Skip to content
GitHub

HotLoop products: eight on five release lines

HotLoop is eight products for the plant floor: four that grow with a site (IoT, Edge, Gateway, and Edge Relay), plus Flow for wiring, Cascade for workflows, HotLoop for CODESYS for the PLC’s own screen, and Transmitter for putting a PLC’s values on the wire.

They ship on five release lines. HotLoop core is IoT, Edge, Gateway, and Edge Relay: four products, one codebase, one tag. Every core release ships all four in lockstep, same version, each with its own image, its own Helm chart and its own Podman Quadlet units. Pick the one that fits the box. Every product speaks every protocol HotLoop has, so you never step up a product just to read the PLC on your bench. Features split them. Protocols don’t.

Flow, Cascade, HotLoop for CODESYS, and Transmitter are their own lines with their own versions, so a Flow release never moves the core, and the other way around. Everything below the table is about the core. Here is where each of the eight starts:

Product Release line Docs Install Release notes Roadmap
HotLoop IoT HotLoop core Docs Install Notes Roadmap
HotLoop Edge HotLoop core Docs Install Notes Roadmap
HotLoop Gateway HotLoop core Docs Install Notes Roadmap
HotLoop Edge Relay HotLoop core Docs Install Notes Roadmap
HotLoop Flow Flow Docs Install Notes Roadmap
HotLoop Cascade Cascade Docs Install Notes Roadmap
HotLoop for CODESYS HotLoop for CODESYS Docs Install Notes Roadmap
HotLoop Transmitter Transmitter Docs Install Notes Roadmap

The table that follows compares the four core products.

Details HotLoop IoT HotLoop Edge HotLoop Gateway HotLoop Edge Relay
What it’s for The automation base. A Pi, a NUC, a box in a closet A box at the machine, with the uplink allowed to die The site brain Poll, forward, survive. No database, no UI
Entities, automations, helpers, scripts, recipes, blueprints, dashboards, logbook, notifications, MCP, MQTT with discovery, ISA-18.2 alarms, backups, every driver yes yes yes drivers only
OPC UA server, Sparkplug edge publishing with store and forward yes yes publishing only
Fleet, multi-site, scheduled reports yes
Image ghcr.io/hotloop-io/hotloop-iot ghcr.io/hotloop-io/hotloop-edge ghcr.io/hotloop-io/hotloop ghcr.io/hotloop-io/hotloop-edge-relay
Chart hotloop-iot hotloop-edge hotloop hotloop-edge-relay
Quadlet units deploy/quadlet/iot/ deploy/quadlet/edge/ deploy/quadlet/gateway/ deploy/quadlet/edge-relay.container
Quadlet units on SQLite deploy/quadlet/iot/sqlite/ deploy/quadlet/edge/sqlite/ deploy/quadlet/gateway/sqlite/ none
database.type in the chart postgresql (default), sqlite timescaledb (default), postgresql, sqlite timescaledb (default), postgresql, sqlite none
Architectures amd64, arm64, arm/v7 amd64, arm64, arm/v7 amd64, arm64 amd64, arm64, arm/v7
Database the chart and Quadlet run PostgreSQL 17 TimescaleDB TimescaleDB none
History kept by default 30 days 90 days 90 days none
Databases the binary runs on SQLite or Postgres SQLite, Postgres or TimescaleDB SQLite, Postgres or TimescaleDB none

The database is a setting, not an image: HOTLOOP_DB_TYPE=sqlite and HOTLOOP_DB_PATH put any of the three on a SQLite file with no database container at all (DEPLOYMENT.md, “SQLite”). The charts take the same choice as database.type and the Quadlet units as the sqlite/ directories, each product offering only the databases in the table above: asking the IoT chart for timescaledb fails the install. Moving from one database to another is a backup and a restore (BACKUP.md).

The Gateway kept the names it shipped under through 4.17.0: image ghcr.io/hotloop-io/hotloop, chart hotloop. Nobody running it has to change a thing.

What’s in a product is what’s in the binary

Section titled “What’s in a product is what’s in the binary”

A product is a build, not a setting. -tags hotloop_iot builds IoT, -tags hotloop_edge builds Edge, no tag builds the Gateway. What a product doesn’t have isn’t switched off, it isn’t there: the OPC UA server isn’t in the IoT image at all, so no environment variable, no values file and no bad day can turn it on.

