Droplet is a low-latency, stateful serverless programming framework built on top of the Anna KVS. Droplet enables users to execute compositions of functions at low latency, and the system builds on top of Anna in order to enable stateful computation. Droplet is co-deployed with the Anna caching system to achieve low-latency access to shared state, and the system relies on Anna's lattice data structures to resolve conflicting updates to shared state.

You can install Droplet's dependencies with pip and use the bash scripts included in this repository to run the system locally. You can find the Droplet client in droplet/client/client.py. Full documentation on starting a cluster in local mode can be found here; documentation for the Droplet client can be found here. An example interaction is modeled below.

$ pip install -r requirements.txt
$ ./scripts/start-droplet-local.sh
...
$ ./scripts/stop-droplet-local.sh

The DropletConnection is the main client interface; when running in local mode, all interaction between the client and server happens on localhost. Users can register functions and execute them. The executions return DropletFutures, which can be retrieved asynchronously via the get method. Users can also register DAGs (directed, acylic graphs) of functions, where results from one function will be passed to downstream functions.

>>> from droplet.client.client import DropletConnection
>>> local_cloud = DropletConnection('127.0.0.1', '127.0.0.1', local=True)
>>> cloud_sq = local_cloud.register(lambda _, x: x * x, 'square')
>>> cloud_sq(2).get
4
>>> local_cloud.register_dag('dag', ['square'], square)
>>> local_cloud.call_dag('dag', { 'square': [2] }).get()
4

To run Anna and Droplet in cluster mode, you will need to use the cluster management setup, which can be found in the hydro-project/cluster repo. Instructions on how to use the cluster management tools can be found in that repo.

The Hydro Project is licensed under the Apache v2 License.