Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Unreleased

0.5.0 - 2025-12-21

Added

• Circuit breaker pattern for automatic failure detection and recovery
  • Prevents cascade failures by stopping traffic to failing backends
  • Three-state machine: CLOSED (normal), OPEN (blocking), HALF-OPEN (testing)
  • Configurable error threshold, minimum requests, and recovery timeout
  • Sliding window for error rate calculation
  • Automatic transition from OPEN to HALF-OPEN after timeout
  • Gradual traffic restoration with 10% test traffic in HALF-OPEN
  • Graceful degradation when all circuits open (traffic continues)
  • Integration with health check events for error tracking
  • Weight computation combines health, drain, and circuit breaker states
  • Prometheus metrics: lb_circuit_breaker_state, lb_circuit_breaker_error_rate
  • Runtime API: circuit-open?, circuit-half-open?, force-open-circuit!, force-close-circuit!, reset-circuit!
  • Event subscription for state change notifications
  • Configuration: {:settings {:circuit-breaker {:enabled true :error-threshold-pct 50 :min-requests 10 :open-duration-ms 30000}}}
  • New module: lb.circuit-breaker
  • Comprehensive examples in examples/circuit_breaker.clj

0.4.0 - 2025-12-21

Added

• Prometheus metrics export for monitoring and observability
  • HTTP endpoint at /metrics with standard Prometheus text format
  • Metrics include: lb_connections_active, lb_bytes_total, lb_packets_total
  • Backend health status: lb_backend_health (1=healthy, 0=unhealthy)
  • Health check latency histogram: lb_health_check_latency_seconds
  • DNS resolution status: lb_dns_resolution_status
  • System info: lb_up, lb_info
  • Configurable port and path via settings
  • Automatic integration with health check system for latency recording
  • Configuration: {:settings {:metrics {:enabled true :port 9090 :path "/metrics"}}}
  • Runtime API: metrics/start!, metrics/stop!, metrics/running?, metrics/get-status
• DNS-based backend resolution for dynamic environments
  • Use :host instead of :ip for DNS-backed targets
  • Periodic re-resolution with configurable intervals (:dns-refresh-seconds)
  • Multiple A record expansion to weighted targets (weight distributed equally)
  • Graceful failure handling with last-known-good IP fallback
  • Background daemon with jitter to avoid thundering herd
  • Health check integration for resolved IPs
  • Event subscription system for DNS resolution changes
  • Mixed static IP and DNS target support in same target group
  • Runtime API: get-dns-status, get-all-dns-status, force-dns-resolve!
  • New modules: lb.dns, lb.dns.resolver, lb.dns.manager
  • Configuration examples:
    • Basic: {:host "backend.local" :port 8080}
    • With refresh: {:host "backend.local" :port 8080 :dns-refresh-seconds 60}
    • Mixed: [{:ip "10.0.0.1" :port 8080 :weight 50} {:host "dynamic.svc" :port 8080 :weight 50}]
• Hot reload configuration without restart
  • File watching with inotify-based detection (Java NIO WatchService)
  • SIGHUP signal handling for manual reload trigger
  • Configuration diffing with incremental apply
  • Validation before apply with rollback on failure
  • Runtime API: reload-config!, reload-config-from-map!
  • Enable/disable: enable-hot-reload!, disable-hot-reload!
  • Status: hot-reload-enabled?, get-reload-state
  • New config diffing functions in lb.config
• Rate limiting with token bucket algorithm
  • Per-source IP rate limiting to prevent client abuse
  • Per-backend rate limiting to protect backend capacity
  • BPF-based enforcement with sub-millisecond overhead
  • LRU hash maps for automatic bucket expiration
  • Token scaling (1000x) for precise sub-token accounting
  • Runtime API: set-source-rate-limit!, set-backend-rate-limit!
  • Disable functions: disable-source-rate-limit!, disable-backend-rate-limit!, clear-rate-limits!
  • Status functions: get-rate-limit-config, rate-limiting-enabled?
  • Configuration via settings: {:rate-limits {:per-source {:requests-per-sec 100 :burst 200}}}

0.3.0 - 2025-12-21

Added

• Connection draining for graceful backend removal
  • Stop new connections while allowing existing ones to complete
  • Background watcher monitors connection counts via conntrack
  • Configurable drain timeout (default 30 seconds)
  • Callback support for drain completion notification
  • Runtime API: drain-backend!, undrain-backend!, wait-for-drain!
  • Status functions: get-drain-status, get-all-draining, draining?
  • Drain-aware weight computation integrates with health system
  • No BPF changes required (uses existing weight=0 handling)
  • Comprehensive examples in examples/connection-draining.clj

0.2.0 - 2025-12-21

Added

• TLS/SNI-based routing for layer 4 passthrough with layer 7 inspection
  • Route TLS traffic based on hostname without terminating TLS
  • XDP-level TLS ClientHello parsing with SNI extraction
  • FNV-1a 64-bit hashing for efficient BPF map lookup
  • Case-insensitive hostname matching
  • Weighted load balancing support per SNI hostname
  • Runtime SNI route management (add-sni-route!, remove-sni-route!, list-sni-routes)
  • Graceful fallback to default target when SNI not found or parsing fails
• New macros for cleaner code patterns
  • with-lb-state macro in core API for cleaner state access (replaces 25 when-let patterns)
  • when-root and root? in test utilities for root privilege checking
  • with-bpf-maps macro for automatic BPF map cleanup in tests
  • with-xdp-attached macro for automatic XDP detachment in tests
• Shared test utilities namespace (lb.test-util) for common test patterns
• Integration with bpf/with-program macro from clj-ebpf library for program lifecycle

0.1.0 - 2025-12-21

Added

• Initial release of clj-ebpf-lb
• XDP-based ingress DNAT for high-performance packet processing
• TC-based egress SNAT for return traffic handling
• Stateful connection tracking with per-CPU hash maps
• Source-based routing with LPM trie for CIDR matching
• Weighted load balancing across multiple backends (up to 8 targets per group)
• Health checking system with TCP, HTTP, and HTTPS checks
• Automatic weight redistribution when backends become unhealthy
• Gradual recovery for backends returning to healthy state
• Runtime configuration API for adding/removing proxies and routes
• Real-time statistics collection via ring buffer
• EDN-based configuration file format
• Command-line interface with multiple options
• ARM64 support with QEMU-based testing infrastructure
• GitHub Actions CI/CD pipeline
• Clojars publishing on version tags