Changelog
This document provides a high-level overview of customer-facing features, improvements, and critical fixes introduced in each version.
[3.44.1]โ
The release where the Agent comes on by default, and where certificates and GDS enrolment do what the Certificates page says.
๐งช Ask your plant, now on by default (preview)โ
In 3.43.0 the Agent, the AI Provider and the sidebar chat stayed hidden until you changed a setting. They now show in every editor. They still do nothing until an AI Provider node has an API key, and the Agent cannot deploy.
- On by default, and easy to switch off. Set
opcUaAgentEnabled: falseinsettings.js, orOPCUA_FOR_NODE_RED_ENABLE_AGENT=falsein the environment.settings.jswins over the environment. The test-tube emoji is gone from the node names. - Follow a value over time. Tick Watch in the sidebar, then ask about a process in motion, such as tell me when the boiler temperature goes over 90. The Agent waits for the change, up to 60 seconds, instead of reading the value again and again. Watch is off by default, because the question stays open while it waits.
- Code is flagged before you deploy it. When a flow that the sidebar prepares holds a Function node, a line under Place on canvas names it, so you read it first. The Agent does not prepare a Function node that reads the environment, loads a module or uses the network.
- Read-only stays read-only. The Agent node uses the read tools, or read and watch, whatever profile a flow file gives it.
- The AI Provider remembers its preset and lists the models. Open the node again and Preset still shows the service you chose. A button on the Model field lists the models that your provider offers, taken from its own API with your key. If the provider does not answer, you get the usual models of the preset and the reason.
๐ Certificates and GDS enrolmentโ
Trusting a CA together with its revocation list, and enrolling with a Global Discovery Server, now work from the Certificates page from start to end.
- Trust a CA and its CRL in one place. The Trusted tab takes the CRL next to the CA, and marks a CA as such. The Trusted and Issuers tabs list the CRLs with their issuer, the CA that signed them, the expiry date and the number of revoked certificates, and you can delete one. The Health tab warns about an expired CRL and about files in a store that are not certificates.
- GDS enrolment keeps what you trusted by hand. The trust list sync moved every trusted certificate that the GDS did not list to Rejected, so enrolling could break connections that worked. It now adds what the GDS lists and leaves the rest. A certificate that the GDS revokes is still refused, through its CRL. A certificate that the GDS only removes from its list stays trusted until you reject it.
- GDS enrolment works without an OPC UA Server node. It failed with
discoveryUrls is empty. Node-RED now registers as a client, or as a client and server at the address of your running OPC UA Server node. - A failed enrolment tells you why. The GDS page showed only that enrolment had failed. It now gives the reason in plain words, for example that the GDS does not trust this certificate, with its thumbprint and what to do.
๐ Bug Fixesโ
- The feedback icon no longer covers the form. It sat in a large box over the node dialog and hid controls such as Manage Certificates. It is now a small icon in the bottom bar of the dialog.
- Trust, Reject and Delete stay visible. A certificate with a long name pushed them out of the table on the Certificates page, so a rejected certificate could not be trusted or deleted. Long names now wrap.
- A CRL in PEM format installs. The page said "CRL uploaded" and installed nothing unless the file was in DER format. Both formats now install, from a
.pemor a.crlfile. - A failed upload says why, in red. Failed uploads looked like successes. A CRL uploaded as a certificate, or the reverse, is now refused with a clear message, instead of landing in the trust store.
- The thumbprint of a certificate that comes with its CA is right. The page showed a hash of the whole file, which matched no other tool, and Delete in the Trusted tab failed for such a certificate.
- The AI Provider keeps the model you choose. Opening a provider with a preset, such as Gemini, replaced your model with the default of the preset, and the next Update saved it.
- Flows from the Agent sidebar use your endpoint. A flow could carry an endpoint of its own, with a made-up name and address, and Place on canvas added it beside yours. Every node now goes on the endpoint that you chose in the sidebar.
- Max retry is documented as it works. The help and the endpoint dialog said that Max retry did not apply to the first connect yet. It does: it limits the attempts of the first connect.
โ ๏ธ Upgrade notesโ
- The Agent nodes and the sidebar chat now show in every editor. To keep them hidden, set
opcUaAgentEnabled: falseinsettings.js. - Certificates that an earlier GDS sync moved to Rejected stay there. Open the Rejected tab and use Trust for the ones you need.
- 3.44.0 was withdrawn shortly after it was published. If you installed it, update to 3.44.1.
