Sprinkle

Run jobs on distributed machines easily. No master negotiation or consensus in sight: All parts take an addr command line argument that refers to a UDP multicast address on which workers can be discovered.

Run some workers on any machines you have lying around¹ (linux servers, raspberry pis) within a network. Start an instance of the ui to monitor them, and use the run command to run stuff.

Commands

Worker

Responsible for doing work. Exposes a gRPC service definition defined in api/proto/sprinkle.proto.

Run

User-facing command line for running work on the most appropriate worker.

UI

Simple HTTP server to monitor the workers and running jobs.

Example command lines

Start the UI

$ ./bin/ui --logtostderr

Start a worker

$ ./bin/worker --logtostderr

Run a command

$ ./bin/run --cmd="sleep 10 && echo hello" --logtostderr 
...
hello

TODO

• take a reference to a command and use groupcache
• test if it's possible to run the UI on a worker!
• TTL on discovery requests.
• unit tests

Footnotes

1. not OSX.. multicast doesn't work on OSX for some reason. ↩