Releases
v0.5.0
Compare
Sorry, something went wrong.
No results found
Added
Observability instrumentation: tracing spans on RPC handlers, job workers, and cron ticks with structured fields; Prometheus-style metrics for request count, latency, and queue depth; slow query logging with configurable threshold; startup summary banner
Consistent query routing via #[forge::query(consistent)] attribute to force reads from primary, bypassing replicas for read-after-write consistency
Health-aware replica selection: background monitor pings replicas every 15s, automatically skips unhealthy replicas and falls back to primary
Workload-isolated connection pools (pools.default, pools.jobs, pools.observability, pools.analytics) with independent size and timeout configuration
Coalesced real-time subscriptions: identical query subscriptions share a single re-execution instead of running per-client
Hybrid rate limiting combining in-memory token bucket with PostgreSQL-backed sliding window for cluster-wide consistency
Cluster-aware cache invalidation via forge_invalidations table so nodes only re-execute queries affected by changes on other nodes
forge-idiomatic-engineer Claude Code skill shipped with scaffolded projects for AI-assisted development
Changed
Removed 20 direct dependencies by reimplementing minimal usages inline (async-stream, hex, regex, regex-lite, walkdir, darling, dialoguer, indicatif, hostname, arc-swap, sysinfo, slab, smallvec, once_cell, futures, axum-extra, tonic, prost, hyper, and gRPC features from opentelemetry-otlp)
Switched OTLP telemetry transport from gRPC (port 4317) to HTTP (port 4318), eliminating duplicate transitive dependency trees (axum 0.7, tower 0.4, matchit 0.7)
Dropped bundled AGENTS.md from project templates in favor of the installed skill
You can’t perform that action at this time.