[3.43.0]โ
The release where a running flow can change its OPC UA connection, tell you when the connection changes, and start to answer questions about your plant.
๐ Change the connection of a running flowโ
A flow used to be tied to the server address, the security settings and the login that you deployed. To move a packaging line from Line A to Line B, or to point a flow at a spare server, you edited the endpoint and deployed again. Two new nodes now do this from a message, and the endpoint has a new setting for when the connection opens.
- Move a running flow to another server with a message. Send
connect,disconnect,reset,get,validateorstatusto the new Endpoint Control node, from a dashboard button, a Function node or any other source. Every Read, Write, Call and Monitor node of that endpoint uses the new settings at once. You do not deploy again, and a Monitor subscribes again without a new message. The message can set the server address, the security mode and policy, the login (user name and password, or a certificate and key as PEM text), the retry delays, the namespace aliases and some parameters of the subscriptions. - A wrong change leaves the running session alone. The node checks the new settings first. If something is wrong, or the new server does not answer within 20 seconds, the previous settings come back. The error messages go to the second output, one for each field.
validateis a dry run, andifRevisionstops two operators from replacing each other's change. - Secrets stay in memory. A password, a passphrase or a private key arrives in the message and appears in no output, log or status. A list of Allowed servers limits where a message can send the endpoint. Note that, as for any endpoint, the client accepts the certificate of a server the first time it connects, so a message that points to a new server makes that certificate trusted. The Certificate Manager stays the place for trust decisions.
- A basic edit dialog. Choose the Action in one field, then fill in the Server and the Login. A sentence under the Action says what the node does, and Show the message this node sends gives you the equivalent message to copy. A node without a name shows its action and its endpoint, such as
connect ยท Plant 1. - Professional licence for Connect and Reset.
get,validate,statusanddisconnectwork with every licence, so an expired licence never leaves an endpoint stuck. - Auto Connect on the endpoint. It is on by default, and your existing flows keep their behaviour. Clear it to keep the endpoint closed until an Endpoint Control node sends
connectorreset. Until then, each message to a client node gets a node error, and the editor can still browse the server. After a deploy or a restart the endpoint is closed again: sendresetfrom an Inject node that fires once at start. - Retry settings on the endpoint. Under Advanced parameters, Initial delay (1000 ms) and Max delay (10000 ms) rule the reconnections after a lost connection. Max retry (2) is stored but does not apply to the first connect yet. A value out of bounds in a flow file is replaced by the default, with a warning at start.
- Node ids with an alias find their node on another server.
na=plant;s=Setpointresolves on a server that numbers its namespaces in another order. - Two example flows. A packaging line that changes between Line A and Line B from dashboard buttons, and a Dashboard 2 form where an operator types a server address, a security mode and a login and moves a running Read flow with one press on Submit. A new documentation page, Two certificates, two jobs, explains why the application certificate belongs to the Node-RED instance and the user certificate is the login.
๐ก Follow the connectionโ
The connection can be opened, lost or closed by any node, and a dashboard used to learn it only from the node that failed.
- One message for every change of the connection. The new Endpoint Events node reports connecting, connected, reconnecting, failed and disconnected, with no input message. It also warns before the user certificate expires. Tick the events you want in the dialog;
retry, one message for each new attempt, is off by default. It works with every licence, and several nodes can follow one endpoint. - A switch tells both sessions. When a
connectorresetreplaces an open session, the events aredisconnected(with the id of the old session),connectingandconnected(with the id of the new one). - Calm on a bad link. At most 10 messages in 10 seconds for each node, and the last state is never lost.
- The name of the connection in every message. The outputs of Endpoint Control and Endpoint Events carry the endpoint name in
msg.connectionName, to route on with a Switch node, and inmsg.payload.name, to show.
๐๏ธ Choose the subscription from the messageโ
- Monitor and MonitorEvent follow
msg.subscription. Set it to the id (s2) or the name (Fast) of a subscription of the endpoint, and the monitored items move to it. The node subscribes again on it after a reconnection. Without the property, the dialog decides, as before. A name that no subscription has, or that two have, gives a node error that lists the choices. - Monitor and MonitorEvent follow a change of the connection. After a switch by Endpoint Control, they subscribe again on the new session without a new message.
๐งช Ask your plant a question (preview)โ
Finding the cause of an alarm means browsing, reading and comparing nodes by hand. In this preview an AI model does the looking, with read-only access to your server through an endpoint you already have.
