Feature Request: Separate output files for each format when multiple --output-format are specified

[pugsswangxs]

Full Issue Content

Summary

Currently, fio supports specifying multiple report formats via comma-separated values for --output-format (e.g. normal,terse,json). However, all formatted output contents are written into a single file defined by the single --output parameter, which mixes human-readable plain text, CSV-style terse data, and structured JSON together in one file. This mixed file cannot be directly parsed by automation scripts, spreadsheet tools or log analyzers.

I request adding support to assign an independent output file for each selected format, so each format’s report is saved to its dedicated file without mixing contents.

Reproduction Command Example

# Current behavior: all formats write into result.log (mixed messy content)
fio --name=randwrite \
--filename=/dev/sdd1 \
--direct=1 --rw=randwrite --bs=4k \
--offset=24005054464 --size=800169984 \
--output-format=normal,terse,json \
--output=result.log

After execution, result.log contains interleaved normal text, semicolon-delimited terse rows and JSON blocks, making post-processing extremely cumbersome.

Desired New Behavior (Two feasible design proposals for maintainers to pick)

Proposal 1: Format-specific output parameters (clean & unambiguous)

Add dedicated flags like --output-normal, --output-terse, --output-json, matching each supported format.
Example usage:

fio --name=randwrite \
--filename=/dev/sdd1 --direct=1 --rw=randwrite --bs=4k \
--offset=24005054464 --size=800169984 \
--output-format=normal,terse,json \
--output-normal=report.txt \
--output-terse=report.csv \
--output-json=report.json

Each format writes exclusively to its own target file. If a format is enabled but its corresponding flag is omitted, fall back to stdout or the generic --output file as a compatibility fallback.

Proposal 2: Suffix auto-append with single base filename

Accept one base name via --output, and fio automatically appends format suffixes to generate separate files:

fio ... --output-format=normal,terse,json --output=test
# Generate test.normal, test.terse, test.json

This keeps command syntax shorter while splitting outputs automatically. Either design is acceptable to end users.

Use Case & Value

1. Automated disk stability test pipelines (like my SSD firmware stability script) need three types of logs simultaneously:
   • Normal human log for manual review
   • Terse CSV for batch statistical analysis
   • JSON for programmatic metric extraction
2. Mixed single-file output forces extra post-processing scripts to split contents, adding error-prone parsing logic.
3. Many enterprise storage testers run long cyclic fio workloads; separate format files simplify log archiving, filtering and failure triage.
4. This matches fio’s existing bandwidth-log design pattern (auto split read/write log files), consistent with existing UX.

Workaround I Currently Use

I have to run 3 separate identical fio jobs with different single --output-format and distinct --output files, which triples test runtime and wastes SSD write endurance, severely reducing test efficiency.

Environment Info

• fio version: 3.19
• OS: Linux (CentOS / Kylin / Ubuntu)
• Scenario: SATA/NVMe SSD long-term stability & firmware upgrade verification tests

Additional Note

I searched existing GitHub issues and found #977 raised the same demand in 2020 without resolution. This feature would bring significant usability improvement for all automation-based fio users.

---

Short Version (If you want a concise comment/mail to mailing list)

Short Feature Request Text

Dear fio maintainers,

I’m writing to request a feature enhancement for multi-format output handling. Right now when multiple formats are set with --output-format=normal,terse,json, all output data is dumped into one single file from --output, mixing text, CSV and JSON content together, which breaks automated parsing workflows.

I hope fio can support separate dedicated output files per selected format. Two implementation ideas:

1. Introduce format-specific output flags (--output-json, --output-terse, etc.)
2. Auto append format suffix to the base filename given by --output

Running duplicate fio instances to generate separate logs doubles IO load and wastes SSD endurance, which is a big pain for long-duration stability testing. This feature will align with the existing split bandwidth log design and greatly improve usability for storage automation test scripts.

Thanks for your consideration.

[axboe]

Three suggestions:

1. Just output to json and have a post-processing tool turn it into eg csv?
2. Write the support for the above yourself
3. Have your LLM do it, it wrote the report/suggestion in the first place clearly

I think 1 makes the most sense, if you go for 2 or 3, then ensure to follow the rules of the repo.