Releases

0.2.1

This release improves shutdown reliability for database clients and TCP-based protocols, and adds application-driven IO execution.

Features

• Database shutdown waits: MySQL connections, Redis clients and both pools provide close_and_wait(). Permanent closure rejects new requests and waits for accepted operations, callbacks and owned IO resources to retire. Retained leases and statement handles become inert.
• Application-driven execution: manual_runtime lets an application drive IO with poll() from its update loop, without creating a worker that executes business handlers. Its standard Asio executor supports awaitable and co_spawn; dedicated public coroutine protocol APIs remain TODO.

Fixes

• TCP, TLS TCP, HTTP and WebSocket shutdown waits cover outstanding IO and callbacks, preventing endpoint resources from being released while still in use. Shutdown also prevents long timer rearming; explicit restart and normal reconnect remain available.
• Database pools complete queued connection requests during shutdown without relying on container allocations, including reentrant close and cancellation. Redis pools preserve diagnostics and recover connection capacity after lease-return submission failures, without replaying commands or changing successful results.

Upgrade notes

Stop application producers before waiting for shutdown, and keep external contexts running until completion. Network shutdown does not drain business tasks or establish dynamic-plugin unload safety. See lifecycle and database shutdown.

The five-second workload comparison records 6.51% lower median throughput and 13.29% higher P99 for the 4 KiB WebSocket case. Five-run throughput ranges overlap; the cause is unconfirmed. Measurements use snapshot 2aa5bc0, before the release’s final pool-shutdown correction. Detailed validation, performance reports and release attachments retain their original source identities; deployment capacity has not been established.

0.2.0

This release adds native single-node MySQL and Redis clients, makes Boost the default networking dependency, and improves process-exit cleanup. arknet remains C++20 and header-only.

Features

• MySQL: text queries, typed prepared parameters and results, explicit transactions, caching_sha2_password and mysql_native_password, verified TLS, and bounded connection pools with exclusive leases.
• Redis: binary-safe RESP2/RESP3 commands, ordered pipelines, authentication, TLS/mTLS, separate RESP3 push delivery, and bounded connection pools with exclusive leases.
• Request control: queue and result limits, deadlines, cancellation, session reset, and explicit indication when a failed command may already have executed. Commands are not automatically replayed.

Both clients support Boost.Asio and Standalone Asio. The database client guide provides complete programs, configuration and pool reuse rules. MySQL pools restore the configured database after reset, or discard returned connections when no database is configured; prepared statements remain session-bound. Redis pools restore protocol, authentication and database settings before reuse. Leases remain exclusive until pending work and return processing finish.

Fixes

• Owned-runtime process cleanup waits for callbacks, worker threads and TLS resources before OpenSSL’s normal cleanup, including supported callback and global-owner lifetimes. The cleanup worker starts with the first owned IO pool; including a header alone does not start one.
• Database requests reject invalid or unrepresentable deadlines before clock arithmetic and restore queue capacity if submission throws. Documented zero per-call overrides still select the configured default.
• MySQL retains its runtime through asynchronous submission and releases command timers before the runtime, including when a completion destroys the connection.

Upgrade notes

ARKNET_USE_BOOST_ASIO now defaults to ON; set it to OFF to retain Standalone Asio. HTTP/HTTPS and WebSocket/WSS require Boost. The bundled BHO tree has been removed: replace arknet/bho/ includes and bundled BHO types with official Boost APIs. The arknet::arknet interface target propagates headers, settings and dependencies without an arknet binary library. Review installation when upgrading.

MySQL authentication requires OpenSSL even with TLS disabled, and its connection API requires ARKNET_ENABLE_SSL. Verified TLS is the default; there is no automatic downgrade. Plain Redis does not require OpenSSL; secure Redis does.

Each translation unit declaring a global IO owner must include the relevant arknet IO header before that declaration. Owners without that include, including forward-declared or type-erased globals, need explicit teardown before main returns. Legacy Asio thread-storage configurations retain their own requirements. See process shutdown.

MySQL supports one Classic Protocol result per command; multi-result execution, LOCAL INFILE, compression, cursors and arbitrary authentication plugins are outside this release. Redis Cluster, Sentinel and Pub/Sub, dedicated database coroutine APIs and other planned protocols remain on the roadmap.

See testing for independent MySQL and Redis server checks and sanitizer coverage, and performance for local measurements. Application and ark plugin regression, cross-host load and deployment endurance remain deployment acceptance; this release does not establish production PCU capacity.

0.1.0

The first regular release provides a C++20 header-only networking library with client/server APIs, CRTP extension points and component composition.

Features

• TCP, native UDP, RUDP, WebSocket/WSS, HTTP/HTTPS with routing, optional TLS/mTLS, and IPv4/IPv6.
• RUDP with ARQ-only, fixed Reed–Solomon FEC and adaptive FEC policies. Local checks cover negotiation, ordered delivery, recovery, resource limits, reconnect and legacy zephyr interoperability.
• Automatic reconnect, timers and post, bounded sends, owned or external IO contexts, and deadlines and cancellation for single HTTP requests.

Fixes

Lifecycle status queries use atomic state to avoid races with socket and acceptor shutdown, including concurrent queries of retained sessions.

Upgrade notes

This version defaults to Standalone Asio and also supports Boost.Asio. The default changes to Boost in 0.2.0.

Testing and RUDP performance identify verified workloads and source revisions. Cross-host load, production-duration endurance and ark business regression remain deployment acceptance. Cluster capacity of 100,000–200,000 PCU has not been measured.

GitHub releases