- The OPC UA Agent node answers a prompt. Send a question, such as why an alarm is active, and the node returns the answer, a trace of the tool calls and progress messages. Send
msg.cancelto stop a run. The new AI Provider node holds the settings for Anthropic or an OpenAI-compatible service, and the API key stays in the credentials store or an environment variable. A Gemini preset is included. - A chat in the editor sidebar. While you build a flow, ask the plant a question and get the answer as Markdown with diagrams. A click on a NodeId copies it, and the nodes you selected go along with the question. Continue and Dig deeper extend an answer. Paths in answers use the namespace aliases of your endpoint, so you can paste them into node settings.
- An example chat page for Dashboard 2, and an alarm triage example.
- Off by default. The nodes stay out of the palette until you set
opcUaAgentEnabled: trueinsettings.js. A deployed Agent node showspreview: offand says how to enable it. - Limits. The Agent needs a Professional licence for unlimited use. Without it, a small allowance of prompts per day applies, shared by Agent nodes, the sidebar and dashboard flows; it comes back at 00:00 UTC. A run that the model never answered, such as a wrong API key, gives its prompt back. The At once field of the Agent node (default 1) limits the runs in progress, and a prompt over the limit gets a node error with
msg.agentBusyset, so a Catch node can drop, log or retry it.
๐ Bug Fixesโ
- Transport settings of the endpoint apply. The values in the endpoint dialog were not passed to the connection, and the editor read them from the wrong fields. They now reach the client.
- The user certificate can be entered. The Certificate login of the endpoint showed a label but no input. A stored secret for an endpoint that is not deployed yet is kept for the certificate too.
- A user name or key that cannot work is reported. An empty user name, or an unreadable or PKCS#1 private key, is now checked before the connection, with a clear message.
- A missing endpoint is reported. A client node whose endpoint was deleted used to do nothing. It now shows a failed status and raises an error. A second node that fails with the same text is no longer silent.
- A refused licence check completes the message. A client node whose licence check failed left the message pending and raised an unhandled error. It now reports the error on the node.
- The user PKI folders of the OPC UA Server are in the right place. The folder was created relative to the working directory instead of beside the server PKI, and the start-up banner showed the trusted and rejected user folders the wrong way round.
- A node picked in the tree always works. Picking the Objects, Root, Types or Views folder wrote a path such as
Root/Objects, and the node then failed at each message withUnrecognised address at position 0. The editor now writes the NodeId (ns=0;i=85). - Seven defects of the endpoint dialog are fixed. Subscriptions keep their ids and names, and only the default one is the default. Transport, retry and subscription values are saved as numbers. A new endpoint saves the security policy
None, andBasic256Sha256when you choose Sign or Sign&Encrypt. Namespace aliases keep their flags. Check connection no longer sends a field the dialog does not have, and Done stays off when the dialog failed to draw. - No false warning at start-up. Node-RED printed
cannot write global("sterfive")at each start, because the flows were not loaded yet. - The GDS connection logs no deprecation warning. The sessions opened to the Global Discovery Server no longer log
endpoint_must_exist is now deprecatedon every connection, and they present the instance certificate with its own application URI. - A connection URL with a user name and password stays out of logs and editor answers. A password in an endpoint URL no longer appears in Test Connection answers, debug output or the library's log.
[3.42.0]โ
The release where what you configure is what you get: every event arrives whole, browse filters are applied, and a method tells you what it is missing.
๐ฏ Events arrive one by oneโ
If you wired two nodes to a MonitorEvent output, the first one saw the wrong events. A burst of alarms or state changes came out with some events missing and the last one repeated, while the second node on the same wire got them all. The node reused a single message and refilled it, and Node-RED hands the first wire that very object.
- Every event is its own message. Each notification now carries its own payload, so the first node on the wire sees each event once, with its own content. If you added a no-op node in front as a workaround, you can remove it.
- The Select Clauses field accepts commas. It expected semicolons while the hint said "comma separated", and a comma-separated list quietly selected nothing: every event arrived with an empty payload. Commas, semicolons and one field per line all work now, and leaving the field empty to get every field no longer marks the node invalid.
- A select clause the server cannot resolve is reported. An unknown field name is dropped with a warning and the rest are kept; if none of the fields exist, the node goes into error instead of delivering empty events.
๐ Browse filters do what they sayโ
The Browse node has a reference type filter, and it had no effect: asking for Organizes returned every reference, including HasTypeDefinition targets, because a reference type written by name never reached the server.
- A reference type given by name is applied.
