Hey guys, what sort of architecture would you choose to go for if you wanted to have separated handling of queries according to the client needs - whether "user interface" or "reporting"? More on this thread.

You can have single Pinot cluster but provide isolation via Pinot tenants (hosting different workloads on different tenant tags)

Got it. That's interesting. But then the mechanics of ingesting data would have to be duplicated as well, right? Like, if I create segments with SegmentTarPushTask for one tenant, I would also have to do the same for the other tenant?

Are the two tables going to mirror the data? I think you can push same data to multiple tables

But then I would end up having Controllers and Servers sharing the duty to respond to queries for both tables, no? Or is there a way to separate this? What I'm thinking of is to have a set of Controllers and Servers for each "department" (UI and reporting), so that there's absolutely no way of one set interfering into the other - for example, if a server gets OOM-killed because of some very large reporting task, I want this to not even be felt by the UI at all. Or like too many reporting tasks taking too much RAM and CPU from all the servers and the end-users starting to see the UI responding too slow.

Controllers can serve thousands of tables. Brokers also have tags, so you can do separate tags for brokers as well. Servers are the ones that do most of the work, so you definitely want to isolate them.

Ah, I didn't know about tags. I'll search more about that. And about isolating Servers, how would you do that?

Same - tenant tags

Ah, nice! OK then, sounds like a good plan 🙂