ntimes

ntimes: time(1) with average time, flaky rate, ..

ntimes is an improved version of the time(1) command.

You can execute an command N times, and measure the average/max/min time taken for the execution. You can also measure the "flaky" rate (i.e. failure rate).

Features

Metrics (performance):

• average time
• max and min
• standard deviation
• percentile
• histogram (TBD output format)

Metrics (flakiness):

• flaky rate

Others:

• JSON output
• stdout and stderr storage
• skip "warming up" (e.g. cache) iterations for stat
• handle signals (e.g. ^C)
• parallel execution

Install

For Linux and macOS (experimental):

curl -L https://github.com/AkihiroSuda/ntimes/releases/download/v0.1.0/ntimes-`uname -s`-`uname -m` >/usr/local/bin/ntimes && \
chmod +x /usr/local/bin/ntimes

For Windows (experimental): https://github.com/AkihiroSuda/ntimes/releases/download/v0.1.0/ntimes-windows-x86_64.exe

Latest development version (requires Go):

$ go get github.com/AkihiroSuda/ntimes

Usage

Example usage:

$ ntimes -n 10 bash -c 'sleep=$((RANDOM%5)); fail=$((RANDOM%2)); echo "id=$NTIMES_ID, sleep=$sleep, fail=$fail"; sleep $sleep; exit $fail'
id=0, sleep=4, fail=0
id=1, sleep=0, fail=0
id=2, sleep=3, fail=1
id=3, sleep=4, fail=0
id=4, sleep=3, fail=1
id=5, sleep=0, fail=1
id=6, sleep=3, fail=1
id=7, sleep=0, fail=0
id=8, sleep=3, fail=1
id=9, sleep=3, fail=1

real average: 2.303595053s, max: 4.004607594s, min: 3.455164ms, std dev: 1.636182569s
real 99 percentile: 4.004607594s, 95 percentile: 4.004607594s, 50 percentile: 3.002190904s
user average: 0, max: 0, min: 0, std dev: 0
sys  average: 0, max: 0, min: 0, std dev: 0
flaky: 60%

You can specify the report format using Go's text/template syntax. Additionally to the standard functions provided by text/template, the json function is available as well. Note that a time.Duration value is expressed in nanoseconds.

$ ntimes --format "{{json .}}" -n 10 dd if=/dev/urandom of=/dev/null bs=512 count=1000
1000+0 records in
1000+0 records out
512000 bytes (512 kB, 500 KiB) copied, 0.0448978 s, 11.4 MB/s
...
{"real":{"average":44155207,"max":68222928,"min":38143407,"stddev":9421337,"percentiles":{"50":39855284,"95":68222928,"99":68222928}},"user":{"average":0,"max":0,"min":0,"stddev":0},"system":{"average":36000000,"max":36000000,"min":36000000,"stddev":0},"flaky":0}

Practical example for debugging flaky tests with Namazu (nmz, osrg/namazu):

$ cd some_maven_project
$ sudo ntimes -n 10 --storage /tmp/logs nmz inspectors -cmd "mvn test"
...
Flaky: 10%

$ find /tmp/logs -name result.json | xargs jq .successful
true
true
true
false
true
true
true
true
true
true

Please refer to ntimes --help for the detailed help.

$ ./ntimes --help
Usage: ./ntimes [OPTIONS] COMMAND [ARG...]
  -f, --format string         format string (in golang text/template, e.g. "{{json .}}")
  -n, --repeat-n-times uint   number of times (default 1)
  --storage string            path to stdout,stderr storage
  --version                   print version to stdout and exit
  --warm-up uint              skip first n iterations for stat

Motivation

Originally, ntimes was designed so that it can be combined with osrg/namazu. Namazu can be used for controlling non-deternimism and increasing reproducibility of flaky test failures. ntimes can be used for measuring the reproducibility of flaky test failures.

Specification

• When running a command, NTIMES_ID is passed to the command as an environment variable. The value of NTIMES_ID denotes a non-negative decimal value corresponding to the iteration count. (0, 1, .., n-1)

• The format of statistics report is defined as Stat structure in common.go.

• If --storage dir is specified, following files are created in dir/$NTIMES_ID:

  • stdout: containes the standard output
  • stderr: containes the standard err
  • result.json: JSON representation of a Result structure (defined in common.go)

Related Project

Some projects similar to ntimes:

• jmcabo/avgtime
• ryanmjacobs/avgtime
• kevinstreit/avgtime