autocos | Peregrine

Repeatable benchmark lifecycle automation over cosbench-rs, Swift, Java COSBench, or mock backends.

autocos converts a task name such as 64KB_write_100 into workload generation, execution, progress tracking, result retention, and CSV/XLSX reporting. Its state is isolated under ~/.autocos/ (config, result, fool, and lib).

Install

cd autocos
cargo build --release --locked
install -m 0755 target/release/autocos /usr/local/bin/autocos
autocos --help

Task names and suites

A single task is <size>_<read|write>_<workers>, for example 4KB_read_128 or 16MB_write_8. fool is the built-in size × operation × concurrency suite. Override counts and runtime explicitly so the report can be reproduced.

autocos run 64KB_write_64 \
  --prepare-worker 16 \
  --object-count 400 \
  --container-count 4 \
  --runtime 25
autocos run fool --start-task 1MB_read_32
autocos run fool --rerun-task 16MB_write_8

Backends

Backend	Credentials or controller	Notes
`cosbench-rs`	S3 env/`~/.s3cfg`, or Swift env	default; selects Swift when `ST_AUTH` is present
`swift`	`ST_AUTH`, `ST_USER`, `ST_KEY`	direct TempAuth path
`java`	`--cosbench-url`	Java COSBench controller
`mock`	none	driver and reporting checks only

Swift example:

export ST_AUTH=http://127.0.0.1:8080/auth/v1.0
export ST_USER=test:tester
export ST_KEY='REPLACE_FROM_SECRET_STORE'
autocos run 64KB_read_32 --backend swift --runtime 30

The current loader does not implement ST_ENDPOINT. The backend follows the storage URL returned by TempAuth. If a benchmark contract needs a fixed route, configure that route in the underlying cosbench-rs workload or the auth/LB environment and record it in the run manifest.

S3 uses accesskey, secretkey, and endpoint (uppercase spellings are also accepted), or access_key, secret_key, and host_base from ~/.s3cfg.

Result lifecycle

Command	Behavior
`list [fool]`	list completed normal or suite runs
`remove <wid>`	remove a normal result
`remove -f <wid>`	remove one result from the suite namespace
`collect [--out PATH]`	build the normal CSV report
`run fool --collect`	collect the suite without rerunning it

Collection emits a CSV and the suite/report path can generate a formatted XLSX with throughput, bandwidth, and latency charts. Keep raw per-run results with the report; a spreadsheet alone loses provenance and error samples.

Production-grade use

Benchmark only a named route and immutable binary set. Record object size, workers, duration/op count, prepare population, cleanup outcome, client host, network path, CPU/memory limits, background traffic, rings/policies, and all error/latency distributions. A historical zero-failure run is not acceptance of the current candidate; see Performance and G0-G8 validation.