Local Metrics

This was an experiment to run metrics in a distributed way: local collection and access, with optional aggregation through a hosted backend.

Layout

We have constructed this - for now - as an umbrella application with the following apps:

• apps/lm_agent_web: the agent application, which users install. It is fully stand-alone.
• apps/lm_backend_web: "our" backend, which agents can optionally connect to for data aggregation, replication coordination, and so on.
• apps/lm_agent: agent-specific code (local storage, sending to backend, etc)
• apps/lm_backend: backend-specific code (PostgreSQL storage, etc)
• apps/lm_web: common code for front-ends (Phoenix deps, etc).
• apps/lm_common: code common to both.

Modelled as a class diagram we get the following dependencies:

classDiagram
    LmAgentWeb <-- LmWeb
    LmAgentWeb <-- LmAgent
    LmBackendWeb <-- LmWeb
    LmBackendWeb <-- LmBackend
    LmAgent <-- LmCommon
    LmBackend <-- LmCommon
    LmWeb <-- LmCommon

We build deployables for lm_agent_web and lm_backend_web. The included Dockerfile will build the latter by default but --build-arg SUBSYSTEM=agent will build the agent container.