Is there performance issues possible with a high cardinality subjects for a stream?

[ashumkin]

We're going to make a stream (file storage, retention 10GB, 5-10 days) with the wildcard, way, subject.>. And put there a lot of (about 1 million) messages like subject.<random-UUID-v4-here>.
Is there any performance issues expected because of subject cardinality?
In any case: consuming, subscribing, publishing, cluster startup, RAFT leader election, etc, whatever.
Going to use a NATS cluster with 3 nodes.
Consumers based on the faststream (Python + Django) and Go SDK library.

Any response is appreciated.

[jnmoyne]

No problem with having a lot of subjects in a stream in terms of performance impact, even with a lot more than a million subjects.

The only thing to keep in mind is that the server will use RAM to store its index of subjects (radix tree index). It depends on the size of the subjects of course but for something around 1 million subjects it would be in the order of 100 megs of RAM (with small subjects).