Poor performance when huge number of tablets

Using the default tablet splitting values, you’d have 24 tablets per yb-tserver, 24*40=960. And this number would stay until you had 100GB tablets Tablet splitting | YugabyteDB Docs

You used “SPLIT INTO 1500” when creating this table, correct?

You can explain some other tables. Or we can lower the defaults and override more tablets only for tables that we know will grow big quickly.

@Meteo

For reference, can you explain:

  1. Version of yugabytedb you’re using (exact commit if custom build)
  2. What changes did you do to the db since you custom build & why?
  3. Can you paste any custom config you’ve set on yb-tserver & yb-master?
  4. What is the cluster RF?
  5. When you mean 1500 tablets, you mean BEFORE RF, correct?
  6. Can you explain why 2000 tables? Is this natural cause it’s just a big app? Or is it because of the partitioning? If partitioning, then please explain some scenarios so we can lower them too.
    Or do many of the 2000 tables have the same schema? Like creating 1 table for “type-of-thing”, to scale them?
  7. Regarding your previous thread, Support of 100.000+ connections, how many connections does each node have? Are you using Connection Manager in YSQL | YugabyteDB Docs ?
  8. When pasting an explain/analyze, please include the query too for clarity.