Architecture

graph TD
    subgraph server ["Server"]
        web([Web UI & API])
        scheduler(["Scheduler"])
        publisher(["Publisher"])
        coordinator([Coordinator])
        db[("Database")]

        web <--> db
        %% db --> publisher breaks the layout; a labeled link is used instead.
        publisher-- Query ---db
        scheduler <--> db
        publisher <--ZMQ--> scheduler
        web --ZMQ--> publisher
    end

    subgraph lab [Lab]
        worker([Worker])
        dut1([DUT<br>Linux])
        dut2([DUT<br>IoT])
        dut3([DUT<br>AOSP])

        worker <--> dut1
        worker <--> dut2
        worker <--> dut3
    end

    publisher --WS--> worker
    web <--HTTP--> worker
    coordinator <--TCP--> worker

LAVA consists of several components with different roles, organized in a server-worker model.

The simplest possible configuration is to run the server and worker components on a single machine. Larger instances can be configured with multiple workers controlling a greater number of attached test devices.

Server

Web interface

The LAVA web interface is built using the apache2 web server, the gunicorn application server and the Django web framework. It also provides XML-RPC access and the REST API.

Danger

In production, do not serve the web application or the publisher over plain HTTP. Put them behind a reverse proxy that terminates TLS.

Publisher

The lava-publisher receives events from LAVA services and forwards them to subscribers.

Scheduler

The lava-scheduler periodically schedules jobs in the queue to idle devices.

Coordinator

The lava-coordinator daemon receives, stores, and dispatches messages for MultiNode jobs.

Worker

The lava-worker daemon pulls scheduled jobs for this worker from the server over HTTP. For each job it starts a lava-run process to manage all the operations on the DUT. It also keeps a WebSocket connection to the publisher for real-time events, reducing latency when jobs are assigned or updated.

DUT

Physical or virtual devices under test. Multiple DUTs can be attached to a single worker.