Generic computation implementation on COINSTAC.
Project description
COINSTAC computations development made easy.
A very intuitive wrapper for writing coinstac based computations:
- Break down your computations in simple phases with automatic transition between phases.
- Add as many phases as you want.
- Even run phases that needs to be run multiple local-remote iterations.
- Installation:
pip install coinstac-computation
- Add entry
coinstac-computation
to the requirements.txt.
Example: Gather max even numbers from each site
A full working use case is in the examples/basic directory where:
- Local sites filters out even numbers and sends to the remote.
- Remote finds the max across sites and returns the final result to each of the sites.
- Sites save final result.
inputspec.json data:
[
{
"data": {
"value": [10, 3, 5, 6, 7, 8, 12, 38, 32, 789, 776, 441]
}
},
{
"data": {
"value": [12, 33, 88, 61, 37, 58, 103, 3386, 312, 9, 77, 41]
}
}
]
Local node pipeline:
import os
from coinstac_computation import COINSTACPyNode, ComputationPhase
class PhaseLoadData(ComputationPhase):
def compute(self):
data = []
for d in self.input['data']:
if d % 2 == 0:
data.append(d)
return {'filtered_data': data}
class PhaseSaveResult(ComputationPhase):
def compute(self):
with open(f"{self.state['outputDirectory'] + os.sep + 'results.txt'}", 'w') as out:
out.write(f"{self.input['aggregated_data']}")
local = COINSTACPyNode(mode='local', debug=True)
local.add_phase(PhaseLoadData)
local.add_phase(PhaseSaveResult)
Remote node pipeline:
from coinstac_computation import COINSTACPyNode, ComputationPhase, PhaseEndWithSuccess
class PhaseCollectMaxEvenData(ComputationPhase):
def compute(self):
data = []
for site, site_vars in self.input.items():
site_max = max(site_vars['filtered_data'])
data.append(site_max)
return {'aggregated_data': data}
remote = COINSTACPyNode(mode='remote', debug=True)
remote.add_phase(PhaseCollectMaxEvenData)
remote.add_phase(PhaseEndWithSuccess)
Entry point:
import coinstac
from local_pipeline import local
from remote_pipeline import remote
coinstac.start(local.compute, remote.compute)
Run:
cd example
docker build -t base . && coinstac-simulator
Advanced use case: Phases with multiple iterations.
Overview:
- Specify a phase as multi-iterations:
local.add_phase(SomeIterativePhase, multi_iterations=True)
- Specify when to end the iterative phase as:
class SomeIterativePhase(ComputationPhase):
def compute(self):
"""Do all the stuff"""
"""Check if the iterative phase is done, and send a phase jump signal."""
should_jump_to_next_phase = ...
return {..., 'jump_to_next': should_jump_to_next_phase}
Full working example where:
- Each sites cast a vote for multiple(default=51) times.
- Remote gathers the votes and returns the final voting result at the end.
- Sites save the final result.
Development notes:
- Make sure you have:
- docker installed and running.
- nodejs installed.
- coinstac-simulator package installed.
npm install --global coinstac-simulator
- Must set
debug=False
while deploying. - Backward compatible to the older library(compspecVersion=1):
- Add the following snippet at the end of local and remote pipeline scripts.
if __name__ == "__main__": local.to_stdout()
- Use version 1.0 compspec format.
- Comment out line
CMD ["python", "entry.py"]
in theDockerfile
. - You can also use a remote debugger in pycharm as here.
Thanks!
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Close
Hashes for coinstac-computation-0.55.tar.gz
Algorithm | Hash digest | |
---|---|---|
SHA256 | 09929ec25d4ab4a940af0de29426b16a1379c66ac9c24e0450767e21027ba795 |
|
MD5 | 3c239d6e5bbaf1cd8fdd721e4cccaa21 |
|
BLAKE2b-256 | cd5c7d0508e0dccd086071fb5486d2250e3ca9b52553e3c1f320f0689d1bc224 |