Switch to Python3

[fio.git] / tools / fio_jsonplus_clat2csv

#!/usr/bin/env python3
# Note: this script is python2 and python3 compatible.

"""
fio_jsonplus_clat2csv

This script converts fio's json+ latency data to CSV format.

For example:

Run the following fio jobs:
$ fio --output=fio-jsonplus.output --output-format=json+ --ioengine=null \
    --time_based --runtime=3s --size=1G --slat_percentiles=1 \
    --clat_percentiles=1 --lat_percentiles=1 \
    --name=test1 --rw=randrw \
    --name=test2 --rw=read \
    --name=test3 --rw=write

Then run:
$ fio_jsonplus_clat2csv fio-jsonplus.output fio-jsonplus.csv

You will end up with the following 3 files:

-rw-r--r-- 1 root root 77547 Mar 24 15:17 fio-jsonplus_job0.csv
-rw-r--r-- 1 root root 65413 Mar 24 15:17 fio-jsonplus_job1.csv
-rw-r--r-- 1 root root 63291 Mar 24 15:17 fio-jsonplus_job2.csv

fio-jsonplus_job0.csv will look something like:

nsec, read_slat_ns_count, read_slat_ns_cumulative, read_slat_ns_percentile, read_clat_ns_count, read_clat_ns_cumulative, read_clat_ns_percentile, read_lat_ns_count, read_lat_ns_cumulative, read_lat_ns_percentile, write_slat_ns_count, write_slat_ns_cumulative, write_slat_ns_percentile, write_clat_ns_count, write_clat_ns_cumulative, write_clat_ns_percentile, write_lat_ns_count, write_lat_ns_cumulative, write_lat_ns_percentile, trim_slat_ns_count, trim_slat_ns_cumulative, trim_slat_ns_percentile, trim_clat_ns_count, trim_clat_ns_cumulative, trim_clat_ns_percentile, trim_lat_ns_count, trim_lat_ns_cumulative, trim_lat_ns_percentile,
12, , , , 3, 3, 6.11006798673e-07, , , , , , , 2, 2, 4.07580840603e-07, , , , , , , , , , , , ,
13, , , , 1364, 1367, 0.000278415431262, , , , , , , 1776, 1778, 0.000362339367296, , , , , , , , , , , , ,
14, , , , 181872, 183239, 0.037320091594, , , , , , , 207436, 209214, 0.0426358089929, , , , , , , , , , , , ,
15, , , , 1574811, 1758050, 0.358060167469, , , , , , , 1661435, 1870649, 0.381220345946, , , , , , , , , , , , ,
16, , , , 2198478, 3956528, 0.805821835713, , , , , , , 2154571, 4025220, 0.820301275606, , , , , , , , , , , , ,
17, , , , 724335, 4680863, 0.953346372218, , , , , , , 645351, 4670571, 0.951817627138, , , , , , , , , , , , ,
18, , , , 71837, 4752700, 0.96797733735, , , , , , , 61084, 4731655, 0.964265961171, , , , , , , , , , , , ,
19, , , , 15915, 4768615, 0.971218728417, , , , , , , 18419, 4750074, 0.968019576923, , , , , , , , , , , , ,
20, , , , 12651, 4781266, 0.973795344087, , , , , , , 14176, 4764250, 0.970908509921, , , , , , , , , , , , ,
...
168960, , , , , , , , , , , , , 1, 4906999, 0.999999388629, 1, 4906997, 0.999998981048, , , , , , , , , ,
177152, , , , , , , , , , , , , 1, 4907000, 0.999999592419, 1, 4906998, 0.999999184838, , , , , , , , , ,
183296, , , , , , , , , , , , , 1, 4907001, 0.99999979621, 1, 4906999, 0.999999388629, , , , , , , , , ,
189440, , , , , , , 1, 4909925, 0.999999185324, , , , , , , , , , , , , , , , , , ,
214016, , , , 1, 4909928, 0.999999796331, 2, 4909927, 0.999999592662, , , , , , , , , , , , , , , , , , ,
246784, , , , , , , , , , , , , , , , 1, 4907000, 0.999999592419, , , , , , , , , ,
272384, , , , 1, 4909929, 1.0, 1, 4909928, 0.999999796331, , , , , , , , , , , , , , , , , , ,
329728, , , , , , , , , , , , , 1, 4907002, 1.0, 1, 4907001, 0.99999979621, , , , , , , , , ,
1003520, , , , , , , , , , , , , , , , 1, 4907002, 1.0, , , , , , , , , ,
1089536, , , , , , , 1, 4909929, 1.0, , , , , , , , , , , , , , , , , , ,

The first line says that there were three read IOs with 12ns clat,
the cumulative number of read IOs at or below 12ns was two, and
12ns was the 0.0000611th percentile for read latency. There were
two write IOs with 12ns clat, the cumulative number of write IOs
at or below 12ns was two, and 12ns was the 0.0000408th percentile
for write latency.