Organizes,HasChild,HasComponentand the rest work, as do NodeIds, including a reference type of your own address space. One that cannot be resolved is reported rather than ignored. - The options on the message are honoured.
msg.referenceTypeId,msg.browseDirection,msg.includeSubtypes,msg.nodeClassMaskandmsg.resultMaskwork alongside theirmsg.payload.*form, as the help describes. A browse direction set on the message used to be ignored. - The mask fields are readable and checked. Type
63in Result Mask and it showsReferenceType | IsForward | NodeClass | โฆ, which is what gets saved; a mask written as text now counts as that number instead of selecting nothing but the NodeId. A misspelled flag marks the node invalid rather than quietly meaning "select nothing".
๐งฉ Methods that take a structureโ
Calling a method whose argument is a structure failed with Cannot read properties of undefined (reading 'Value') โ a message about nothing you had written. The cause was almost always a property name: the payload holds one property per argument, named exactly as the server declares it, and Settings is not Setting.
- A missing argument says which one. The node names the argument, lists the arguments the method expects and the properties your payload holds, and shows the shape to use. An argument that is a number or a string still goes out with its default value, now with a warning, so a typo is visible instead of producing a call with
0. - A plain object is enough for a structure. You do not have to write the type of the ExtensionObject: the node reads it from the server. The node help and a new documentation page show both routes โ with and without an ExtensionObject node in the flow โ and the Function node that puts the structure under the argument's name.
๐ Bug fixesโ
- BrowseNames with reserved characters resolve. Names such as
HMI_RT_1::@RedundancyRecoveryState_2, as some servers expose, produced a browse path that could not be read back, or failed as a malformed path. The characters the OPC UA text form reserves are now escaped everywhere in a path, including in the parent names. - Browsing with no reference type set no longer fails. With neither a reference type in the node nor one in the message, the node reported an internal error instead of browsing every reference type.
- The bundled examples run against the live demo servers again. Several had drifted: wrong method targets, models that had moved, a Windows-unsafe file path.
๐ ๏ธ Under the hood โ node-opcua 2.186.17 โ 2.187.10โ
- The embedded OPC UA Server answers reads and writes far faster. Requests the server can serve at once skip most of the per-item work: on the maintainers' benchmark machine, reads of 1000 values went from 38 to about 440 calls per second, writes of 1000 values from 20 to about 80.
- A large address space costs less memory. Optional fields take no space until set, and references are stored more compactly โ about 6% less heap per node over a hundred thousand variables.
- Encrypted connections are quicker. With Sign & Encrypt, a read of a 10 KB value is about 45% faster, and the server spends roughly a third less CPU per small read.
- A server under heavy pipelined reads stays within its memory. A channel now admits new requests only while the work in flight stays within a bound, instead of building every large response at once.
- A namespace's default access restrictions are enforced. They were read under the wrong property name, so nodes without their own restrictions were treated as unrestricted.
โ ๏ธ Upgrade notesโ
- A Browse node with a reference type set may now return fewer references. The filter is applied for the first time. If a flow depended on getting everything, clear the reference type field or set it to
Referenceswith subtypes. - A MonitorEvent select clause that resolves to nothing now stops the subscription with an error, where it used to subscribe and deliver empty events. Check the field names against the ones the picker offers.
- If you run the embedded OPC UA Server with default access restrictions on a namespace, they now take effect. Verify your clients still have the access you intend.
[3.41.1]โ
Your Node-RED instance is now a single OPC UA application โ one identity, one certificate, for both the nodes you connect with and the server you expose.
๐ One identity for the whole instanceโ
Until now an instance carried two OPC UA identities: the client nodes presented one certificate, the embedded server another, both sitting on the same private key. It mostly went unnoticed โ until a key was renewed from the server side, after which every client connection silently stopped working.
- Client nodes and the embedded server now share one ApplicationUri and one
certificate,
PKI/own/certs/certificate.pem. Renewing the certificate, by signing request or from a GDS, renews it for both at once instead of breaking the other half. - Your existing client certificate is the one that survives. On first start it becomes the instance certificate; the server's old certificate is set aside, not deleted. Servers that already trust your Node-RED client keep working untouched.
- One instance runs one embedded OPC UA server. A second one is refused with a message that says which server already holds the endpoint and names its owner, instead of pointing at your licence.
๐ Certificate enrolment that waits for approvalโ
Pulling a certificate from a GDS that requires an operator to approve requests used to end at a bare Error once the polling window closed โ with no way to tell a pending request from a refused one.
- The GDS tab now shows a held request, which one it is and how long it has been waiting, and the approver's reason when a request is rejected.
