OpenAI released their real-time voice API today, enabling bidirectional voice streaming with sub-second latency. The WebSocket-based API handles interruptions naturally and supports function calling mid-conversation.

Early access users report GPT-4 quality responses with natural interruption handling. The API also supports streaming transcripts, making it useful for call center automation and voice-first interfaces.

The Rust team shipped version 2.0 with two major ergonomic improvements: async fn in traits (no more Box<dyn Future> gymnastics) and a stable effect system for compile-time capabilities.

Migration guide shows most code compiles without changes. The effect system is particularly interesting for embedded and systems programming where you need compile-time guarantees about I/O or memory allocation.

Meta released Canopy, the distributed tracing system they use to debug production issues across 100,000+ services. It handles 50 billion spans per day with automatic trace sampling and intelligent retention.

The ClickHouse-based storage design is particularly clever—they use materialized views for common queries and get sub-second P99 query times on a month of production traces. Compatible with OpenTelemetry.

GitHub added built-in DORA metrics (deployment frequency, lead time, MTTR, change failure rate) to Actions. No configuration needed—it automatically detects production deployments and calculates trends.

Brandur Leach's detailed exploration of database connection pooling, covering the math behind pool sizing, common misconfiguration patterns, and why the "connections = 2 × cores + 1" rule is outdated.

The section on pgBouncer vs. application-level pooling is especially good. He shows real metrics from Crunchy Data demonstrating that pooler overhead is negligible compared to connection establishment cost.

University of Washington researchers analyzed 6 months of Lambda telemetry across 50 production applications. They found cold starts consume 23-47% of total execution time in typical workloads—far higher than AWS publishes.

Most valuable section: how to instrument your functions to measure cold start percentage in production. They provide CloudWatch Insights queries that actually work.

Martin Fowler's updated take on modular monoliths, featuring case studies from Shopify, GitHub, and Stack Overflow on how they structure large monolithic codebases with clear module boundaries.