feat: Implement graceful shutdown with connection tracking

[dcoppa]

Description

This PR implements graceful shutdown support for chproxy, allowing the proxy to properly handle SIGTERM and SIGINT signals and wait for active connections to complete before terminating.

Motivation: When chproxy is deployed in orchestrated environments like Kubernetes, abrupt termination can kill in-flight queries, leading to failed client requests and incomplete data operations. This implementation ensures clean shutdowns by waiting for active connections to complete (or timeout) before process termination.

Key changes:

• Signal handling using signal.NotifyContext for SIGTERM/SIGINT
• Connection tracking via http.Server.ConnState callback with atomic counter
• Configurable graceful_shutdown_timeout (default: 25s, fits within K8s 30s grace period)
• Concurrent shutdown of HTTP and HTTPS servers
• Resource cleanup (caches, heartbeat goroutines, idle connections)
• Debug logging for connection lifecycle
• Test coverage for shutdown scenarios

Example shutdown flow:

INFO: Shutdown signal received
INFO: Starting graceful shutdown with 3 open connections
INFO: Shutting down HTTP server...
DEBUG: Connection closed from 127.0.0.1:41826 (active: 2)
DEBUG: Connection closed from 127.0.0.1:41832 (active: 1)
DEBUG: Connection closed from 127.0.0.1:41836 (active: 0)
INFO: HTTP server stopped
INFO: Closing proxy resources...
INFO: Graceful shutdown completed successfully (all connections closed)

Pull request type

Please check the type of change your PR introduces:

• Bugfix
• Feature
• Code style update (formatting, renaming)
• Refactoring (no functional changes, no api changes)
• Build related changes
• Documentation content changes
• Other (please describe):

Checklist

• Linter passes correctly
• Add tests which fail without the change (if possible)
• All tests passing
• Extended the README / documentation, if necessary

Does this introduce a breaking change?

• Yes
• No

Further comments

Implementation Highlights:

1. Connection Tracking: Uses http.Server.ConnState callback. This correctly tracks TCP connections and handles HTTP keep-alive.

2. Concurrent Server Shutdown: HTTP and HTTPS servers shut down in parallel using sync.WaitGroup for efficiency.

3. Thread Safety:

   • Atomic operations for connection counter
   • Consistent lock ordering (configLock -> lock) to prevent deadlocks
   • Safe channel close pattern for reloadSignal

4. Error Handling: Errors collected via buffered channel and aggregated with errors.Join

5. Testing Strategy: Uses subprocess pattern to test shutdown behavior in isolation

Files Changed:

• main.go: Signal handling, connection tracking, graceful shutdown logic
• proxy.go: Resource cleanup (caches, heartbeat goroutines, transport)
• config/config.go: Added GracefulShutdownTimeout field
• main_test.go: Two comprehensive test cases (normal shutdown + timeout)
• Documentation: English, Chinese, and configuration reference

Testing:

# Run tests
go test -v -run TestGracefulShutdown

# Manual test
curl -u default:password http://localhost:9090 -d "SELECT sleep(10)" &
pkill -TERM chproxy  # Observe it waits for query completion