Logging

With logging, it’s possible to record all Goose activity. This can be useful for debugging errors, for validating the load test, and for creating graphs.

When logging is enabled, a central logging thread maintains a buffer to minimize the IO overhead, and controls the writing to ensure that multiple threads don’t corrupt each other’s messages. All log messages are sent through a channel to the logging thread and written asynchronously, minimizing the impact on the load test.

Each log option takes a file name, such as --request-log requests.log, and writes that file. Given without one, as --request-log, it writes a file named for the log and its format into the run’s folder: requests.json, requests.csv, or requests.log for raw and pretty. A name ending in .gz writes the log gzipped, and --compress gzips every log in the run’s folder. See Output files.

Every log takes a format option: --request-format, --transaction-format, --scenario-format, --error-format and --debug-format. Each accepts json (the default), csv, raw or pretty. A JSON log writes one object per line, with its keys in the order of the logged structure’s fields. A CSV log starts with a header row. The elapsed field of the request, transaction, scenario and error logs counts milliseconds since the load test started, a clock every user shares. The raw format writes Rust’s debug output of each entry on one line, and pretty spreads the same output over several indented lines.

The request, transaction, scenario and error logs are built from Goose metrics, so they can’t be combined with --no-metrics. The debug log can.

What logging costs

With Goose on two cores running examples/simple with 1,000 users and no waits, the JSON request log added about 3.4 µs of CPU to each request, a tenth of what Goose spent per request. Gzip at the default level 6 roughly doubles that. Each request writes about 500 bytes, so at 25,000 requests a second the log grows by about 45 GB an hour. Write it uncompressed unless the disk can’t keep up or space is short; then gzip it at the default level.