I am considering the schema from Signoz as opposed to OpenTelemetry contrib Clickhouse. Mainly the use of a json

string

column for labels, rather than a

map

What was the decision point for this approach? It seems like this would limit the optimizations for read. I expect all metrics would have 10+ attributes, using a json string, and constantly parsing would not be optimal. If I am storing 200+ PBs of metric data, that could add up, making a map an viable alternative.

The string is used in the time seires table not the samples table. What is the maximum number of unique time series do you expect to have per metric