Need Guidance on OPC UA Extractor Configuration with Flexible Data Modeling (CDF)

​@Mohan Narayanan sorry for the late reply. Please find the answer to your questions below.

1. The only view the extractor writes to directly is CogniteExtractorTimeSeries, which implements CogniteTimeSeries .
2. I have attached a sample .yml file below.
3. The docs.cognite.com link is correct and has some help about recommended practices.
4. OPC-UA Events can be mapped to CDM records, or CDF classic events only.

Thank you for your response and for sharing the sample configuration file.

Over the past couple of weeks, I have been working with the OPC UA Extractor and have successfully configured and ingested time series data into CDF. However, event data is currently being pushed only to Classic Events. I'm still exploring the best approach to move beyond that and would appreciate any guidance.

I have a follow-up question: if events are mapped to Classic Events, what is the recommended approach to make them available in a Flexible Data Model or the Cognite Data Fusion Core Data Model (CDM)?

My end goal is to have these events represented within my own data model space, the Cognite Core Data Model (CDM), or Cognite Activities, so they can be consumed by downstream applications such as Atlas AI RCA Agents. Ideally, when an RCA Agent is created, it should be able to seamlessly access, correlate, and analyze these events together with other contextual operational and asset data to support root cause analysis and decision-making.

If possible, could you share your thoughts or recommended architecture for achieving this without using Transformations? I am particularly interested in any native or configuration-based approaches that may be available.

Thank you in advance for your help and insights.

Thank you for the detailed context; great to hear you've successfully ingested time series into CDF!

Regarding your question about moving events beyond Classic Events into a Flexible Data Model or CDM, here is what is currently available based on the OPC UA extractor's capabilities:

Native/Config-Based Options for Events

1. BETA: Write Events to Data Modeling Records (Streams/ILA)

The OPC UA extractor has a BETA records section in the cognite configuration that writes OPC UA events directly to data modeling records (backed by the Industrial Log Analytics / Streams service). Containers corresponding to OPC-UA event types are created automatically in the configured model space:
 

cognite:
  records:
    log-space: "your-log-space"
    model-space: "your-model-space"
    stream: "your-stream-name"

This requires the service account to have the following capabilities: streamrecords:read, streamrecords:write, streams:read, and streams:write.

 Note: This is a BETA feature, so exercise caution in production environments. Also, if your server produces many different event types, this can create a large number of containers. 

2. CogniteActivity / CDM Activity(Not Currently Supported Natively)

Unfortunately, the OPC UA extractor does not currently support natively mapping events to CogniteActivity or other CDM schedulable types. As noted by the Cognite engineering team, OPC UA events are typically records of occurrences with a single timestamp, which doesn't cleanly map to the schedulable CDM activity concept (which is designed for planned events like work orders). This mapping is something the team is still working through.
 

Recommended Architecture (Without Transformations)
 

Given your goal of making events available for Atlas AI RCA Agents alongside asset and operational data, here is what I'd suggest considering:

1. Use the records BETA feature if your use case aligns with stream-based event logs and you have access to the ILA streams capability. This gives you events in data modeling natively.
2. Keep Classic Events short-term while the OPC UA extractor's full CDM event support matures; OPC UA support for stream records is actively being developed.
3. For custom event correlation and enrichment (e.g., associating counter/frequency data with events), the knowledge sources indicate this level of customization is currently beyond what the extractor can handle natively and would require a custom script using the Python SDK or forking the open-source extractor.

I'd recommend reaching out to your Cognite account team to get the latest roadmap status on CDM event support in the OPC UA extractor, as this is an actively evolving area.