- The renewal loop runs on its own: it starts at boot, after an enrolment that left a request pending, and when your OPC UA nodes come back from a redeploy. An approved certificate is collected on its own, without you enrolling a second time.
- It stops cleanly when Node-RED shuts down, instead of holding the process.
๐ Bug Fixesโ
- Reading an Int64 or UInt64 array no longer crashes Node-RED. Any array of
64-bit integers raised an uncaught exception about
arrayTypeand took the flow down with it. Scalars were never affected. - Browse paths survive an escaped colon (issue 9). A browse name
containing
::โ written&:&:inside a path, as Siemens WinCC exposes them โ was mistaken for a namespace prefix, so the path the node browser proposed could not be resolved back, and a hand-typed one failed as malformed. - Explore hands Monitor a usable
nsu=address (issue 10). When two namespace aliases shared a name, or a regular-expression alias matched several namespaces, the address Explore produced could name another namespace; a Monitor fed with it failed onns=-1. - The bundled example flows work on FlowFuse (issue 8). The server
demo failed with
global.get('sterfive') is not seton hosts whose generated configuration never declared the slot those helpers were written into. - Two flows configured for the same server get a clear warning, instead of the second one quietly adopting the first one's server and failing later with an unrelated error.
๐ ๏ธ Under the hood โ node-opcua 2.181.1 โ 2.186.17โ
- History Read can ask for bounding values again โ raw-data reads honour the request, and a server with a historian advertises the matching profile.
- Node-RED starts noticeably faster: loading the OPC UA stack now opens about a third fewer files.
- A server that sends its certificate as a chain now connects โ signatures computed over the full chain are accepted, which is what CA-issued server certificates do.
- Clearer answers when a connection is refused: the message a server sends before closing is reported instead of being swallowed, and a server warns when a client reaches it on a hostname its certificate does not cover.
- Alarms and events follow the specification more closely โ a condition refresh reaches only the subscription that asked for it, and events are delivered only to sessions entitled to receive them.
- More companion specifications out of the box, including Surface Technology, with information-model details preserved through a load and save round trip.
โ ๏ธ Upgrade notesโ
- Clients of your embedded server must trust its new certificate. The server now presents the instance certificate โ the one your client nodes already use. Its previous certificate is kept beside it, so you can see what changed, but anything that trusted it needs to trust the new one. Servers your nodes connect to are unaffected.
- An instance that ran several embedded servers now starts one. If you deployed more than one OPC UA Server node in a single Node-RED instance, only the first starts; move the others to another instance.
- Seeded server users are written once, at first boot. Deleting a seeded account now sticks โ it is no longer recreated on the next restart.
- Anonymous subscriptions are no longer transferred over an unsecured
endpoint. An anonymous client reconnecting to the embedded server through a
Noneendpoint rebuilds its subscription instead of resuming it, as the specification requires. Signed and encrypted sessions are unaffected.
[3.41.0]โ
Enrolment under a Global Discovery Server that finishes on its own, and an end to four errors that stopped a flow for the wrong reason.
๐ Enrolment that completes without youโ
Under a Global Discovery Server in manual approval mode, an enrolment used to stall. The panel showed an error once the polling window closed, and an approved certificate was collected only the next time someone enrolled by hand.
- The GDS tab says where an enrolment stands. It shows which request is held and since when, or the approver's reason when a request was rejected, instead of a bare error.
- An approved certificate is collected on its own. The renewal loop now runs by itself: at start-up from the saved configuration, after an enrolment that left a request held, and again when the nodes come back after a redeploy. It runs only for an enrolled application or a held request, and stops cleanly when Node-RED shuts down.
๐ Bug Fixesโ
- An Int64 or UInt64 array no longer crashes Node-RED. Reading one raised an uncaught error about a missing array type, which took the runtime down instead of failing the node.
- The
sterfivehelper is available on hosts that do not declare it. On FlowFuse the embedded-server example flows failed withglobal.get('sterfive') is not set: the helper was written to a setting Node-RED only reads when the host declared it before start-up, and FlowFuse's generated settings do not. It is now also set through a live node's global context, at start-up and whenever the flows start. - A browse name containing an escaped colon is read as a name, not a prefix. A path whose first name holds
&:โ as WinCC servers produce โ was taken for a namespace or alias prefix and resolved against the wrong namespace. - An
nsu=NodeId names the namespace it came from. The value was built by converting the NodeId to an alias and back. With two aliases sharing a name, or a regular-expression alias matching several namespaces, the way back could return a different namespace's URI, and a Monitor fed with it failed onns=-1.