The job had one write IO complete at 168960ns and 4906999 write IOs
completed at or below this duration. Also this duration was the
99.99994th percentile for write latency. There was one write IO
with a total latency of 168960ns, this duration had a cumulative
frequency of 4906997 write IOs and was the 99.9998981048th percentile
for write total latency.

The last line says that one read IO had 1089536ns total latency, this
duration had a cumulative frequency of 4909929 and represented the 100th
percentile for read total latency.

Running the following:

$ fio_jsonplus_clat2csv fio-jsonplus.output fio-jsonplus.csv --validate
fio-jsonplus_job0.csv validated
fio-jsonplus_job1.csv validated
fio-jsonplus_job2.csv validated

will check the CSV data against the json+ output to confirm that the CSV
data matches.
"""

from __future__ import absolute_import
from __future__ import print_function
import os
import json
import argparse
import itertools
import six

DDIR_LIST = ['read', 'write', 'trim']
LAT_LIST = ['slat_ns', 'clat_ns', 'lat_ns']

def parse_args():
    """Parse command-line arguments."""

    parser = argparse.ArgumentParser()
    parser.add_argument('source',
                        help='fio json+ output file containing completion '
                             'latency data')
    parser.add_argument('dest',
                        help='destination file stub for latency data in CSV '
                             'format. job number will be appended to filename')
    parser.add_argument('--debug', '-d', action='store_true',
                        help='enable debug prints')
    parser.add_argument('--validate', action='store_true',
                        help='validate CSV against JSON output')
    args = parser.parse_args()

    return args


def percentile(idx, run_total):
    """Return a percentile for a specified index based on a running total.

    Parameters:
        idx         index for which to generate percentile.
        run_total   list of cumulative sums.

    Returns:
        Percentile represented by the specified index.
    """

    total = run_total[len(run_total)-1]
    if total == 0:
        return 0

    return float(run_total[idx]) / total


def more_bins(indices, bins):
    """Determine whether we have more bins to process.

    Parameters:
        indices     a dict containing the last index processed in each bin.
        bins        a dict contaiing a set of bins to process.

    Returns:
        True if the indices do not yet point to the end of each bin in bins.
        False if the indices point beyond their repsective bins.
    """

    for key, value in six.iteritems(indices):
        if value < len(bins[key]):
            return True

    return False


def debug_print(debug, *args):
    """Print debug messages.

    Parameters:
        debug       emit messages if True.
        *args       arguments for print().
    """

    if debug:
        print(*args)


def get_csvfile(dest, jobnum):
    """Generate CSV filename from command-line arguments and job numbers.

    Paramaters:
        dest        file specification for CSV filename.
        jobnum      job number.

    Returns:
        A string that is a new filename that incorporates the job number.
    """

    stub, ext = os.path.splitext(dest)
    return stub + '_job' + str(jobnum) + ext


def validate(args, jsondata, col_labels):
    """Validate CSV data against json+ output.

    This function checks the CSV data to make sure that it was correctly
    generated from the original json+ output. json+ 'bins' objects are
    constructed from the CSV data and then compared to the corresponding
    objects in the json+ data. An AssertionError will appear if a mismatch
    is found.

    Percentiles and cumulative counts are not checked.

    Parameters:
        args        command-line arguments for this script.
        jsondata    json+ output to compare against.
        col_labels  column labels for CSV data.

    Returns
        0 if no mismatches found.
    """

    colnames = [c.strip() for c in col_labels.split(',')]

    for jobnum in range(len(jsondata['jobs'])):
        job_data = jsondata['jobs'][jobnum]
        csvfile = get_csvfile(args.dest, jobnum)

        with open(csvfile, 'r') as csvsource:
            csvlines = csvsource.read().split('\n')

        assert csvlines[0] == col_labels
        debug_print(args.debug, 'col_labels match for', csvfile)

