gaainf / tanktools

Yandex-tank tools

Yandex-tank input/output file parser

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.

Generate Yandex-tank ammo from pcap or har files using pcap2ammo or har2ammo scrips. HTTP requests are extracted completely with headers and body.

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
• convert pcap file to ammo
• convert har file to ammo
• filter out and modify requests on ammo generating

Installation

pip install tanktools

Examples

Select DataFrame by timestamp

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

from tanktools import phout
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('phout.log', 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

pcap2ammo

Convert pcap file to Yandex-tank ammo

pcap2ammo file.pcap

73
GET https://rambler.ru/ HTTP/1.1\r\n
Host: rambler.ru\r\n
Content-Length: 0\r\n\r\n

Count statistics about HTTP requests

pcap2ammo -S file.pcap

Stats:
    total: 1
    complete: 1
    incorrect: 0
    incomplete: 0

Print to file

pcap2ammo -o out.ammo file.pcap

Add or delete headers

Applyed for all requests, containing specified headers

pcap2ammo --add-header 'Referer: http://domain.com' --add-header 'X-Ip: 1.1.1.1' file.pcap

pcap2ammo --delete-header 'Content-Length' file.pcap
pcap2ammo --delete-header 'Connection' --add-header 'Connection: keep-alive' file.pcap

Filter TCP/IP packets

pcap2ammo -f 'ip.src==10.10.10.10 and tcp.dport==8080' file.pcap

Filter HTTP packets

pcap2ammo -F '"rambler.ru" in http.uri' file.pcap

You can use logical expressions in filters

pcap2ammo -F '"keep-alive" in http.headers["connection"] or "Keep-alive" in http.headers["connection"]' file.pcap

String functions over HTTP headers

pcap2ammo -F '"keep-alive" in http.headers["connection"].lower()' file.pcap

Use excluding filters also

pcap2ammo -F '"rambler.ru" != http.headers["host"]' file.pcap

har2ammo

Convert pcap file to Yandex-tank ammo

har2ammo file.har

73
GET https://rambler.ru/ HTTP/1.1\r\n
Host: rambler.ru\r\n
Content-Length: 0\r\n\r\n

Count statistics about HTTP requests

har2ammo -S file.har

Stats:
    total: 1
    complete: 1
    incorrect: 0
    incomplete: 0

Print to file

har2ammo -o out.ammo file.har

Add or delete headers

Applyed for all requests, containing specified headers

har2ammo --add-header 'Referer: http://domain.com' --add-header 'X-Ip: 1.1.1.1' file.har

har2ammo --delete-header 'Content-Length' file.har
har2ammo --delete-header 'Connection' --add-header 'Connection: keep-alive' file.har

Filter HTTP packets

har2ammo -F '"rambler.ru" in http.uri' file.har

You can use logical expressions and python functions in filters

har2ammo -F '"keep-alive" in http.headers["connection"] or "Keep-alive" in http.headers["connection"]' file.har
har2ammo -F '"keep-alive" not in http.headers["connection"].lower()' file.har

Please, see more information about filters in pcaper package description.

tanktools's Issues

Wrong request size calculation in pcap2ammo tool

Yandex Tank expects the request size in bytes (in utf-8 encoding) whereas the tool calculates size as count of characters in the request. This solution worked for most of requests because they contained only 1-byte characters. Unfortunately, UTF-8 uses 1-4 bytes per character in general.