

import (
	"encoding/json"
	"net/http"
	"sync/atomic"
	"time"
)

var (
	ready    int32
	connections int64
	queueDepth  int64
)

func healthHandler(w http.ResponseWriter, r *http.Request) {
	status := map[string]interface{}{
		"status":        "healthy",
		"timestamp":     time.Now().UTC().Format(time.RFC3339),
		"connections":   atomic.LoadInt64(&connections),
		"queue_depth":   atomic.LoadInt64(&queueDepth),
		"ready":         atomic.LoadInt32(&ready) == 1,
	}

	// Fail if queue depth exceeds threshold or not ready
	if atomic.LoadInt64(&queueDepth) > 500 || atomic.LoadInt32(&ready) != 1 {
		status["status"] = "degraded"
		w.WriteHeader(http.StatusServiceUnavailable)
	} else {
		w.WriteHeader(http.StatusOK)
	}

	json.NewEncoder(w).Encode(status)
}

func main() {
	http.HandleFunc("/healthz", healthHandler)
	http.ListenAndServe(":8080", nil)
}

Integration Notes

• Metrics Export: Expose Prometheus metrics (http_requests_total, request_duration_seconds, backend_queue_depth) to feed Envoy's dynamic weighting or external autoscalers.
• Connection Pooling: Envoy's max_connections and max_requests thresholds prevent backend saturation. Tune based on your runtime's goroutine/thread limits.
• TLS Session Resumption: Pre-generate session ticket keys and rotate them weekly to balance security and handshake overhead.

---

Pitfall Guide (6 Critical Mistakes)

1. Ignoring Connection Exhaustion & Keep-Alive Misconfiguration

Symptom: Backend threads pile up, file descriptor limits hit, and latency spikes despite low request rates. Root Cause: Load balancers keep connections open indefinitely, or backends close them prematurely, causing TCP handshake storms. Mitigation: Align keepalive_timeout (LB) with keep-alive (backend). Set max_requests_per_connection to force periodic renegotiation. Monitor TIME_WAIT sockets and tune tcp_tw_reuse/tcp_fin_timeout at the OS level.

2. Health Check Misconfiguration & Thundering Herds

Symptom: All healthy backends receive a traffic surge simultaneously after a failed node recovers. Root Cause: Synchronous active health checks or zero jitter in recovery timing. Mitigation: Add jitter to health check intervals (interval: 5s ± 1s). Use passive failure detection alongside active checks. Implement gradual traffic injection (e.g., max_ejection_percent: 20 initially, scaling up).

3. Over-Reliance on Round-Robin Without Backend Awareness

Symptom: CPU-heavy instances saturate while I/O-bound nodes sit idle. P99 latency degrades unevenly. Root Cause: Round-robin assumes homogeneous backends, which is false in microservices and autoscaled environments. Mitigation: Switch to LEAST_REQUEST or RING_HASH with weighted endpoints. Inject real-time metrics (CPU, memory, queue depth) via gRPC or Prometheus to dynamically adjust weights.

4. TLS Termination Bottlenecks

Symptom: Load balancer CPU hits 90%+ during traffic spikes, increasing handshake latency. Root Cause: Full TLS handshakes for every connection, missing session resumption, or weak cipher suites. Mitigation: Enable TLS session tickets and OCSP stapling. Prefer ECDHE ciphers. Offload TLS to dedicated hardware or use kernel TLS (kTLS) where supported. Monitor ssl_handshakes vs ssl_session_reuses.

5. Lack of Observability & Blind Routing

Symptom: Traffic shifts to degraded nodes without alerting; latency spikes go undetected until user complaints. Root Cause: Load balancers operate on static configs without real-time backend telemetry. Mitigation: Integrate distributed tracing (OpenTelemetry) and expose LB metrics to Prometheus. Implement latency-based routing and alert on P95/P99 breaches. Use canary deployments with traffic mirroring for safe rollouts.

6. Static Scaling in Bursty Environments

Symptom: Backends scale too late, causing 5xx errors during traffic spikes; over-provisioning during lulls. Root Cause: Autoscaling reacts to CPU/memory thresholds rather than queue depth or request latency. Mitigation: Use predictive autoscaling (KEDA, Prometheus Adapter) that scales on http_requests_in_flight or custom queue metrics. Configure LB to drain terminating pods gracefully (drain_timeout: 30s).

---

Production Bundle

✅ Deployment Checklist

Pre-Flight

• TLS certificates valid, chain complete, session tickets generated
• Backend /healthz composite check verified (dependencies, queue, readiness)
• OS limits tuned: fs.file-max, net.core.somaxconn, net.ipv4.tcp_max_syn_backlog
• Load balancer resource requests/limits aligned with expected RPS and connection counts

Runtime

• Health check intervals jittered and passive detection enabled
• Circuit breakers and rate limits tested with chaos engineering (e.g., Litmus, Gremlin)
• Observability stack connected: Prometheus, Grafana, OpenTelemetry, log aggregation
• Drain and redeploy cycles validated (zero-downtime deployments)

Post-Deployment

• P50/P95/P99 latency baselines recorded
• Autoscaling policies verified under synthetic load (k6, wrk2)
• Rollback procedure documented and tested
• Incident runbook created for LB-specific failures (e.g., config drift, certificate expiry)

