3 # fio_jsonplus_clat2csv
5 # This script converts fio's json+ completion latency data to CSV format.
9 # Run the following fio jobs:
10 # ../fio --output=fio-jsonplus.output --output-format=json+ --name=test1
11 # --ioengine=null --time_based --runtime=5s --size=1G --rw=randrw
12 # --name=test2 --ioengine=null --time_based --runtime=3s --size=1G
13 # --rw=read --name=test3 --ioengine=null --time_based --runtime=4s
14 # --size=8G --rw=write
17 # fio_jsonplus_clat2csv fio-jsonplus.output fio-latency.csv
19 # You will end up with the following 3 files
21 # -rw-r--r-- 1 root root 6467 Jun 27 14:57 fio-latency_job0.csv
22 # -rw-r--r-- 1 root root 3985 Jun 27 14:57 fio-latency_job1.csv
23 # -rw-r--r-- 1 root root 4490 Jun 27 14:57 fio-latency_job2.csv
25 # fio-latency_job0.csv will look something like:
27 # clat_nsec, read_count, read_cumulative, read_percentile, write_count,
28 # write_cumulative, write_percentile, trim_count, trim_cumulative,
30 # 25, 1, 1, 1.50870705013e-07, , , , , , ,
31 # 26, 12, 13, 1.96131916517e-06, 947, 947, 0.000142955890032, , , ,
32 # 27, 843677, 843690, 0.127288105112, 838347, 839294, 0.126696959629, , , ,
33 # 28, 1877982, 2721672, 0.410620573454, 1870189, 2709483, 0.409014312345, , , ,
34 # 29, 4471, 2726143, 0.411295116376, 7718, 2717201, 0.410179395301, , , ,
35 # 30, 2142885, 4869028, 0.734593687087, 2138164, 4855365, 0.732949340025, , , ,
37 # 2544, , , , 2, 6624404, 0.999997433738, , , ,
38 # 2576, 3, 6628178, 0.99999788781, 4, 6624408, 0.999998037564, , , ,
39 # 2608, 4, 6628182, 0.999998491293, 4, 6624412, 0.999998641391, , , ,
40 # 2640, 3, 6628185, 0.999998943905, 2, 6624414, 0.999998943304, , , ,
41 # 2672, 1, 6628186, 0.999999094776, 3, 6624417, 0.999999396174, , , ,
42 # 2736, 1, 6628187, 0.999999245646, 1, 6624418, 0.99999954713, , , ,
43 # 2768, 2, 6628189, 0.999999547388, 1, 6624419, 0.999999698087, , , ,
44 # 2800, , , , 1, 6624420, 0.999999849043, , , ,
45 # 2832, 1, 6628190, 0.999999698259, , , , , , ,
46 # 4192, 1, 6628191, 0.999999849129, , , , , , ,
47 # 5792, , , , 1, 6624421, 1.0, , , ,
48 # 10304, 1, 6628192, 1.0, , , , , , ,
50 # The first line says that you had one read IO with 25ns clat,
51 # the cumulative number of read IOs at or below 25ns is 1, and
52 # 25ns is the 0.00001509th percentile for read latency
54 # The job had 2 write IOs complete in 2544ns,
55 # 6624404 write IOs completed in 2544ns or less,
56 # and this represents the 99.99974th percentile for write latency
58 # The last line says that one read IO had 10304ns clat,
59 # 6628192 read IOs had 10304ns or shorter clat, and
60 # 10304ns is the 100th percentile for read latency
69 parser = argparse.ArgumentParser()
70 parser.add_argument('source',
71 help='fio json+ output file containing completion '
73 parser.add_argument('dest',
74 help='destination file stub for latency data in CSV '
75 'format. job number will be appended to filename')
76 args = parser.parse_args()
81 def percentile(idx, run_total):
82 total = run_total[len(run_total)-1]
86 return float(run_total[idx]) / total
89 def more_lines(indices, bins):
90 for key, value in indices.iteritems():
91 if value < len(bins[key]):
100 with open(args.source, 'r') as source:
101 jsondata = json.loads(source.read())
103 for jobnum in range(0, len(jsondata['jobs'])):
106 ddir_set = set(['read', 'write', 'trim'])
109 for ddir in ddir_set:
110 if 'bins' in jsondata['jobs'][jobnum][ddir]['clat_ns']:
112 elif 'bins' in jsondata['jobs'][jobnum][ddir]['lat_ns']:
115 raise RuntimeError("Latency bins not found. "
116 "Are you sure you are using json+ output?")
118 bins[ddir] = [[int(key), value] for key, value in
119 jsondata['jobs'][jobnum][ddir][bins_loc]
120 ['bins'].iteritems()]
121 bins[ddir] = sorted(bins[ddir], key=lambda bin: bin[0])
123 run_total[ddir] = [0 for x in range(0, len(bins[ddir]))]
124 if len(bins[ddir]) > 0:
125 run_total[ddir][0] = bins[ddir][0][1]
126 for x in range(1, len(bins[ddir])):
127 run_total[ddir][x] = run_total[ddir][x-1] + \
130 stub, ext = os.path.splitext(args.dest)
131 outfile = stub + '_job' + str(jobnum) + ext
133 with open(outfile, 'w') as output:
134 output.write("{0}ec, ".format(bins_loc))
135 ddir_list = list(ddir_set)
136 for ddir in ddir_list:
137 output.write("{0}_count, {0}_cumulative, {0}_percentile, ".
142 # Have a counter for each ddir
143 # In each round, pick the shortest remaining duration
144 # and output a line with any values for that duration
146 indices = {x: 0 for x in ddir_list}
147 while more_lines(indices, bins):
148 min_lat = 17112760320
149 for ddir in ddir_list:
150 if indices[ddir] < len(bins[ddir]):
151 min_lat = min(bins[ddir][indices[ddir]][0], min_lat)
153 output.write("{0}, ".format(min_lat))
155 for ddir in ddir_list:
156 if indices[ddir] < len(bins[ddir]) and \
157 min_lat == bins[ddir][indices[ddir]][0]:
158 count = bins[ddir][indices[ddir]][1]
159 cumulative = run_total[ddir][indices[ddir]]
160 ptile = percentile(indices[ddir], run_total[ddir])
161 output.write("{0}, {1}, {2}, ".format(count,
165 output.write(", , , ")
168 print "{0} generated".format(outfile)
171 if __name__ == '__main__':