So a smaller product answers for what it doesn’t have instead of pretending:

  • The API keeps every route in every product. A route for a feature this product doesn’t have answers 404 and names the product that does: {"error":"scheduled reports are part of HotLoop Gateway; this is HotLoop IoT","product":"iot","feature":"reports"}. GET /api/status says product and features, so a script asks instead of probing.
  • A Gateway-only setting on IoT or Edge (HOTLOOP_REPORTS_ENABLED=true carried down in an old values file, say) warns at startup and on the Settings screen, and HotLoop runs. HotLoop keeps the plant running over a report it cannot produce, and the warning makes sure somebody sees it.
  • The charts don’t have the setting at all. charts/hotloop-iot has no reports, opcuaServer or edgePublish block, and hotloop-edge has no reports. Set one anyway and it renders nothing.
  • hotloop --product prints iot, edge or gateway. --version still prints only the version, because scripts parse it.
  • A Pi, smart plugs, a few sensors, an MQTT broker full of Tasmota and Zigbee2MQTT, or one PLC on a bench: IoT. It’s the smallest image (about 29 MB on arm) and plain PostgreSQL’s image is a sixth of TimescaleDB’s, which on an SD card is the number that matters.
  • A box bolted next to a line, feeding SCADA over OPC UA, or publishing upstream over Sparkplug and riding out the uplink going down: Edge.
  • The site: every line, reports mailed on a schedule, edge relays reporting in, more than one site on one screen: Gateway.
  • A box that should poll and forward and nothing else, with no database to look after: Edge Relay. See Edge Relay reference.

The image is the binary plus about 3 MB of base.

Details amd64 image / binary arm64 image / binary arm/v7 image / binary
IoT 31.2 / 28.0 MB 29.1 / 26.0 MB 29.2 / 26.0 MB
Edge 36.5 / 33.3 MB 34.2 / 31.0 MB 34.3 / 31.1 MB
Gateway 37.1 / 33.9 MB 34.8 / 31.6 MB not built
Edge Relay 15.7 / 12.5 MB 14.9 / 11.7 MB 15.1 / 11.9 MB

Those are on disk. Running with no devices, HotLoop sits around 10 MB of RSS and PostgreSQL around 12 MB, and that’s the floor, not the number: what a product uses with devices polling depends on the devices.

One repository, all four charts:

Terminal window
helm repo add hotloop https://hotloop.io/hotloop/
helm install hotloop hotloop/hotloop-iot --namespace hotloop --create-namespace
helm install hotloop hotloop/hotloop-edge --namespace hotloop --create-namespace
helm install hotloop hotloop/hotloop --namespace hotloop --create-namespace

Call the release hotloop whichever product it is. The release name is what the database’s volume, the Secrets and the Service are named after, and keeping it is what lets a site move up a product later (below).

On EmberNet, each one is its own card in the App Store: “HotLoop IoT”, “HotLoop Edge”, “HotLoop Gateway” and “HotLoop Edge Relay”. A store deploy folds the dashboard’s tenant labels into everything the chart renders, and Running Apps shows them as HotLoop-IoT, HotLoop-Edge, HotLoop-Gateway and HotLoop-Edge-Relay.

For a box that isn’t in a cluster. Each product directory under deploy/quadlet/ has five units: hotloop.network, hotloop.container (the app), hotloop-db.container (its database, the same image and version its chart runs), hotloop-db.volume and hotloop-data.volume. The install steps are in the header of each hotloop.container, and Podman Quadlet has the rest.

These need Podman 5.0 or newer. Raspberry Pi OS Bookworm ships 4.3, which has no Quadlet at all, so a Pi wants Raspberry Pi OS Trixie or newer. The app follows :stable and Podman’s auto-update moves it, with the database pinned: the Quadlet README has how and why. Whichever way it’s installed, put the database on a USB SSD and not the SD card. History writes all day, every day, and SD cards die of exactly that.

IoT to Edge to the Gateway keeps every device, tag, entity, automation, alarm, user and hour of history. Every product runs every migration, so the database is the same shape in all three, and the bigger product starts with its extra features empty.

  • Quadlet: the unit names, the volumes and the secrets are the same in every product directory. Copy the bigger product’s hotloop.container and hotloop-db.container over yours, systemctl daemon-reload, restart.
  • Helm: helm upgrade hotloop hotloop/hotloop-edge (or hotloop/hotloop) on the same release. The three charts select their pods with the same labels and name the database’s StatefulSet and volume the same, so the upgrade is allowed and the volume stays bound.

Take a backup first (POST /api/backup, see Backups and restore) and copy it off the box. The move has been run on Quadlet, IoT then Edge then the Gateway on one database volume with the same device there at every step, with the IoT database’s PostgreSQL directory starting under the Gateway’s TimescaleDB image, and as a helm upgrade on a real cluster. Where IoT’s plain directory has no TimescaleDB extension in it, history stays in native partitions after that move, and a hypertable takes a backup and a restore into a fresh TimescaleDB. Going back down a product is not a move to make on a box you care about.

One v tag releases all four, so IoT, Edge, the Gateway and the Edge Relay always share a version. Before anything is built, every chart’s version and image is checked against the tag, every Quadlet app unit is checked to follow :stable with a healthcheck, and every database unit is checked to match its chart and never be auto-updated. The IoT and Edge charts are generated from the Gateway chart, so a fix to one reaches all three.