pachyderm-sdk

Pachyderm's Python SDK

Official Python client/SDK for Pachyderm. The successor to https://github.com/pachyderm/python-pachyderm.

This library provides the autogenerated gRPC/protobuf code for Pachyderm, generated using a fork of the betterproto package, along with higher-level functionality.

Installation

pip install pachyderm_sdk

A Small Taste

Here's an example that creates a repo and adds a file:

from pachyderm_sdk import Client
from pachyderm_sdk.api import pfs

# Connects to a pachyderm cluster using your local config
#   at ~/.pachyderm/config.json
client = Client.from_config()

# Creates a pachyderm repo called `test`
repo = pfs.Repo(name="test")
client.pfs.create_repo(repo=repo)

# Create a new commit in `test@master` and upload a file.
branch = pfs.Branch.from_uri("test@master")
with client.pfs.commit(branch=branch) as commit:
    file = commit.put_file_from_bytes(path="/data/file.dat", data=b"DATA")

# Retrieve the uploaded file.
with client.pfs.pfs_file(file) as f:
    print(f.readall())

How to load a CAST file into a pandas dataframe

from pachyderm_sdk import Client
from pachyderm_sdk.api import pfs
import pandas as pd

client = Client.from_config()
file = pfs.File.from_uri("test@master:/path/to/data.csv")
with client.pfs.pfs_file(file) as f:
    df = pd.read_csv(f)

Changes from Python-Pachyderm

This package is a successor to the python-pachyderm package. Listed below are some of the notable changes:

1. Organization of the API
   • Methods and Message objects are now organized according to the service they are associated with, i.e. auth, pfs (pachyderm file-system), pps (pachyderm pipelining-system).
   • Message objects can be found within their respective submodule of the pachyder_sdk.api module, i.e. pachyderm_sdk.api.pfs.
   • Methods can be found within their respective attribute of the Client class, i.e. client.pps.create_pipeline.
     • Some methods have been renamed to remove redundancy due to this organization, i.e. python_pachyderm.Client.get_enterprise_state -> pachyderm_sdk.Client.enterprise.get_state
2. The autogenerated code is generated using a fork of the betterproto compiler.
   • Messages are now python dataclasses.
   • Methods require keyword arguments.
   • Pachyderm resources are specified using types.
     • python-pachyderm (old): client.create_repo("test")
     • pachyderm_sdk (new): client.pfs.create_repo(repo=pfs.Repo(name="test"))

Contributing

Please see the contributing guide for more info (including testing instructions)

Developer Guide

Generate python APIs from protobuf:

./generate-protos.sh

Generate HTML documentation (writes to docs/pachyderm_sdk):

make docs

Running Tests:

pytest -vvv tests