README.md

sneks

Get your snakes in a row.

sneks lets you launch a Dask cluster in the cloud, matched to your local software environment*, in a single line of code. No more dependency mismatches or Docker image building.

from sneks import get_client

client = get_client()

*your local Poetry or PDM environment. You must use poetry or PDM. Locking package managers are what sensible people use, and you are sensible.

Neat! Sneks also supports ARM clusters! Just pass ARM instances in scheduler_instace_types=, worker_instace_types= and cross your fingers that all your dependencies have cross-arch wheels!

Installation

poetry add -G dev sneks-sync

A full example:

mkdir example && cd example
poetry init -n
poetry add -G dev sneks-sync
poetry add distributed==2022.5.2 dask==2022.5.2 bokeh pandas pyarrow  # and whatever else you want

from sneks import get_client
import dask.dataframe as dd

client = get_client(name="on-a-plane")
ddf = dd.read_parquet(
    "s3://nyc-tlc/trip data/yellow_tripdata_2012-*.parquet",
)
print(ddf.groupby('passenger_count').trip_distance.mean().compute())

Oh wait, we forgot to install a dependency!

poetry add foobar

When we reconnect to the cluster (using the same name), the dependencies on the cluster update automatically.

from sneks import get_client
import dask.dataframe as dd
import foobar  # ah, how could we forget this critical tool

client = get_client(name="on-a-plane")
ddf = dd.read_csv(
    "s3://nyc-tlc/csv_backup/yellow_tripdata_2012-*.csv",
)
means = ddf.groupby('passenger_count').trip_distance.mean()
means.apply(foobar.optimize).compute()

Caveats

This is still a proof-of-concept-level package. It's been used personally quite a bit, and proven reliable, but use at your own risk.