Skip to content

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.3.0]

Added

  • Multi-step scenarios — define named sequences of HTTP steps (login → browse → checkout) that VUs execute in order. Each step has its own host, method, headers, and request/response templates.
  • Weighted VU distribution — assign relative weights to scenarios to control what proportion of VUs run each flow. Assignment is deterministic by VU index.
  • Step failure handling — configurable on_step_failure per scenario: continue (default) completes all steps, abort_iteration skips remaining steps and starts the next iteration.
  • Three-layer header merging — headers cascade global → scenario → step with case-insensitive last-wins semantics.
  • Per-scenario and per-step metrics — CLI output and JSON reports include latency, throughput, error rate, and status code breakdowns at both scenario and step granularity.
  • scenarios field in JSON output — new top-level array in the report schema with nested step data.
  • ScenarioVu execution engine — new VU type that loops through steps sequentially, with budget claiming per iteration (not per request) in fixed mode.
  • ScenarioResolver — structured config resolution with ${ENV_VAR} expansion, method parsing, and template loading per step.
  • Scenario config validation — scenarios are mutually exclusive with run.host/run.method; unique names enforced; weight in [1, 10_000]; scenarios count capped at 64.
  • Step chaining with response captures — new capture map on steps extracts values from response bodies via JSON paths ($.data.access_token) into a per-iteration, per-VU CaptureState. Captured values are injected into subsequent step headers, inline bodies, and template output via {{capture.KEY}} placeholders. Captures are string-valued (objects/arrays stringified as compact JSON).
  • Inline body field on steps — mutually exclusive with request_template. Supports {{capture.KEY}} injection and is capped at 1 MiB.
  • Startup static validation of capture references — during config resolution, every {{capture.KEY}} reference in step headers and bodies is checked against the cumulative set of aliases defined by preceding steps. Undefined references fail the run before any load is generated.
  • Skipped-step accounting — new skipped: bool field on RequestRecord and total_skipped: u64 on RequestStats. Steps skipped due to unresolvable captures or abort_iteration emit skipped records that contribute to request counts but not to latency histograms or status-code breakdowns. Surfaced in both CLI output (0 skip) and JSON output (requests.skipped).
  • Dependency-aware iteration abort — if a step references {{capture.KEY}} that isn't in the capture state (prior step failed or server omitted the field), the iteration aborts immediately regardless of on_step_failure.
  • Scenarios guide, recipe, and config reference documentation (including the capture feature and size caps).
  • 5 new functional tests covering scenarios in fixed, curve, abort, JSON output, and per-step stats modes.
  • --rps flag and execution.rps config field — cap aggregate requests-per-second across all VUs. Implemented as a shared token-bucket limiter (via governor), so output is smoothed rather than bursted at the boundary of each second. Works in both fixed and curve modes; in scenario mode the cap applies per HTTP request, not per iteration. Omit for no rate limit (full-throttle behaviour unchanged).
  • Functional test coverage for --rps (CLI flag with timed-elapsed assertion + YAML config form).

Changed

  • RequestSpec is now an enum — Single { ... } for single-endpoint mode, Scenarios(Vec<ResolvedScenario>) for multi-step mode. This is a breaking change for programmatic users of lmn-core.
  • DrainMetricsAccumulator — extracted shared drain logic from both fixed and curve executors, eliminating code duplication.
  • HashMap keys use Arc<str> — scenario/step accumulator maps avoid per-request String heap allocation on the hot path.
  • Single canonical sort — scenarios and steps are sorted once in into_stats() rather than redundantly in three layers.
  • JSON output schema version bumped to 2.

[0.2.0]

Breaking

  • JSON output: sampling field removed from the output schema. Latency percentiles are now exact (computed from a full HDR histogram) — no approximation, no sampling state to report.
  • Config: result_buffer and sample_threshold config fields removed. The old reservoir sampling pipeline has been replaced; these settings have no effect and will cause a parse error if present.

Changed

  • Histogram-based statistics pipeline — exact latency percentiles for any run size, replacing the reservoir sampling approach
  • Worker-pool VU model with Arc-based clone reduction — lower memory overhead at high concurrency
  • CompiledTemplate struct — template compilation errors are now reported at startup, before any requests fire
  • Write-to-buffer renderer with pre-resolved globals — faster per-request body generation
  • Replaced expect/unwrap with a typed RunError — no more panics during a run, errors surface cleanly

Fixed

  • CLI reference: default value for --concurrency corrected to 10 (was incorrectly documented as 100)

[0.1.7]

Changed

  • Restore publish pipeline
  • Drop Windows from release targets
  • Bump CI dependencies: actions/setup-python, docker/login-action, docker/build-push-action

[0.1.5]

Changed

  • Improve documentation structure and content
  • Set custom domain lmn.talek.cloud for docs site
  • Fix cargo-deb release pipeline

[0.1.4]

Changed

  • Update docs URLs to lmn.talek.cloud

[0.1.3]

Added

  • Custom request headers via --header CLI flag and headers: config section
  • ${ENV_VAR} interpolation in header values for secret management
  • .env file loading at startup via dotenvy
  • SensitiveString type — secrets are redacted in debug output
  • {{ENV:VAR_NAME}} built-in template placeholder — resolved from environment at parse time
  • Load curve support — time-based VU scaling with stages: config
  • Threshold-gated CI — exit code 2 on threshold failure
  • Response template extraction — track specific fields from response bodies
  • OpenTelemetry tracing support via OTEL_EXPORTER_OTLP_ENDPOINT
  • JSON output mode (--output json) and file artifact (--output-file)
  • GitHub Actions workflows — CI, release (5-target cross-compiled binaries), publish, audit
  • Apache 2.0 license