# The Skia Perf Format

The Skia Perf format is a JSON file that contains measurements. For example:

    {
      "version": 1,
      "git_hash": "cd5...663",
      "key": {
          "config": "8888",
          "arch": "x86",
      },
      "results": [
        {
          "measurements": {
            "key": {
              "test": "some_test_name"
            }
            "ms": [
              {
                "value": "min",
                "measurement": 1.2,
              },
              {
                "value": "max"
                "measurement": 2.4,
              },
              {
                "value": "median",
                "measurement": 1.5,
              }
            ]
          }
        }
      ]
    }

The format is documented
[here](https://pkg.go.dev/go.skia.org/infra/perf/go/ingest/format?tab=doc#Format).

# Storage

Each Perf data file should be stored in Google Cloud Storage in a location
of the following format:

    gs://<bucket>/<one or more dir names>/YYYY/MM/DD/HH/<zero or more dir names><some unique name>.json

Where:

    YYYY - Year
    MM - Month, e.g. 02 for February.
    DD - Day, e.g.  01, 02, etc.
    HH - Hour in 24 hour format, e.g. 00, 01, 02, ..., 22, 23

## Example

    gs://skia-perf/nano-json-v1/2018/08/23/22/Android-Clang/7989dad6c3b2efc10defb8f280f7a8a1a731d5d0.json

The Perf ingester will attempt to ingest all files below /HH/ that end in `.json`.
Nothing about the file location or the file name is ingested as data.

# Notes

- Perf only uses the data in the file, and does not parse the GCS file location to get data.
- The YYYY/MM/DD/HH should represent the time the file was written, not the
  time the data was gathered.
- Perf is robust to duplicate data, i.e. a file written at a later time can
  contain data that will replace data that has appeared in an older file.
  Where 'older' and 'newer' are defined in terms of the data/time in the GCS
  file path.
- See
  [IngestionConfig](https://pkg.go.dev/go.skia.org/infra/perf/go/config?tab=doc#IngestionConfig)
  for configuring the ingestion of new data.