Skip to content
GitHub

HotLoop Flow 2.16.4 release notes

HotLoop Flow 2.16.4 makes a big flow hold less memory, and brings Node-RED’s own example flows closer to what Node-RED does. Nothing in your flows, settings or chart changes.

  • A node that never takes a message holds no inbox. Every node’s inbox was a queue of 1,024 slots made when the node started, 24 KB each, so a flow of 12,700 nodes held about 300 MB of inboxes that the garbage collector scans on every cycle. On a plant running 12,753 nodes, 6,306 of them had never received a message. The inbox is now made when the first message is offered to the node. An idle node holds 2.5 KB instead of 29.7 KB. Order, capacity, the overflow policies and what happens on a redeploy are unchanged.
  • Node-RED’s example flows give what Node-RED gives. The 71 example flows Node-RED ships were run in Node-RED 5.0.10 and in Flow, and what every Debug printed was compared: 27 matched before and 52 do now, and 39 that do not depend on a clock or the machine are held in a test. Debug with complete set to false, Inject with a bare payload row, CSV with no columns, Batch in concatenate mode with a repeated topic, and Switch recreating sequences are fixed. The editor’s Import dialog is driven in a browser with one of the flows.
  • FlowLab reads a LabVIEW VI as far as the file lets it. A VI is a binary that NI does not document, so FlowLab reads the container, the description, the version, the front panel names and the VIs the diagram calls. The Import dialog lists what it cannot read, the wires and the primitives, before anything is added.
  • The soak test can fill inboxes. HOTLOOP_FLOW_SOAK_SLOW=n adds n chains of an inject in front of a slow node to the plant-shaped soak, which is how a slow Modbus read backs a queue up. A test measures what a queued message holds in memory (489 bytes in that flow).

An idle node, measured on one machine: 29,734 bytes before, 2,502 after. On the plant’s flow that is about 6,300 nodes and about 150 MB less live heap.

On a synthetic plant flow with 184 injects in front of slow nodes, so that inboxes fill the way slow Modbus reads fill them, and a memory limit close to the live heap the way the plant’s pod is, 2.16.3 climbed from 0.4 cores to 0.9 cores by 277,000 queued messages and 1.3 by 360,000. 2.16.4, with the same flow and limit, held 0.53 cores at 277,000 queued and 0.52 at 305,000, where 2.16.3 was at 0.91 and 0.85.

None.

A plain helm upgrade to the 2.16.4 image and chart, or the new tag in your Quadlet unit. Nothing happens on first start and nothing needs migrating. A flow file loads and saves byte for byte as before. This release builds the amd64 image. The 2.16 and 2 tags keep the arm64 image they already hold from 2.16.1, and arm64 comes back in 2.16.5. If you are on arm64, stay on 2.16.1 until then.

Image: ghcr.io/hotloop-io/hotloop-flow:2.16.4, linux/amd64. Chart: https://hotloop.io/hotloop-flow/. The README at this tag has the rest, with pictures: https://github.com/HotLoop-io/hotloop-flow/blob/v2.16.4/README.md