IdbMigrationStrategy for Device ExtensionPoints

Kevin.Herron · February 25, 2025, 7:10pm

~~new DataValue(new Variant(StatusCodes.Bad_AttributeIdInvalid))?~~

pturmel · February 25, 2025, 7:13pm

Tried new DataValue(e.getStatusCode()) Is that not the same?

Kevin.Herron · February 25, 2025, 7:13pm

Ah shit, that's actually correct and what I suggested was not.

Kevin.Herron · February 25, 2025, 7:15pm

In "easy mode" when you have actual UaNode instances and defer to the "Managed" impls that's basically what's happening:

github.com/eclipse-milo/milo

opc-ua-sdk/sdk-server/src/main/java/org/eclipse/milo/opcua/sdk/server/AttributeReader.java

580bc5bbf


      
          public static DataValue readAttribute(
              AccessContext context,
              UaServerNode node,
              AttributeId attributeId,
              @Nullable TimestampsToReturn timestamps,
              @Nullable String indexRange,
              @Nullable QualifiedName encodingName) {
          
            if (!AttributeId.getAttributes(node.getNodeClass()).contains(attributeId)) {
              return new DataValue(StatusCodes.Bad_AttributeIdInvalid);
            }

pturmel · February 25, 2025, 7:20pm

Ok. Let me prune back my experiments and try again.

Edit: Oy! My driver is choking on disable then re-enable, and I hadn't noticed.

My subscriptions are working. Writes are writing. Phew!

The IA Logix driver isn't browsing at all, even when everything else is disabled. Weird results from your MB server driver.

FWIW, I'm not using the ManagedDevice layer, but I am using a node manager for my address space, and creating UaNodes on demand.

Kevin.Herron · February 25, 2025, 8:02pm

Hmm, I'll have to look into this. Hopefully whatever this is happened recently.

Kevin.Herron · February 25, 2025, 8:26pm

It's a UI/config bug... I think the Logix driver had its UI built before we had the generic extension point config page, and some of the settings moved but the move isn't reflected on the frontend/UI...

anyway end result is that the Concurrent Requests setting in the UI is ignored and the driver ends up getting created with 0 allowed, i.e. 0 CIP Connections allowed, and there's not sufficient assertions/validations to blow up when this happens.

Didn't get noticed because we've been using the API or restoring backups.

Kevin.Herron · February 25, 2025, 8:30pm

I'm gonna be hammering the Milo sample server with the UA compliance test tool soon-ish as I get ready for an IOP event in March. Hopefully I'll shake out any remaining browse bugs or performance issues.

I strongly suspect I wrecked browse performance in some way when I implemented the access control stuff + refactored all the browse helpers. It feels like something went exponential.

pturmel · February 25, 2025, 10:16pm

I replaced my open-coded attribute reads and writes with AttributeReader and AttributeWriter static methods. My ClientDriver is screaming along now. (Having to look at my own diagnostic tags, as the Web diagnostics aren't there for devices yet.) One of my torture tests is running 183k items ~500 requests ~1400ms actual sampling time (L81 front port, ~99% Class3 load).

I'll have to look at why it doesn't want to disable properly.

I'll test that, and my HostDevice driver type, tomorrow. Time for an adult beverage.

pturmel · February 27, 2025, 9:47pm

For posterity: this bug was a race between device creation, channel creation, and device .startup(). v8.3's startup speed is both delightful and annoyingly good at exposing this kind of problem.

(My compliments to IA for mitigating my biggest concern about killing module hot-loading.)