๐ ๏ธ Under the hood โ node-opcua 2.181.1 โ 2.186.17โ
- Node-RED starts faster. The stack loads about a third fewer files at start-up.
- A failed connection reports what the server said. The error message a server sends just before it closes the channel is surfaced instead of being lost.
- Writing a structure to a loosely typed variable works again. A variable declared as BaseDataType accepts any extension object.
- History on the embedded server is more complete. A raw history read honours a request for bounding values, and the server advertises the historical-access facet when a historian is installed.
- Certificate handling on the embedded server follows Part 12 more closely. Adding and removing trust-list certificates, and listing the rejected ones, behave as the specification requires.
โ ๏ธ Upgrade notesโ
- An anonymous client that reconnects over an unsecured endpoint rebuilds its subscriptions. The stack now enforces the OPC UA rule that a subscription moves to a new session only when the channel is signed or encrypted and the client certificate matches the one of the original session. If a client reconnects anonymously to the embedded OPC UA Server node over a
Noneendpoint and expects to keep its subscription, it will re-create it instead. Use a signed or encrypted endpoint if you need the subscription to survive.
[3.40.0]โ
The editor release: dark mode everywhere, a reworked address browser, and an end to the NodeId that silently changed under you.
๐ Dark mode, following Node-REDโ
- The configuration panel and every node dialog now follow the editor theme. Not the operating system โ Node-RED's own Theme setting, so a light editor on a dark desktop stays light. Certificates, licence, onboarding, the address browser, its menus and hints were all reworked for contrast on both grounds.
๐ฒ A better address browserโ
- Browse before you deploy. The address-space tree can be opened as soon as an endpoint is configured, so you can wire and fill in a flow in a single pass instead of deploying just to look up a NodeId.
- The first click on a branch shows its children instead of an empty branch.
- Picking a node writes a browse path, the notation that survives a namespace re-index.
- Verify says what actually went wrong โ the server reason is surfaced instead of a generic failure.
๐ฏ Explicit NodeId sourceโ
-
The prefix of every address field now has two stops: the first says where the value comes from at runtime โ this node, this node, templated, or the message โ and the second says how the address is written. Two questions that used to share one seven-entry menu.
One sentence: a message always triggers a node; whether it also targets it is now your choice. The classic trap โ a stock inject node whose timestamp
payloadsilently replaced a carefully configured NodeId and failed far downstream with a cryptic parse error โ is gone: under this node the message cannot retarget the node at all.Available on Read, Write, Browse, Call (ObjectId and MethodId independently), Explore, File Operation, Monitor, Monitor Event and History Read โ the last two gained it in this release, so every node that takes an address now answers the same question the same way.
-
Notations are shown as their own syntax โ
abc,ns=,na=,nsu=,/,/a:โ so the button and the field agree, replacing four stacked icon pairs that never resolved at small sizes. The last notation you pick is remembered for the next field; new fields start onns=. -
Templated addresses are a declared source. A value such as
ns=1;s={{msg.payload.name}}used to read the message while claiming to come from the node. -
msg.mode keeps every shape. A message path pointing at an array or an object tree drives a multi-node read or an array/structure subscription exactly as a payload would โ including on the Monitor node.
๐ Bug Fixesโ
- Valid addresses were being rejected. The guard that catches "this is a data value, not a
NodeId" was narrower than the resolver itself, so four legal forms raised a hard error: browse
paths starting with
., the root-anchored form, thensa=prefix, and bare well-known names such asServerorObjectsFolder. This affected the Monitor node in every mode, including a NodeId configured in the node itself. - A node no longer stays yellow for ever. When a message overrode a configured NodeId, the
status turned yellow permanently and lost its real colour โ a
sampling...Monitor never showed green again. - The Call node reports both fields. With ObjectId and MethodId overridden, only the first was ever logged.
- New nodes are explicit from birth. A node dragged onto the canvas used to start in the legacy mode and immediately warn about behaviour it had no history of. Existing flows are untouched.
- Help and documentation links on the OPC UA nodes are corrected, and the missing help widgets are restored.
- The licence panel survives a blocked payment provider and still offers the quote path, with the correct URL and a negotiated locale.
๐ ๏ธ Under the hoodโ
- Platform bump to node-opcua 2.180.0.
- A generated software bill of materials, shared with the node-opcua GDS tooling.
โ ๏ธ Upgrade notesโ
- Browse and verify now appear only under this node. They act on an address the node owns, so they are absent under the message, under a template, and while the source is unanswered. On a flow saved before 3.40 they appear as soon as you pick a source.
