GitHub announced native caching for Actions that automatically detects dependencies (npm, pip, cargo, maven) and caches them without explicit configuration. Average build time improvements of 40-60% in early testing.

The implementation uses content-addressable storage and can share cache layers across workflows. Free tier gets 10GB per repo; Team tier gets 100GB. Rolling out over the next two weeks.

PostgreSQL 17 beta includes a rewritten query planner that delivers 40% faster OLAP queries and 25% faster OLTP in benchmark tests. The new incremental sort algorithm and better parallel query scheduling are the main contributors.

Also notable: improved JSONB indexing, native SQL/JSON path queries, and logical replication for large objects. The migration path from 16 is straightforward—no pg_upgrade gotchas in testing so far.

AWS extended Lambda SnapStart to Python, enabling sub-100ms cold starts for Python functions by pre-initializing and snapshotting the runtime. Works with common frameworks (FastAPI, Flask, Django) out of the box.

Loki 3.0 introduces "adaptive indexing" that automatically creates indexes for high-cardinality fields you actually query, eliminating the upfront index configuration tax. Query times improved 10-100× for common patterns.

A practical walkthrough of distributed consensus algorithms by the etcd team, covering why Paxos is hard to implement, how Raft makes it understandable, and the subtle bugs that emerge in production (split-brain, network partitions, clock skew).

Includes real outage examples from HashiCorp and CockroachDB showing how consensus bugs manifest. The section on testing consensus algorithms is particularly valuable—they demonstrate how to use Jepsen-style fault injection to catch edge cases.

ACM Queue article examining the technical debt of feature flags, based on survey data from 150 engineering teams. Shows that unmaintained flags increase code complexity by 40% and slow down feature velocity by 25%.

Charity Majors (Honeycomb) argues that comprehensive staging environments are a lie we tell ourselves. Real reliability comes from progressive rollouts, observability, and the ability to test safely in production.

The framework for "testability requirements" is useful—structured way to think about what observability and controls you need before shipping. Includes runbooks from incidents where testing in production saved them.