This Week in Ekala #6
Welcome to This Week in Ekala #6! Here's what happened this week.
Highlights
- The ekapkgs-cli cache server received a massive performance overhaul — streaming NAR serving, SQLite-backed search indexes, tiered zstd compression, and a dedicated compression worker pool
- KDE Frameworks and KDE Gear landed in ekapkgs with ~300 packages in a new
kde/directory - Default ecosystem versions bumped across the board: Linux kernel 6.18, Ruby 4.0, Node.js 26, Erlang/OTP 28
- eka-ci gained AI agent skill gates that run model evaluations against PR diffs
corepkgs
2 packages were updated in a batch sweep and best-practices cleanups were applied treewide.
- buildBunPackage: init
Introduced a dedicated builder for Bun packages, enabling first-class support for the growing Bun ecosystem in ekapkgs. - Bump default ecosystem versions
Major default version bumps: Linux kernel 6.12 → 6.18 LTS, Ruby 3.4 → 4.0, Node.js 24 → 26 LTS, and Erlang/OTP 27 → 28. Also adds new variants for LLVM 22, Perl 5.44, and PHP 8.5.
ekapkgs
86 packages were added across Rust (34), prebuilt binaries (26), Bun (19), Python (3), and database packages (4). Contributed by @ic4-y and @jonringer.
- postgresql: wire extension dir into postgresqlPackages
Wired a dedicatedpostgresql/pkgs/directory into thepostgresqlPackagesscope, establishing the scaffolding for PostgreSQL extension packages. Contributed by @ic4-y. - Mass additions 2 (KDE)
Added ~300 KDE Frameworks and KDE Gear packages in a newkde/directory, establishing foundational KDE packaging infrastructure for ekapkgs.
ekapkgs-cli
Bug fixes and observability improvements were made alongside the major changes below.
- Optimize IO
Reader lock optimizations to reduce SQLite contention and improve concurrent read throughput in the cache server. - Reduce memory use
Switched to streaming throughout NAR serving and compression paths, eliminating large in-memory allocations. - search: migrate indexes from json.zst to SQLite
Replaced in-memory zstd-compressed JSON with a singlesearch.dbSQLite database for package, option, and file search indexes. - serve: replace storage_format with nullable zstd_level column
Unified chunk compression state into a single nullable column with background recompression from zstd-1 to zstd-9. - serve: performance improvements across storage, CPU, and memory
11 targeted improvements including SQLite tuning, bloom filters, LRU caches, zero-copy reads, and parallel delta computation. - template: add --forge flag for self-hosted GitHub/GitLab instances
New--forgeoption enables template generation from self-hosted GitHub and GitLab instances, with GitLab API metadata fetching and automatic tag prefix detection. - serve: dedicated zstd-9 worker pool for chunk compression
Added a dedicated rayon thread pool for zstd-9 compression with zstd-1 overflow fallback to keep latency bounded under load.
eka-ci
- feat: AI agent skill gates for CI
Introduced a new "skills" gate type that runs AI model evaluations against PR diffs. Skills are defined with a prompt file and optional model ID; providers are configured server-side to keep API keys out of repository config. The model returns structured pass/fail verdicts mapped to GitHub check runs.
python-packages
Broad fix sweep applied across hundreds of Python package definitions, plus a new best-practices AI skill definition and GitHub Actions Nix workflow.