HotLoop core 4.19.11 release notes
The HMI dashboard cards and the CODESYS work that landed after 4.19.8, two fixes found in earlier releases. There are no migrations, and the one API change is a new route (below). A plain helm upgrade or the Quadlet auto-update takes it, and the previous release runs on the same database.
This build is amd64 only, and the next release is one small version bump that builds the arm images, the way the two runs go. stable keeps the arm images it holds in the meantime. 4.19.10 is the last multi-arch build, and 4.19.8 lists everything that changed since the one before it.
What changed
Section titled “What changed”-
A setpoint card. A dashboard card for a number somebody types: a temperature, a speed, a level. It knows the tag’s unit and limits (a card can only narrow the tag’s range, never widen it), says under the field what is wrong with an entry and keeps Set off until it is right, asks first with the current value and a reason for the audit trail, and writes through
POST /api/write, so the master write switch, the tag’s writable flag, the role’s permission and the audit row are the same as for every other write. It then says what the device holds: “Read back 70 C from the device: it is the value sent”, or “The device reports 60 C, not 70 C” if the PLC took the write and holds something else. It watches the reading for two seconds after it matches, because a PLC that brings a written value back to its own limit shows the value for an instant first. While the tag has no good reading the card says the device is not answering and turns its field and Set off. See DASHBOARDS.md. -
A symbol page card. A picture of a machine: pumps, motors, fans, valves, tanks, heaters, lamps, pipes, readouts and labels placed by pixel on a background and drawn from the state of their tags, in words as well as color, red or warm in an alarm, faded when a tag has no good reading, and still when the person asked their system for less motion. HotLoop draws every symbol. It commands nothing: a setpoint or button card beside it does. See DASHBOARDS.md.
-
A write to a CODESYS or OPC UA tag is read back from the PLC right after it goes in. A subscription sends only changes, so a recipe that wrote what the PLC already held was never confirmed. The runtime now reads the point itself after a write that went in, and the setpoint card’s “read back” is that reading.
-
A CODESYS device is read once a second, not every five. That is how often the driver asks the runtime’s OPC UA server to look at each variable; the runtime’s own default of five seconds showed a pump starting five seconds late. Change the device’s poll interval or its
samplingIntervalMsto ask for more. See INTEGRATIONS.md. -
Alarms import from a CODESYS project.
POST /api/alarms/import/codesys(permissionalarms.write) reads the file the CODESYS IDE’s Export Alarms writes, UTF-16 as the IDE wrote it or UTF-8, with?deviceId=and optionally?application=and?defaultPriority=. Without?apply=truenothing is saved and the answer says what would come across. Digital alarms, UpperLimit and LowerLimit with their messages and delays carry over; anything else the IDE can write is not read and comes back inskippedwith its row and why, because CODESYS does not document the layout and only what a real export has shown is read. See API.md. -
The old screens’ copy has no dashes in its sentences. A plant reads every one.
-
The UniFi screens draw a trace before the number beside it. The old screen’s Overview, and a port’s detail on both screens, read the number first and then the recent readings behind it. A reading that landed between the two sat in the trace and not in the number, so a WAN card said 5 ms beside a line that ended at 6. They ask for the traces first now, with the traces of the last look, so a refresh makes no request it did not make before; the first look of a screen with no WAN yet reads them after. Proved in a browser against both screens: the old order fails the new test (5 ms beside a trace ending at 6, a port’s traffic one count behind its trace), the new order passes, and putting the old order back fails it again.
-
A Kafka or Redpanda history target says why a broker refuses writes. A broker with no room on its disk (Redpanda below its free-space threshold, Kafka with a log directory it cannot write) answers every write with an error code, and the client retries behind it until the record times out, so the target showed only “records have timed out before they were able to be produced”. After a timeout the target now asks the broker once, with a write that carries no records and so cannot store a reading, and shows what it answers:
KAFKA_STORAGE_ERROR, with a line that it is the disk under the broker. A working broker answers that request with an error that means nothing, and the target says nothing extra. The readings were never lost and still are not: the target keeps the batch, owes the span and sends it from HotLoop’s own history when the broker takes writes again. Proved against Redpanda with its free-space threshold set above its disk (the target is down with the reason, and the client’s timeout is still behind it), and against a working Redpanda and Apache Kafka (no reason given, and nothing written by the question). This closes the one item 4.19.8’s notes list as known.
- Nothing new. What is still open under Broken or missing in the roadmap: OPC UA X.509 user tokens, the deprecated
Basic128Rsa15andBasic256policies our OPC UA server offers and passwords over 64 bytes (both the library’s), and S7comm and EtherNet/IP silent-device tests against an in-process server.