Release Mako v0.6.35 · loreste/mako

Actor performance and scaling

Single-port loops receive up to 16 ready messages per lock acquisition without waiting to fill a batch. FIFO ordering, early returns, and typed cleanup of unprocessed messages are preserved.
Single boolean messages and pairs of signed 24-bit integers avoid envelope allocations. Larger pairs fall back to full-width envelopes without truncation.
Shared and borrowed slice payloads remain isolated snapshots. Exclusive heap backing can transfer without an additional copy, with recursive checks for nested slices.
Named-port polling avoids temporary Result allocations and preserves per-message control priority.
Added bounded, keyed sharding examples, multi-producer regression tests, and throughput/latency benchmarks. Cross-shard transactions still require application coordination.
Fixed cleanup of channel arrays and native channel fields/aliases, raw actor handles, inline message strings, and mangled port names such as exec.

Earlier actor correctness fixes included

Includes the fixes for #60–#63: typed payloads, constructor arguments, early returns, multi-file self resolution and state persistence, nested-slice double-free fixes, and safe rejected-send/shutdown cleanup. The documented self-aliasing local-variable workaround remains available.

Validation and measured performance

CI for this revision: all 16 jobs passed, including ASan, UBSan, TSan, memory-safety gates, all three platforms, and the unchanged strict 1.5× performance contract. Timing jobs required a fresh-runner retry.
Local full test suites: 470/470 native and 470/470 C with ASan/UBSan.
Local strict 1.5×-Rust gate passed all seven kernels; actor200k measured 1.35× Rust and chan50k 0.91× Rust.
Seven alternating payload benchmark runs on the local Apple arm64 host: integer pairs improved from 17.20 ms to 5.32 ms, and booleans from 16.98 ms to 5.19 ms, for 200,000 messages. Measured send-to-handler latency remained 4/12/15 microseconds at p50/p95/p99.
Local memory-safety gate passed, including 800/800 HTTP soak requests with stable RSS. Focused C/native actor and sharding leak checks reported zero leaks.

Measurements are workload- and host-specific. The mailbox remains mutex-based; an experimental direct-lock send path did not improve contended throughput and was removed.

actor_len counts queued messages and excludes the bounded in-flight batch. Raw failed sends retain ownership; prefer generated consuming send helpers, and use Name_drop_message for discarded owned messages.

Full changelog: v0.6.34...v0.6.35

Full Changelog: v0.6.34...v0.6.35

Navigation Menu

Repository navigation

Mako v0.6.35

Choose a tag to compare

Sorry, something went wrong.

Sorry, something went wrong.

Uh oh!

No results found

Actor performance and scaling

Earlier actor correctness fixes included

Validation and measured performance

Uh oh!