        # create 'bins' objects from the CSV data
        counts = {}
        for ddir in DDIR_LIST:
            counts[ddir] = {}
            for lat in LAT_LIST:
                counts[ddir][lat] = {}

        csvlines.pop(0)
        for line in csvlines:
            if line.strip() == "":
                continue
            values = line.split(',')
            nsec = values[0]
            for col in colnames:
                if 'count' in col:
                    val = values[colnames.index(col)]
                    if val.strip() != "":
                        count = int(val)
                        ddir, lat, _, _ = col.split('_')
                        lat = lat + '_ns'
                        counts[ddir][lat][nsec] = count
                        try:
                            assert count == job_data[ddir][lat]['bins'][nsec]
                        except Exception:
                            print("mismatch:", csvfile, ddir, lat, nsec, "ns")
                            return 1

        # compare 'bins' objects created from the CSV data
        # with corresponding 'bins' objects in the json+ output
        for ddir in DDIR_LIST:
            for lat in LAT_LIST:
                if lat in job_data[ddir] and 'bins' in job_data[ddir][lat]:
                    assert job_data[ddir][lat]['bins'] == counts[ddir][lat]
                    debug_print(args.debug, csvfile, ddir, lat, "bins match")
                else:
                    assert counts[ddir][lat] == {}
                    debug_print(args.debug, csvfile, ddir, lat, "bins empty")

        print(csvfile, "validated")

    return 0


def main():
    """Starting point for this script.

    In standard mode, this script will generate CSV data from fio json+ output.
    In validation mode it will check to make sure that counts in CSV files
    match the counts in the json+ data.
    """

    args = parse_args()

    with open(args.source, 'r') as source:
        jsondata = json.loads(source.read())

    ddir_lat_list = list(ddir + '_' + lat for ddir, lat in itertools.product(DDIR_LIST, LAT_LIST))
    debug_print(args.debug, 'ddir_lat_list: ', ddir_lat_list)
    col_labels = 'nsec, '
    for ddir_lat in ddir_lat_list:
        col_labels += "{0}_count, {0}_cumulative, {0}_percentile, ".format(ddir_lat)
    debug_print(args.debug, 'col_labels: ', col_labels)

    if args.validate:
        return validate(args, jsondata, col_labels)

    for jobnum in range(0, len(jsondata['jobs'])):
        bins = {}
        run_total = {}

        for ddir in DDIR_LIST:
            ddir_data = jsondata['jobs'][jobnum][ddir]
            for lat in LAT_LIST:
                ddir_lat = ddir + '_' + lat
                if lat not in ddir_data or 'bins' not in ddir_data[lat]:
                    bins[ddir_lat] = []
                    debug_print(args.debug, 'job', jobnum, ddir_lat, 'not found')
                    continue

                debug_print(args.debug, 'job', jobnum, ddir_lat, 'processing')
                bins[ddir_lat] = [[int(key), value] for key, value in
                                  six.iteritems(ddir_data[lat]['bins'])]
                bins[ddir_lat] = sorted(bins[ddir_lat], key=lambda bin: bin[0])

                run_total[ddir_lat] = [0 for x in range(0, len(bins[ddir_lat]))]
                run_total[ddir_lat][0] = bins[ddir_lat][0][1]
                for index in range(1, len(bins[ddir_lat])):
                    run_total[ddir_lat][index] = run_total[ddir_lat][index-1] + \
                        bins[ddir_lat][index][1]

        csvfile = get_csvfile(args.dest, jobnum)
        with open(csvfile, 'w') as output:
            output.write(col_labels + "\n")

#
# Have a counter for each ddir_lat pairing
# In each round, pick the shortest remaining duration
# and output a line with any values for that duration
#
            indices = {x: 0 for x in ddir_lat_list}
            while more_bins(indices, bins):
                debug_print(args.debug, 'indices: ', indices)
                min_lat = 17112760320
                for ddir_lat in ddir_lat_list:
                    if indices[ddir_lat] < len(bins[ddir_lat]):
                        min_lat = min(bins[ddir_lat][indices[ddir_lat]][0], min_lat)

                output.write("{0}, ".format(min_lat))

                for ddir_lat in ddir_lat_list:
                    if indices[ddir_lat] < len(bins[ddir_lat]) and \
                       min_lat == bins[ddir_lat][indices[ddir_lat]][0]:
                        count = bins[ddir_lat][indices[ddir_lat]][1]
                        cumulative = run_total[ddir_lat][indices[ddir_lat]]
                        ptile = percentile(indices[ddir_lat], run_total[ddir_lat])
                        output.write("{0}, {1}, {2}, ".format(count, cumulative, ptile))
                        indices[ddir_lat] += 1
                    else:
                        output.write(", , , ")
                output.write("\n")

            print("{0} generated".format(csvfile))


if __name__ == '__main__':
    main()