Protect the throughput of large CSV and JSON processing

Your tool processes large files, and what matters is how much it gets through per second, not how long one file takes.

# yaml-language-server: $schema=../../schema/himorime.schema.json
#
# Recipe: protect the throughput of large CSV and JSON processing.
# https://nao1215.github.io/himorime/cookbook/#protect-the-throughput-of-large-csv-and-json-processing
version: "1"

name: throughput
description: Bytes and records per second of the word counter on large CSV and JSON Lines inputs.

build:
  command: [go, build, -o, "${artifact}", ../tools/wordcount]

defaults:
  warmup: 1
  runs: 10

benchmarks:
  - name: csv bytes
    setup:
      - command: ["${artifact}", -gen, "200000", -gen-format, csv, -o, "${workdir}/input.csv"]
    metrics:
      throughput:
        # Throughput is work divided by latency, and himorime never guesses the
        # work: here it is the size of the input file, read before every run.
        work:
          file_size: "${workdir}/input.csv"
          unit: bytes
    commands:
      wordcount:
        command: ["${artifact}", "${workdir}/input.csv"]
        budget:
          # Higher is better, so a throughput budget is a floor.
          throughput:
            median: ">= 5MiB/s"
  - name: jsonl records
    setup:
      - command: ["${artifact}", -gen, "200000", -gen-format, jsonl, -o, "${workdir}/input.jsonl"]
    metrics:
      throughput:
        # A fixed amount of work in your own unit: records, lines, files.
        work:
          value: 200000
          unit: records
    commands:
      wordcount:
        command: ["${artifact}", "${workdir}/input.jsonl"]
        budget:
          throughput:
            median: ">= 50000 records/s"
            min: ">= 20000 records/s"

$ himorime run examples/throughput

A throughput table in MiB/s for the CSV case and records/s for the JSON Lines case, and the budgets table. Throughput is computed per run, as the declared work divided by that run’s latency.

Declare value or file_size; himorime does not infer work. See Throughput for work units and statistic meanings.
Throughput budgets are floors (>= or >), and their unit must match the work: MiB/s uses unit: bytes, while records/s uses unit: records.

Example: examples/throughput