mircealungu / benchmark

Measuring the overhead of Flask Monitoring Dashboard

Benchmark

Framework for measuring the overhead of Flask Monitoring Dashboard on the monitored Flask application.

How to run

There are 3 steps you have to take:

1. Set up the environment

Install a virtual environment with virtualenv:

virtualenv --python=python3.7 ENV

Activate it:

source ENV/bin/activate

Go in the root of the project and install the requirements:

pip install -r requirements.txt

2. Run benchmarks

Make sure to configure the benchmark parameters. Use the config_micro.ini and config_macro.ini files for the micro and macro benchmarks respectively.

2.1. Micro benchmarks

To run the micro benchmarks, go in the root of the project and run:

make run_micro

The results are saved into json files, in the /results/micro folder.

2.2. Macro benchmark

The macro benchmark consists of a real world Flask application, Conduit. To set up the Conduit app type the following commands from the root of the project:

cd macro
export FLASK_APP=./autoapp.py
Flask db init
Flask db migrate
Flask db upgrade

To generate a load run from the root:

python -m caller --load true

This will create an SQLite db of ~63MB, named as in the config file. The db consists of 500 users, each with 10 articles of 10KB each, 30 comments per user, 3 tags per article, each user follows 3 other users and favorites 3 articles.

To run the macro benchmarks, go in the root of the project and run:

make run_macro

The results are saved into json files, in the /results/macro folder.

TODO

3. Visualize results

The results are saved into json files, in the /results folder. There are 2 ways of viewing them:

3.1 Using the pyperf package

From the root run:

python -m pyperf compare_to results/micro/190308_15:25:37/-1.json results/micro/190308_15:25:37/3.json --table

This will show a table similar to:

+---------------------+--------+------------------------------+
| Benchmark           | -1     | 3                            |
+=====================+========+==============================+
| pidigits            | 284 ms | 314 ms: 1.11x slower (+11%)  |
+---------------------+--------+------------------------------+
| float               | 255 ms | 304 ms: 1.19x slower (+19%)  |
+---------------------+--------+------------------------------+
| json_loads          | 250 ms | 291 ms: 1.16x slower (+16%)  |
+---------------------+--------+------------------------------+
| pathlib             | 235 ms | 700 ms: 2.98x slower (+198%) |
+---------------------+--------+------------------------------+
| sqlalchemy_combined | 207 ms | 589 ms: 2.84x slower (+184%) |
+---------------------+--------+------------------------------+
| sqlalchemy_writes   | 211 ms | 531 ms: 2.51x slower (+151%) |
+---------------------+--------+------------------------------+
| sqlalchemy_reads    | 208 ms | 353 ms: 1.69x slower (+69%)  |
+---------------------+--------+------------------------------+

Make sure to replace the files in the command above with your own.

3.2 Using the viewer (local) package

To get plots of all the benchmarks, type:

python -m viewer --path results/micro/190415_16:25:30 --type t

where t can be:

• violin for violin plots
• line for line graphs
• both

To get line charts of the overheads, give a list of results directories to the --path argument, as shown below:

python -m viewer --path results/micro/190415_16:25:30 results/micro/190415_13:06:04

To get violin plots of the most recent results, type:

make view

Clean up

To remove all the databases created by the benchmarks and FMD, run:

make clear

To remove all the html files containing the plots, run:

make free_space

If there's a failure, the server will not shutdown and port 5000 is unusable. To shut down the server, run:

make shutdown