v2.0.1

A maintenance release: one fix for following live streams, plus faster parsing and aggregation across the board. Output is unchanged from 2.0.0 — no new options and nothing to migrate.

Fixed

• --metrics, --drain, and --discover now print their results when you stop a live stream with Ctrl-C - These modes normally show their summary once the input ends. When following an endless stream — for example tail -f app.log | kelora -m … — that end only comes when you press Ctrl-C, and until now doing so threw away everything collected so far, printing nothing. Pressing Ctrl-C once now prints the metrics table, drain templates, or discover table you'd built up. (Press Ctrl-C a second time to quit immediately; --silent still prints nothing.)

Performance

Kelora is faster on this release, with the biggest gains on large files. How much depends on what you're doing — filtering by level is the standout, and everyday parsing and field selection are meaningfully quicker too. The improvements are internal only; results are identical to 2.0.0.

• Filtering by level (--levels) is dramatically faster when few lines match - kelora now does a quick scan of each raw line and skips parsing lines that can't possibly match the levels you asked for. On a file where only ~1% of lines are the level you want, filtering runs around 8× faster (e.g. 4.8s → 0.6s over a large log); the win shrinks as more lines match, and files where most lines match are unchanged.
• Faster field selection with -k - When you keep only a few fields (-k time,level,msg) from records with many fields, kelora skips building the fields you didn't ask for — up to ~30% faster on wide JSON records, ~20% on wide logfmt.
• Faster logfmt and JSON parsing - The logfmt parser was rewritten to do far less work per line, and JSON string handling was streamlined — roughly 10–13% faster on logfmt-heavy workloads.
• Faster reading from a pipe under --parallel - Lines from standard input are now handed to the worker threads in batches instead of one at a time, so the reader is no longer the bottleneck when a filter is doing most of the discarding.
• Faster --stats and --drain grouping - The grouping behind these aggregations does less work per line, about 7% faster on large inputs.

What's Changed

• ci: target OpenBSD 7.9/7.8 in release builds by @dloss in #323
• Flush metrics/drain/discover aggregations on signal termination by @dloss in #324
• bench: add level pre-filter benchmark harness and corpus generator by @dloss in #325
• bench: add allocation-counting harness for grouping work by @dloss in #326
• perf(parallel): batch line delivery in the plain reader by @dloss in #327
• bench: record field-name small-strings baseline by @dloss in #328
• perf(parsers): projection pushdown — materialize only observed fields by @dloss in #329

Full Changelog: v2.0.0...v2.0.1