Skip to content
/ pumpkin Public

Distributed data processing network in python work in progress....

License

Notifications You must be signed in to change notification settings

recap/pumpkin

Repository files navigation

Pumpkin

Pumpkin is a framework for distributed Data Transformation Network (DTN). It is similar to an actor model but data is dynamically routed from one actor to the next. In our model actors are seeds and these can be added to the network dynamically, transformation paths are learnt and data can automatically start flowing to the new seeds.

Download

by cloning

git clone https://github.com/recap/pumpkin.git

by archive

https://github.com/recap/pumpkin/archive/master.zip

Install

If you want to install local

python setup.py install --prefix=~/.local

If you want to install pumpkin to the system:

sudo python setup.py install

Hello World

run:

./bin/pmk --endpoint.type zmq.IPC

If all goes well you should get something like:

INFO:Node assigned UID: slimme-783d9470
INFO:Exec context: e4a460ff
INFO:Node bound to IP: 192.168.1.50
INFO:Starting broadcast listener on port 7700
INFO:Discovered new peer: extract at ipc:///tmp/slimme-783d9470
INFO:Discovered new peer: inject at ipc:///tmp/slimme-783d9470
INFO:Discovered new peer: greet at ipc:///tmp/slimme-783d9470
Greeting: Hello World

you can exit with ctrl-c

Short Explanation

What is happening is that by default Pumpkin il loading seeds from examples/helloworld. If you take a look in that directory you notice 3 files inject.py, greet.py and extract.py. These 3 seeds form a chain to produce the output "Hello World". inject.py is injecting "World" string as data, the greet.py attaches the greeting "Hello" and the extract.py prints the output. These seeds are loosely coupled so much so that they can run on different machines (more on this later).

+----------+         +----------+       +-----------+
|          |         |          |       |           |
|  inject  |-------> |  greet   |------>|  extract  |
|          |         |          |       |           |
+----------+         +----------+       +-----------+

Injecting Data

There are several ways of injecting data. The above demonstrates the use of an injector seed. Another way is through files. if we rerun the above example but first:

create a tmp directory e.x.

mkdir /tmp/rx

then run:

./bin/pmk --endpoint.type zmq.IPC --rx /tmp/rx

This will output the same as above and show "Hello World". At theis point we can inject more data into /tmp/rx directory. To do this you can copy a data packet file [examples/helloworld/data.pkt] to /tmp/rx and check the console output for pmk. If all goes well a new output should be:

Greeting: Hello Mars

Distributing Seeds

Pumpkin allows seeds to be disperesed on different computers. To try this out lets force greet to run on a different host. First install Pumpkin on a differetn host we will call host B. Our current host is host A. On A locate the file greet.py under examples/helloworld/. Edit the file and change the line

##"auto-load" : true,

to

##"auto-load" : false,

This instruct Pumpkin not to load the seed greet automatically. Now on host A run Pumpkin with:

./bin/pmk

If you notice the output only lists 2 seeds

INFO:Discovered new peer: extract at tcp://192.168.1.50:7900
INFO:Discovered new peer: inject at tcp://192.168.1.50:7900

The greet seed is missing. We will host greet on B. After installing Pumpkin on B run the following:

./bin/pmk --seed ./examples/helloworld/greet.py

This instructs Pumpkin to only load the greet seed. After a few seconds you should see "Hello World" as output on A. This is because Pumpkins on A and B discover each other given they are on a LAN and data can flow.

Data Transformation Network

Using the meta data given for each seed about its inputs and outputs, Pumpkin builds data automaton which represents the possible states data can be in the network. The DTN for the hello world example is given below.

dtn

About

Distributed data processing network in python work in progress....

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published