Logman is a simple elixir app that leverages distributed elixir to log messages across multiple nodes. This implementation uses docker-compose to simulate nodes running on different hosts, with 3 containers running simultaneously, each with a Logman instance. Each instance exposes an http interface with a single /event endpoint for incoming messages.
Make sure you have elixir, docker-compose and their respective dependencies installed on your host machine.
./run.sh
NOTE: This script will attempt to build the image each time it's run, which can take a little time. If the image is already built on your machine, see the Step by step setup section below for some one-off commands to get things up and running faster.
If containers exit with non-zero status code on first try, run command again. See note below in Run containers section for why sometimes containers can fail on first startup.
docker build -t logman .
mix deps.get && mix deps.compile && mix compile
NOTE: with this setup, you need to install and compile deps on host machine. Some of the reasons for needing to do this in development environment when running elixir in docker are outlined in this article.
docker-compose up
NOTE: on first time running docker-compose up, you may see containers exit with a non-zero status code. I didn't have time to debug this thoroughly, but I believe it happens because of non-deterministic behavior of files in _build directory that can occur when all 3 containers are trying to initialize simultaneously for the first time (e.g. at least once the stack trace showed File.Error could not write to file "/app/_build/dev/lib/plug_cowboy/ebin/Elixir.Plug.Cowboy.beam": no such file or directory). The issue was always fixed for me by stopping containers (ctrl + c) and restarting (docker-compose up).
Logman uses cowboy and plug to service http requests. Each node/container is run on its own port. The url for the logging endpoint for each container is as follows:
node_1: http://localhost:5555/event
node_2: http://localhost:5556/event
node_3: http://localhost:5557/event
To make a request to node_1:
curl -X POST --data "some message" http://localhost:5555/event
Upon receiving the request, node_1 will log the message "locally" (in a subdirectory specified in Logman.Handler.get_log_file_path, which will be distinct for each node), and also send message to all other nodes (node_2 and node_3 in this case) to be logged, using distributed elixir and Task api for communication across nodes.
The same applies for requests to other nodes (e.g. a request to node_2 endpoint will log message "locally" on node_2 and also send message to node_1 and node_3 to be logged).
To run tests within container:
./test.sh
Testing is less robust than I would like. Currently there is no testing for sending messages to other nodes, mostly because of challenges in setting up testing with multiple nodes. I tried using LocalCluster lib, which is designed to simplify creation of nodes for testing. However things get complicated when each node is running an http server and you get address in use errors. I didn't have time to delve into this more deeply, though I believe there are solutions.
The approach here uses docker-compose to run each node as a container, though I'd like to explore running multiple nodes in a single container. I didn't take this approach here because 1) I wanted to simulate running nodes across multiple hosts and 2) it didn't feel like idiomatic usage of docker.