From 154f0f6f3d6f0ee7c81c9e56ccc198e1a75de3ed Mon Sep 17 00:00:00 2001
From: Puneeth <85823685+puneethkotha@users.noreply.github.com>
Date: Tue, 17 Mar 2026 01:27:55 -0700
Subject: [PATCH] docs: add CHANGELOG.md for v1.0.0 release

Documented notable changes for version 1.0.0, including new features, performance improvements, and infrastructure updates.
---
 CHANGELOG.md | 55 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 55 insertions(+)
 create mode 100644 CHANGELOG.md

diff --git a/CHANGELOG.md b/CHANGELOG.md
new file mode 100644
index 0000000..947d078
--- /dev/null
+++ b/CHANGELOG.md
@@ -0,0 +1,55 @@
+# Changelog
+
+All notable changes to Falcon are documented here.
+Format follows [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
+Versioning follows [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+---
+
+## [1.0.0] - 2026-02-01
+
+### Added
+- Production ML inference API: `POST /infer` for single-text sentiment classification (negative, neutral, positive)
+- Batch inference endpoint: `POST /infer/batch` supporting up to 50 texts per request
+- Nginx load balancer routing across multiple FastAPI workers (least-connections algorithm)
+- Redis response caching keyed by normalized input for sub-millisecond cache hits
+- Redis-backed idempotency via `X-Idempotency-Key` header to prevent duplicate processing
+- PostgreSQL async request logging with in-memory buffer (max 1,000 entries) for Postgres failover
+- Circuit breaker: opens on 5 consecutive failures, 60s timeout, half-open recovery probe
+- Exponential backoff retry: 3 attempts, 100ms–5s delay range
+- Graceful shutdown: SIGTERM handler drains in-flight requests and flushes log buffer
+- Prometheus metrics: 20+ metrics including latency histograms, cache hit ratio, circuit breaker state, dropped log count
+- Grafana pre-provisioned dashboard: RPS, p50/p95/p99 latency, error rate, cache hit ratio, worker health
+- Prometheus alert rules: high p95 latency, error rate spike, worker down, Redis/Postgres unhealthy
+- k6 load testing suite: baseline (50 VUs/5 min), stress (ramp to 500 VUs), spike (10→300 VUs), soak (100 VUs/10 min)
+- Failure injection scripts: `kill_worker.sh`, `redis_down.sh`, `postgres_slow.sh`, `cpu_spike.sh`
+- Docker Compose local dev stack: API workers, Nginx, Redis, PostgreSQL, Prometheus, Grafana
+- GitHub Actions CI/CD pipeline with automated deployment to GitHub Pages (demo site)
+- Full documentation suite: `RUNBOOK.md`, `CAPACITY_PLAN.md`, `SECURITY.md`, `TRADEOFFS.md`, `PERFORMANCE_NOTES.md`, `UBUNTU_DEPLOYMENT.md`
+
+### Performance
+- p95 latency reduced by 30% vs single-worker baseline under 500 VU load
+- Cache hit ratio: 70%+ on repeated inference requests
+- Graceful shutdown completes within configured `GRACEFUL_SHUTDOWN_TIMEOUT_SECONDS`
+- Circuit breaker eliminates cascading failure propagation to upstream clients
+
+### Infrastructure
+- Python 3.11 runtime
+- FastAPI 0.109 + Uvicorn 0.27
+- Nginx 1.25
+- Redis 7
+- PostgreSQL 15
+- Prometheus 2.48 + Grafana 10.2
+
+---
+
+## [Unreleased]
+
+### Planned
+- GPU inference support with CUDA-accelerated model serving
+- Multi-model registry with hot-swap capability
+- Horizontal autoscaling via Kubernetes HPA
+- OpenTelemetry distributed tracing integration
+- gRPC endpoint alongside REST for lower-latency clients
+- A/B testing framework for model version comparison
+- Token-based authentication (JWT) for API access control