메뉴
Work on Rally's own codebase, not running benchmarks with it. Use when setting up the dev environment, running Rally's tests or linters, navigating its source, debugging Rally's own code, or making changes to Rally itself.
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/elastic/rally --skill developing-rally명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
SOC 직업 분류 기준
SKILL.md 표시 중
| name | developing-rally |
| description | Work on Rally's own codebase, not running benchmarks with it. Use when setting up the dev environment, running Rally's tests or linters, navigating its source, debugging Rally's own code, or making changes to Rally itself. |
For working on Rally, not just running it. The docs under References are the source of truth — prefer them over guessing.
Development uses uv and needs jq. Rally runs on Linux/macOS
only (no Windows) and targets Python 3.10–3.13. From a clone:
make install # create .venv and editable-install with dev extras
source .venv/bin/activate
esrally --help # runs your working-tree code (editable install)
make install editable-installs the workspace, so esrally / esrallyd (or uv run esrally …)
run your working-tree code directly. Avoid the ./rally / ./rallyd wrappers while developing:
they auto-update from git (via run.sh) and only run on a clean master, otherwise erroring
There are uncommitted changes … — pass --skip-update if you must use them.
make lint / make format — run/apply pre-commit linters (black, isort, …).make test — unit tests. Scope with make test ARGS=tests/foo_test.py; make test-all
runs across Python 3.10–3.13.make it — integration tests (slow, CI-oriented). make benchmark runs the micro-benchmarks.make docs / make serve-docs — build/serve the docs/ site.Before committing run make lint test (unit tests + lint; integration tests are too slow for
the inner loop). Optionally make install-pre-commit to run lint automatically on each commit.
tests/ mirroring esrally/, named *_test.py; integration tests in
it/, micro-benchmarks in benchmarks/.asyncio_mode = strict, so async tests need an explicit @pytest.mark.asyncio.xfail_strict = true, so an xfail test that unexpectedly passes fails the suite.[tool.mypy].overrides in pyproject.toml. Match the
surrounding code rather than typing everything.driver/ runners, schedulers, and parameter
sources runs per operation during measurement, so extra overhead (logging, allocations,
blocking I/O, locks) perturbs the throughput/latency/service_time numbers Rally exists to
report. Do expensive work outside the measured loop.pytest-benchmark micro-benchmark under
benchmarks/ — coverage there is sparse, so treat this as a tool to reach for, not a standing
requirement. Take the benchmark fixture, call benchmark(fn), decorate with
@pytest.mark.benchmark(group="...", warmup="on", disable_gc=True), and compare against a
baseline variant. See benchmarks/driver/runner_test.py.esrally/)rally.py — CLI entry point: argument parsing and subcommand dispatch.racecontrol.py — orchestrates a race end-to-end via the pipelines registry.mechanic/ — provisions, launches, and tears down Elasticsearch (only for from-* pipelines).driver/ — generates load: schedules tasks and runs runner.py operations against the cluster.track/ — loads and parses tracks/challenges; tracker/ generates a track from an existing cluster.metrics.py — the metrics store (in-memory or Elasticsearch); reporter.py — the summary report.client/ — Elasticsearch client factory; telemetry.py — telemetry devices; config.py / types.py — config.Rally is a distributed system built on Thespian; even a
single-machine race runs as actors. Three logical roles (docs/rally_daemon.rst) that, by
default, all run on the one machine where you invoke esrally:
--load-driver-hosts and --target-hosts split the load driver and provisioner onto other
machines; multi-machine runs use the esrallyd daemon. esrally/actor.py holds the shared
RallyActor base and the bootstrap logic; docs/architecture/actor_system.md walks through the actors
and their message flow with sequence diagrams. If actor-system startup fails (common on a VPN),
set THESPIAN_BASE_IPADDR to a routable address.
Rally logs to ~/.rally/logs/rally.log (human-readable) and rally.json (structured);
profiling output goes to profile.log. Console output is only a summary — the full stack
trace of a failure is usually in the log file, not on stdout.
Each log line carries the emitting actor's address (%(actorAddress)s), so you can trace a
failure across the actor system. The actor framework's own internal log is separate:
~/.rally/logs/actor-system-internal.log (path via THESPLOG_FILE) — check it when actors
fail to start or communicate, and set THESPLOG_THRESHOLD (default INFO) to DEBUG for
more detail on the actor system itself.
Raise the log level to see more: edit ~/.rally/logging.json (created on first run from
esrally/resources/logging.json) and set "level": "DEBUG" on the root logger for
everything, or add a per-module entry under loggers to scope it (matching the existing
entries, which just set a level and inherit the root handlers), e.g.:
"loggers": {
"esrally.driver": {"level": "DEBUG"}
}
Then re-run. Revert to INFO when done — DEBUG is verbose.
Integration tests in it/ drive real races, so a failure usually reflects a real bug.
To debug any IT failure:
Re-run the single failing test in isolation: make it ARGS="it/<file>.py -k <pattern>".
Reproduce outside the test harness by running the equivalent race yourself — fast, against a
throwaway local cluster — then read the logs (see Debugging, raising the level
to DEBUG as needed):
esrally race --distribution-version=9.4.0 --track=geonames --test-mode --on-error=abort
If the failure is in the actor system, add THESPLOG_THRESHOLD=DEBUG and inspect a running
system with the Thespian shell: python3 -m thespian.shell (see docs/rally_daemon.rst).
docs/developing.rstCONTRIBUTING.mddocs/rally_daemon.rstdocs/architecture/actor_system.mdesrally <subcommand> --help): docs/command_line_reference.rstrunning-benchmarks skill