FOSSASIA Summit 2018

# Building microservices with firefly!

## Nabarun Pal

---

# About Me

### Nabarun Pal
- Final Year Undergraduate @IITRoorkee

- Software Engineering Intern @rorodata

- Builds autonomous robots @TeamRoboconIITR

- Sources of my dopamine -> [Code, Kindle, Food, (Sugar, Caffeine)]

.footnote.right.bottom[www.nabarun.in github.com/palnabarun twitter.com/_palnabarun]

---

# The Problem

???
exposing functions as API's for others to use

---

# But why?

- Use functions in a different environment

- Loose couple functionalities of an appplication

- Run functions on different compute resources

---

# Use Cases

- Deploy machne learning models

- Run functional utilities remotely

- Build microservices

---

# Challenges

- Need to writing a web application

- Requirement of a client library

- How would you authenticate users?

- How do you validate the data sent by users?

---

# Why firefly?

- Deploying functions is a cake walk

- Deploying predictive models by adding minimal code

- Functions accessible through a REST based API or a RPC like client

- It puts your python functions on steroids!

---

# Install

Release version:
```bash
$ pip install firefly-python
```

Development version:
```bash
$ git clone https://github.com/rorodata/firefly
$ cd firefly
$ pip install -e .
```

---

# Code

Your usual code:

```python
# funcs.py

def square(n):
  return n**2
```

---

# Run

Invoke firefly:

```bash
$ firefly funcs.square
2017-11-04 01:55:02 firefly [INFO] Starting Firefly...
http://127.0.0.1:8000/
```

Multiple function invokation:

```bash
$ firefly funcs.square funcs.cube
2017-11-04 01:55:02 firefly [INFO] Starting Firefly...
http://127.0.0.1:8000/
```

or, using Gunicorn:

```bash
$ gunicorn firefly.main:app -e FIREFLY_FUNCTIONS="funcs.square"
[2017-11-04 01:59:36 +0530] [14182] [INFO] Starting gunicorn 19.7.1
[2017-11-04 01:59:36 +0530] [14182] [INFO] Listening at: http://127.0.0.1:8000 (14182)
[2017-11-04 01:59:36 +0530] [14182] [INFO] Using worker: sync
[2017-11-04 01:59:36 +0530] [14185] [INFO] Booting worker with pid: 14185
2017-11-04 01:59:36 firefly [INFO] Starting Firefly...
```

---

# Usage

But, how would you use the functions?

We have a client for you.

```python
>>> import firefly
>>> client = firefly.Client("http://127.0.0.1:8000/")
>>> client.square(n=4)
16
```

But, how does that work?

It is a simple RESTful API.

```bash
$ curl -d '{"n": 4}' http://127.0.0.1:8000/square
16
```

The client supports any JSON-friendly data type or file objects.

---

# Config File

firefly includes support for specifying functions and paths from config files.

```yaml
# firefly.yml

functions:
  square:
    path: "/square"
    function: "funcs.square"
  cube:
    path: "/cube"
    function: "funcs.cube"
  ...
```

Specify the config file as:

```bash
$ firefly -c firefly.yml
http://127.0.0.1:8000/
...
```

or,

```bash
$ gunicorn firefly.main:app -e FIREFLY_CONFIG="firefly.yml"
...
```

---

# Authentication

Ability to specify an auth token for requests

```bash
$ firefly --token qwerty123 funcs.square
...
```

And you can use it like this:

```python
>>> import firefly
>>> client = firefly.Client("http://127.0.0.1:8000/", auth_token="qwerty123")
>>> client.square(n=4)
16
```

---

# Deploying ML models

```python
# model.py
from sklearn.externals import joblib

clf = joblib.load('fisheriris.pkl')

def predict(a):
    predicted = clf.predict(a)    # predicted is 1x1 numpy array
    return int(predicted[0])
```

Start firefly:

```bash
$ firefly model.predict
...
```

Using the client for accessing the remote function:

```python
>>> import firefly
>>> remote_model = firefly.Client("http://<server_ip>/")
>>> remote_model.predict(features=[5.9, 3, 5.1, 1.8]))
2
```

---

# Under the hood

- Client invokes a function

- Client processes the inputs

- Client makes a request to the appropriate endpoint on the server

- Server checks whether the endpoint requested is exposed

- Server processes the request and validates the inputs

- The function is then called passing in the inputs

- Function result is processed to either JSON/binary stream

- Server responds to the client with appropriate status code and data

---

# Side effects

???
- Microservices
- rorocloud
- rorolite

---

# Example microservices

???
- firefly.yml
- write a simple firefly.yml and show how you deploy it
- background-removal-demo

---

# Coming Soon...

- Authentication methods like OAuth, Social Login
- True RESTful API with support for all HTTP methods
- Canonical URL's
- Input Validation using Python 3 Type Annotations
- Support for NumPy arrays
- API browser 
- JavaScript Client library

---

# Questions?

---

# References

https://firefly-python.readthedocs.io

https://github.com/rorodata/firefly

https://github.com/rorodata/rorolite

---

# Feedback