strawgate
commented
2 weeks ago If we decide that an lru is not a good idea we could just allowlist and bypass normalization for common attributes (every object includes the same core attributes like name, id, friendly name, etc)
We can likely do the same for the sensor domain.
My guess is this would remove the vast majority of calls to the normalize
Roughly 1/2 of processing time within the publish loop is spent sanitizing datastream names and normalizing attribute names.
These both are good candidates for caching as the values don't change and the sanitization/normalization is not dynamic