Pantavisor Mocker v0.2.0: simulate whole device fleets as swarms — locally, on Docker, or on Kubernetes

News
1

When we introduced Pantavisor Mocker it simulated one device at a time. The last two releases (v0.1.0 and v0.2.0) turn it into a proper fleet-simulation tool: describe a whole fleet in one JSON file and run it as a swarm — in local tmux sessions, as Docker containers, or as Kubernetes pods.

One file per device, one file per fleet

Everything is now driven by single-file JSON configs. A device.json describes one device — Pantahub endpoint, auto-join token, device metadata, automation behavior, sync intervals:

{
  "pantahub": { "host": "api.pantahub.com", "port": "443", "autojoin_token": "YOUR_AUTO_TOKEN" },
  "device-meta": { "pantavisor.dtmodel": "My Test Device" },
  "automation": { "enabled": true }
}

pantavisor-mocker start -s my-device -c device.json

Re-applying the config is always safe: the registration credentials the device obtains on first run are never touched, so the device keeps its identity across restarts.

A swarm.json does the same for a whole fleet — endpoint, token, metadata templates, model/channel definitions, generation counts and automation weights:

pantavisor-mocker swarm init -d my-fleet   # writes a swarm.json template
vim my-fleet/swarm.json                    # set pantahub.host + autojoin_token
pantavisor-mocker swarm run -d my-fleet    # generate the fleet + simulate it

swarm generate-devices, swarm status, swarm simulate --auto and swarm clean are there for step-by-step control, replacing the old pvmocks bash script with native subcommands.

Container-native swarms: one device per container

The v0.2.0 headline is the new swarm device entrypoint: each container provisions one device from a shared swarm.json on first start (picking a model, and a random channel with --channel random) and then runs it in the foreground. One device = one container = one log stream — and the fleet size is just the replica count.

With the ready-made compose file in examples/docker/:

services:
  device:
    image: ghcr.io/pantavisor/pantavisor-mocker:latest
    command: ["swarm", "device", "-c", "/swarm.json", "--channel", "random"]
    volumes:
      - ./swarm.json:/swarm.json:ro
    restart: unless-stopped
    deploy:
      replicas: 5

docker compose up -d --scale device=10
docker compose logs -f          # all devices
docker logs -f docker-device-3  # one device
docker compose down             # add -v to also discard the device identities

Each container keeps its device state in a per-container volume, so identities survive restarts and are dropped with the container.

On Kubernetes

examples/kubernetes/ ships two ready-to-apply manifests, both driven by the same swarm.json as a ConfigMap:

  • swarm-devices.yaml (recommended) — a StatefulSet where every pod runs swarm device and keeps its identity on its own PersistentVolumeClaim, so reschedules resume the same registered devices. Per-device kubectl logs, and scaling the fleet is one command:

    kubectl apply -f examples/kubernetes/swarm-devices.yaml
kubectl scale statefulset pantavisor-mocker-device --replicas=20
kubectl logs -f pantavisor-mocker-device-3

  • swarm.yaml — the whole fleet in a single pod via swarm run --headless, for the smallest footprint (hundreds of devices in one container, per-device output in tmux sessions inside it).

Also new along the way

  • Automation mode — devices answer update invitations and acceptance tests on their own (weights configurable in swarm.json), which is what makes TTY-less container swarms possible in the first place.
  • HTTPS auto-off for private endpoints — pointing mocks at localhost or a private address automatically drops TLS, so a swarm against a local self-hosted Hub just works.
  • pvcontrol server — a Unix-socket implementation of the Pantavisor Control API with local-revisions support, so container-side tooling can talk to the mock like a real Pantavisor.

Get it

Docker image: ghcr.io/pantavisor/pantavisor-mocker:latest (also tagged per release). Source, examples and the full README: github.com/pantavisor/pantavisor-mocker. It pairs nicely with running your own dev Hub — spin up the cluster from the self-hosting guide, mint an auto-join token, and point a 20-device swarm at it.

Questions and feedback welcome below!