Do you support storing such high cardinality metrics in disk alone and not in the in-memory index?
Secondly, we only look at the unique time series in an hour when computing the billing. This should help us handle ephemeral metrics a lot better compared to many disk based data stores.
Finally, we do support high cardinality since the underlying datastore is columnar. Existing customers are sending on the order of few million cardinality per metric.
re: storing high cardinality metrics in disk vs in-memory index, vijay will comment shortly.
2. We do not have a global index - each hour of metrics have their own index. This helps with ephemeral labels as size of the index remains small