- Flows saved before 3.40 keep their historical per-node precedence until you choose a source. Nothing changes on deploy.
[3.39.0]โ
๐ New Featuresโ
- Embedded OPC UA Server โ role-based security, on by default. The embedded server ships with a spec-conformant identity and role layer (OPC 10000-18): well-known roles, per-role identity mapping, user management with a configurable password policy, a must-change-password flow for seeded credentials, and an encrypted-at-rest user store โ credentials never live in the flow.
- Registration without internet access. Machines with no outbound connection can register offline: the onboarding wizard renders a QR code with your details pre-filled, and opens it automatically when an online attempt cannot reach the service.
- Recovering from an expired or stuck licence is no longer a dead end โ the configuration panel offers a renew/change action and can rebuild its licensing runtime in place, so a repaired environment does not need a Node-RED restart.
๐ Bug Fixesโ
- Nodes no longer flash red on deploy โ a node status is cleared cleanly on close instead of being left in a transient error state.
- The Certificates page no longer renders blank on some installations.
- Firefox: toggle checkboxes in the configuration panel render correctly.
- Monitor: namespace context is passed to the subscription, so monitored items using namespace aliases or browse paths resolve correctly.
๐ ๏ธ Under the hoodโ
- Platform bump to node-opcua 2.175.2, bringing reconnection fixes (publishing resumes after a
RepublishreturnsBadMessageNotAvailable, as seen with CoDeSys), stricter event-filter validation, a hardened transport handshake, and all outstanding upstream security advisories resolved.
Node.js: minimum 20.20.0; Node 22 LTS strongly recommended.
[3.38.0]โ
๐ New Featuresโ
- FlowFuse-aware experience. On FlowFuse instances the configuration panel now shows only what applies to that environment, and hides actions an immutable container cannot honour.
๐ Bug Fixesโ
- Configuration panel reachable behind sub-paths and reverse proxies โ it derives its API base from the page it is served from, instead of assuming the editor sits at the server root.
[3.36.0]โ
๐ New Featuresโ
- Self-healing monitored items (
Monitornode). Monitoring now survives variables that are not there yet, and servers that come back in stages:- a node that does not exist yet is tracked as pending and retried until it appears, rather than becoming a dead monitored item โ for single nodes, arrays and structures, at scale;
- after every reconnection the node re-validates its entire monitored set and re-monitors only what came back missing (a PLC still warming up after a reboot). This closes a long-standing gap where a staged restart could leave some tags silently frozen until a redeploy;
- Retry unknown nodes (default on) and Retry interval (default 5 s) are configurable.
- Mark values bad on disconnect (
Monitor, default off) โ re-emit the last value of each item taggedBadServerNotConnected, so downstream flows can tell the data is stale instead of trusting a frozen value. - Array monitoring output gains
msg.statusCodesandmsg.nodeIdsโ parallel arrays in input order, so per-item status and the resolved nodeId are visible. - Output type and Trigger are first-class options on the
Monitornode, and are correctly saved.
๐จ Changedโ
- Duration fields accept units. Sampling Interval, Retry Interval and the subscription Publishing Interval take a value plus a unit (ms / s / m / h). Values are still stored in milliseconds, so existing flows are unaffected.
๐ Bug Fixesโ
- Migrated nodes get correct defaults โ opening a
MonitororEndpointnode saved before these options existed no longer shows blank combos or a0retry interval.
[3.35.0]โ
๐ New Featuresโ
- Embedded OPC UA Server (preview). Host an OPC UA server directly from your flow, including PubSub alongside the classic client-server endpoint. In the Evaluation tier it runs with
[Evaluation]branding, stops after 2 hours, and is capped to a small number of nodes.
๐จ Changedโ
- Minimum Node.js is now 20.20.0 (was 16.x); Node 22 LTS is recommended, and the toolchain warns on anything older. This lifts the hard Node-22-only constraint of the 3.34.x line.
๐ Bug Fixesโ
- Installing on a fresh Node-RED instance no longer fails with a missing-module error โ the dependency closure is resolved at install time instead of relying on host hoisting.
[3.34.0]โ
๐ New Featuresโ
- Certificate pull reports progress while it retries, and can be cancelled. First shipped in the 3.34 line.
- First version of the embedded OPC UA Server node, as an experiment. Iterated further in 3.35.0.
