tanktools 1.0.2

Yandex-tank tools

Yandex-tank prepare phout file with statistics after load testing. tanktools module helps to parse such files and convert to DataFrame. You can use pandas module in manual mode to handle DataFrame or use build-in functions.

So you can:

• calc quantiles

• get information about timings, latency, status codes

• extract requests by timestamp, tag and other columns

• group and analyze specific data like total/partial RPS, average request/response size

• calc statistical metrics

Installation

pip install tanktools

Import

from tanktools import phout

Examples

Select DataFrame by timestamp

It is possible to parse a part of staistics by time and overal count

flags = {
    'from_date': '2018-01-18 20:09:50.123',
    'to_date'  : '2018-01-18 20:10:00.456',
    'limit': 100
}
data = phout.parse_phout(args.input, flags)
print("Total count: %d" % phout.size(data))

Total count: 100

Print percentiles

data = phout.parse_phout('phout.log')
phout.print_quantiles(data, 'receive_time')

Percentiles for 150030 requests
    from 2018-01-18 20:09:42.983
    to   2018-01-18 20:10:55.108:
quantile (%)  receive_time (mks)
        10.0                   9
        20.0                   9
        30.0                  10
        40.0                  10
        50.0                  10
        60.0                  10
        70.0                  11
        80.0                  12
        90.0                  13
        95.0                  14
        98.0                  16
        99.0                  17
        100.0                 716

Note

Pay attention, timings are calculated in microseconds.

Print latency median

data = phout.parse_phout('phout.log')
# Convert and print timing in milliseconds
print("\n\nLatency median: %d ms" % int(data.latency.median() / 1000))

Latency median: 30 ms

Get RPS

data = phout.parse_phout('phout.log')
rps = phout.get_rps(data)

Print HTTP response statistics

data = phout.parse_phout('phout.log')
phout.print_http_reponses(data)

HTTP code   count  percent (%)
     500   83429        56.38
     200   61558        41.60
     502    2944         1.99
       0      41         0.03

Select 200 OK responses and print latency median

data = phout.parse_phout('phout.log')
selected_http_responses = data[data.proto_code == 200]
print("Latency median for 200 OK: %d" %
      selected_http_responses.latency.median())

Latency median for 200 OK: 3539

Print average request/response size

print("Avg. Request / Response: %d / %d bytes" % (
    data.size_in.astype(float).mean(),
    data.size_out.astype(float).mean()
))

Avg. Request / Response: 364 / 26697 bytes

Note

Pay attention it is required to convert data to float for correct work of mean function

Print RPS at Nth request

print("RPS at request:")
chunk_size = int(phout.size(data) / 2)
for start in range(0, phout.size(data), chunk_size):
    data_subset = phout.subset(data, start, chunk_size)
    print("\t%s: %.2f" %
          (start + chunk_size, phout.get_rps(data_subset)))

RPS at request:
    73986: 2062.50
    147972: 2530.56