Hoppa till huvudinnehållet

Hardware requirements for the OPC UA extractor

The hardware requirements of the extractor depend on the configuration and server size.


At some points, the extractor replicates the entire node hierarchy and buffers up to a few minutes of data points and events at a time. Running the extractor against a small server should take no more than 100 MB of memory, while running against a larger server (> 1 million nodes) may take as much as 3–4 GB. If the server is relatively small, a VM with a single core and 1 GB of memory should be sufficient.


In most cases, the bottleneck is the OPC UA server and the network, not the extractor. The OPC UA client operates over a single TCP connection, and the extractor will generally not have to do more than some simple transformations of the data arriving over that connection. Heavy use of complicated regex transformations can cause issues, but those only apply on startup. Regular operation of the extractor doesn't require much processing power.


By default, the extractor doesn't write to disk, apart from these exceptions:

  • If failure-buffer is configured to buffer data points to disk, the extractor may write an arbitrarily large amount of data. The data points are buffered directly, and if the connection to Cognite Data Fusion (CDF) is down for a few hours, with a few hundred data points per second, this may grow fairly large.

    If the intent is to use the buffer, and the server doesn't support history, we recommend giving the extractor a decent amount of permanent storage. The data points are removed once the connection to CDF is re-established.

  • The extractor is configured to write logs to disk. These will generally not be very large unless verbose logging is enabled. We don't recommend verbose logging for production use.

CDF capabilities

The extractor needs the following CDF capabilities on the given data set if this is specified:

  • timeseries:read, timeseries:write, unless there are no variables mapped to timeseries on the server. This can be facilitated by adding a transformation to ignore or transform to properties of the Variable NodeClass.
  • extractionpipelines:read, extractionruns:write for extraction pipelines.
  • assets:read, assets:write if raw-metadata or skip-metadata aren't set.
  • events:read, events:write, if events are enabled.
  • raw:read, raw:write, raw:list, if raw-metadata is enabled, or the state-store is set to use CDF RAW.
  • relationships:read, relationships:write, if relationships are enabled.
  • datasets:read, if using dataset-external-id.
  • projectsAcl:list, and groupsAcl:list if using OIDC tokens for authentication.