v0.7.0

[0.7.0] - 2026-07-17

Changes

• V0.7.0 (#113)
• Fix worker-pool shutdown tax + Docker harness for reproducible benchmarking (#112)
• RFC process + initial baseline RFCs on the docs site (#99)
• Give Prefect real concurrency on benchmarks with parallel branches (#111)
• Add Scheduling guide surfacing the cron task-scheduler use case (#110)
• Standardize benchmark CPU/RAM fairness, fix partition-wiring regression, fix etl_duckdb serialization loss (#94)
• Install pandas/pyarrow for etl_duckdb_dataframes smoke test
• Address Indent review findings on PR #94
• Fix stale benchmark counts and README benchmark suite table
• Fix drift between docs site and implementation (reference + comparisons) (#98)
• Add Astro + Starlight documentation site (#95)
• Document the etl_duckdb serialization investigation and fix
• Add etl_duckdb_dataframes, a DataFrame/parquet variant
• Use pickle serializer for etl_duckdb's heaviest payloads
• Correct per-step timing to include artifact serialization
• Correct the noise-vs-real diagnosis, add peak memory table
• Record full 18-benchmark standardized re-run with variance
• Wire benchmark examples and partition correctness into CI
• Fix dependency wiring for partitioned steps across streams
• Opt-in whole-process-tree peak memory measurement
• Add bench.sh to the remaining script-mode benchmarks
• Fix run.sh path bugs and cache/collision bugs found rerunning suite
• Adaptive pull-queue executor with measured-cost batch sizing (#92)
• Standardize CPU pinning and worker counts across frameworks
• Allow BARCA_POOL_SIZE env override for worker pool size
• Async-native core — runtime owned by the caller, cancellable runs (#88)
• Switch ruff crates from git pin to crates.io =0.0.4 (#89)
• Update git workflow to trunk-based development model (#91)

Install

uv add barca