โ ๏ธ Known limitationโ
- Requires Node.js 22 or above. The 3.34.x line will not start on Node 20 or earlier; this is fixed in 3.35.0. Note that 3.34.2 is the first working publish of the Sterfive distribution to npmjs.org โ from that version you can install directly with
npm install @opcua/for-node-red.
[3.31.0]โ
๐ Bug Fixesโ
-
All persistent state now survives a container rebuild. On the official
nodered/node-redimage, the PKI and the licence state were written outside/dataโ the only volume that image mounts. Every container rebuild therefore wiped the trust store, the rejected list, the client's own certificate and the licence activation together, silently. All of it now lives under the/data-backed user directory, and if anything ever resolves a persistent path too early again, start-up fails loudly instead of quietly writing to the wrong place.Migration. An existing install whose PKI sits at the old location gets a one-time warning at start-up naming the new one. Copy the folder across yourself โ nothing is moved automatically, because certificate material is sensitive โ or set
STERFIVE_CONTAINER_CONFIG_FOLDERto the old parent directory to keep the previous behaviour. See Setting up a Custom Certificate.FlowFuse is unaffected: its PKI stays exactly where it was.
-
Self-signed certificates are written where the OPC UA client actually reads them, so creating one from the Certificate Manager takes effect immediately. If you hit this before upgrading, a stray
own/certs/self_signed_certificate.pemleft in your PKI folder can safely be deleted. -
Clearer activation failures โ when a key is rejected, the reason is shown instead of a generic error.
[3.30.0]โ
๐ New Featuresโ
- Certificate enrollment is editable. The enrollment form exposes the DNS names, IP addresses and application URI that go into the request, pre-filled from your PKI rather than from hard-coded values, and the certificate information panel shows them back to you.
- One application URI across the board โ the same identity is used consistently by every OPC UA subsystem, so a certificate issued for it is accepted everywhere.
๐ Bug Fixesโ
- Two memory leaks fixed: an event-listener leak in the
Monitornode, and a re-fetch loop in the certificate-pull screen that could grow without bound. - Confirmation dialogs in the configuration panel are inline rather than browser pop-ups, and the upload button no longer wraps awkwardly.
[3.28.0]โ
๐ New Featuresโ
- You are told when the product cannot reach the network. Operating in the offline grace period now shows a prominent warning, instead of failing quietly later.
[3.27.0]โ
๐จ Changedโ
-
Output ports are consistent across every Client node. The experimental secondary output has been removed, and all errors now flow through the standard Node-RED error mechanism.
Migration: wire a standard Catch node beneath your client nodes to intercept errors.
๐ Bug Fixesโ
- Fixed timeouts caused by concurrency and certificate-store races during connection setup.
[3.26.0]โ
๐ New Featuresโ
- Certificate Manager Integration: Introduced a visual Certificate Manager tab directly inside the endpoint configuration panel. It issues, trusts and re-generates certificates without leaving the editor.
- Streamlined Trial Registration: Added a "Welcome Registration" flow that allows users to rapidly register and acquire a short-term Trial ID straight from the UI. This drastically reduces the time-to-value for new trial evaluators.
- FlowFuse Capability: Improved compatibility with FlowFuse deployments, specifically fixing PKI API registrations and licensing verification within containers.
[3.24.0]โ
๐ New Featuresโ
- X509 User Identity: Upgraded the core OPC UA client components (to
node-opcua5.20+), which introduces full support for X509 User Identity when connecting to secure servers. - Telemetry & PKI Configuration: Enhanced folder configuration capabilities for PKI and telemetry data.
- UI Polish: Replaced crude alerts with native, proper modal dialogs when interacting with Certificate details in the config panel.
[3.23.0] & [3.22.0]โ
๐ New Featuresโ
- Extensive Reverse Proxy & Subpath Support: Natively brought in support for
RED.settings.httpNodeRoot. The module APIs consistently respect Nginx and custom subpaths context, crucial for enterprise Kubernetes and Docker deployments. - Frictionless Purchasing: Overhauled the Support and Licensing overlay to feature a direct "Buy Online" workflow, heavily minimizing the steps required to acquire or upgrade a professional license.
- Licensing UI Revamp: The license expiration overlay and the support tab were beautifully redesigned to cleanly expose your product information and telemetry APIs.
๐ ๏ธ Core Improvementsโ
- Advanced White-labeling: Substantial under-the-hood refactoring to support configurable API namespaces and global RED instances. This lays the groundwork for OEM partners wanting to white-label the extension.
- License Limitations: Refactored the core logic determining offline vs. online license constraints, providing a more robust enforcement model.