📊 Decision Matrix

Criteria	L4 Load Balancer (IPVS/iptables)	L7 Load Balancer (Envoy/Nginx)	Cloud Provider LB (ALB/NLB)
Protocol Support	TCP/UDP only	HTTP/1.1, HTTP/2, gRPC, WebSocket	HTTP/HTTPS, TCP, UDP
Intelligent Routing	❌	✅ (path, header, weight, canary)	✅ (path, host, weight)
TLS Termination	❌ (requires sidecar)	✅ (native, session resumption)	✅ (managed, ACM integration)
Observability	Limited (conntrack stats)	Rich (metrics, tracing, logs)	Moderate (CloudWatch, basic logs)
Cost & Ops	Low infra, high ops	Medium infra, medium ops	Low ops, pay-per-use
Best For	High-throughput TCP/UDP, gaming, IoT	Microservices, API gateways, complex routing	Quick deployment, managed TLS, standard web apps

📄 Config Template (Envoy Cluster + Rate Limit + Circuit Breaker)

# envoy-production.yaml
static_resources:
  listeners:
  - name: https_listener
    address: { socket_address: { address: 0.0.0.0, port_value: 443 } }
    filter_chains:
    - transport_socket: { name: envoy.transport_sockets.tls, typed_config: { "@type": type.googleapis.com/envoy.extensions.transport_sockets.tls.v3.ServerTlsContext } }
      filters:
      - name: envoy.filters.network.http_connection_manager
        typed_config:
          "@type": type.googleapis.com/envoy.extensions.filters.network.http_connection_manager.v3.HttpConnectionManager
          stat_prefix: prod_ingress
          route_config:
            name: prod_route
            virtual_hosts:
            - name: api
              domains: ["api.example.com"]
              routes:
              - match: { prefix: "/v1/" }
                route:
                  cluster: prod_backend
                  timeout: 8s
                  retry_policy: { retry_on: "5xx,reset", num_retries: 2, per_try_timeout: 2s }
          http_filters:
          - name: envoy.filters.http.local_ratelimit
            typed_config: { "@type": type.googleapis.com/envoy.extensions.filters.http.local_ratelimit.v3.LocalRateLimit, stat_prefix: rl, token_bucket: { max_tokens: 2000, tokens_per_fill: 1000, fill_interval: 1s } }
          - name: envoy.filters.http.router
  clusters:
  - name: prod_backend
    connect_timeout: 1.5s
    type: STRICT_DNS
    lb_policy: LEAST_REQUEST
    outlier_detection: { consecutive_5xx: 4, interval: 8s, base_ejection_time: 20s, max_ejection_percent: 40 }
    health_checks: [{ timeout: 2s, interval: 4s, unhealthy_threshold: 2, healthy_threshold: 2, http_health_check: { path: "/healthz" } }]
    load_assignment:
      cluster_name: prod_backend
      endpoints: [{ lb_endpoints: [{ endpoint: { address: { socket_address: { address: "backend-prod.default.svc.cluster.local", port_value: 8080 } } } }] }]
    circuit_breakers: { thresholds: [{ priority: DEFAULT, max_connections: 1500, max_pending_requests: 750, max_requests: 3000, max_retries: 2 }] }

🚀 Quick Start Guide (5 Steps)

1. Containerize & Expose Health Endpoints

   • Ensure backend services expose /healthz with readiness, liveness, and dependency checks.
   • Add Prometheus metrics exporter to track request latency, error rates, and queue depth.

2. Deploy Envoy as Sidecar or Edge Proxy

   kubectl apply -f envoy-configmap.yaml
   kubectl apply -f envoy-deployment.yaml
   kubectl apply -f envoy-service.yaml
   

   Mount TLS certs and session ticket keys via Kubernetes secrets.

3. Configure Dynamic Routing & Resilience

   • Set lb_policy: LEAST_REQUEST with weighted endpoints.
   • Enable outlier_detection and circuit_breakers to prevent cascade failures.
   • Apply local_ratelimit to protect backend from burst traffic.

4. Validate Under Load

   • Run synthetic traffic: wrk -t12 -c400 -d60s https://api.example.com/v1/test
   • Monitor P99 latency, error rates, and connection counts in Grafana.
   • Simulate backend failure: kubectl delete pod <backend-pod> and verify traffic rerouting.

5. Automate & Observe

   • Integrate with Prometheus Adapter for HPA: scale on http_requests_in_flight.
   • Configure alerts on envoy_cluster_upstream_cx_destroy_remote, ssl_handshake_errors, and P95 > SLA.
   • Implement GitOps for LB config changes; validate with envoy --mode validate -c envoy.yaml.

---

Final Thoughts

Load balancing for high-traffic backends is no longer a set-and-forget infrastructure component. It is a dynamic, metrics-driven control plane that must adapt to backend health, network conditions, and traffic patterns in real time. By combining intelligent algorithms, rigorous health management, TLS optimization, and deep observability, you transform the load balancer from a simple router into a resilience engine. Use the configurations, pitfalls, and production bundle provided here to deploy systems that absorb spikes, isolate failures, and maintain sub-100ms P99 latency under sustained load.