87d1b5356187fb4a35aa11204223ed2c6586268d,perfkitbenchmarker/linux_packages/fio.py,,ParseResults,#Any#Any#Any#Any#Any#,141

Before Change

  timestamp = time.time()
  parameter_metadata = ParseJobFile(job_file)
  io_modes = DATA_DIRECTION.values()
  for (idx, job) in enumerate(fio_json_result["jobs"]):
    job_name = job["jobname"]
    parameters = parameter_metadata[job_name]
    parameters["fio_job"] = job_name
    if base_metadata:
      parameters.update(base_metadata)
    for mode in io_modes:
      if job[mode]["io_bytes"]:
        metric_name = "%s:%s" % (job_name, mode)
        bw_metadata = {
            "bw_min": job[mode]["bw_min"],
            "bw_max": job[mode]["bw_max"],
            "bw_dev": job[mode]["bw_dev"],
            "bw_agg": job[mode]["bw_agg"],
            "bw_mean": job[mode]["bw_mean"]}
        bw_metadata.update(parameters)
        samples.append(
            sample.Sample("%s:bandwidth" % metric_name,
                          job[mode]["bw"],
                          "KB/s", bw_metadata))

        // There is one sample whose metric is "<metric_name>:latency"
        // with all of the latency statistics in its metadata, and then
        // a bunch of samples whose metrics are
        // "<metric_name>:latency:min" through
        // "<metric_name>:latency:p99.99" that hold the individual
        // latency numbers as values. This is for historical reasons.
        clat_section = job[mode]["clat"]
        percentiles = clat_section["percentile"]
        lat_statistics = [
            ("min", clat_section["min"]),
            ("max", clat_section["max"]),
            ("mean", clat_section["mean"]),
            ("stddev", clat_section["stddev"]),
            ("p1", percentiles["1.000000"]),
            ("p5", percentiles["5.000000"]),
            ("p10", percentiles["10.000000"]),
            ("p20", percentiles["20.000000"]),
            ("p30", percentiles["30.000000"]),
            ("p40", percentiles["40.000000"]),
            ("p50", percentiles["50.000000"]),
            ("p60", percentiles["60.000000"]),
            ("p70", percentiles["70.000000"]),
            ("p80", percentiles["80.000000"]),
            ("p90", percentiles["90.000000"]),
            ("p95", percentiles["95.000000"]),
            ("p99", percentiles["99.000000"]),
            ("p99.5", percentiles["99.500000"]),
            ("p99.9", percentiles["99.900000"]),
            ("p99.95", percentiles["99.950000"]),
            ("p99.99", percentiles["99.990000"])]

        lat_metadata = parameters.copy()
        for name, val in lat_statistics:
          lat_metadata[name] = val
        samples.append(
            sample.Sample("%s:latency" % metric_name,
                          job[mode]["clat"]["mean"],
                          "usec", lat_metadata, timestamp))

        for stat_name, stat_val in lat_statistics:
          samples.append(
              sample.Sample("%s:latency:%s" % (metric_name, stat_name),
                            stat_val, "usec", parameters, timestamp))

        samples.append(
            sample.Sample("%s:iops" % metric_name,
                          job[mode]["iops"], "", parameters, timestamp))
    if log_file_base and bin_vals:
      // Parse histograms
      hist_file_path = vm_util.PrependTempDir(
          "%s_clat_hist.%s.log" % (log_file_base, str(idx + 1)))
      samples += _ParseHistogram(
          hist_file_path, bin_vals[idx], job_name, parameters)
  return samples

After Change

    if log_file_base and bin_vals:
      // Parse histograms
      aggregates = collections.defaultdict(collections.Counter)
      for _ in xrange(int(job["job options"]["numjobs"])):
        clat_hist_idx += 1
        hist_file_path = vm_util.PrependTempDir(
            "%s_clat_hist.%s.log" % (log_file_base, str(clat_hist_idx)))
        hists = _ParseHistogram(hist_file_path, bin_vals[clat_hist_idx - 1])

        for key in hists:
          aggregates[key].update(hists[key])
      samples += _BuildHistogramSamples(aggregates, job_name, parameters)

  return samples

Instances