Also just fixed a "bug" in my yesterday's writing - it was late and I had picked a number from a wrong column for one explanation! :-)

And that's exactly why I added the computed "optimizer cardinality misestimate" column to my XB scripts, so I wouldn't have to do such lookups & math manually.

But yeah, most of my demos use fewer partitions to keep things manageable

Telco CDR & usage records etc, seen 15 min time partitions with 1024 hash sub-partitions for each time range in the past 😀

Or if you have hourly time partitioning then 10 years of history will be 100k partitions

I've already figured out how to add (dynamically at-runtime chosen) any function call latency tracking into the next release too 😎

In big Oracle data warehouse world, some setups relied on the ETL engine for referential consistency (no DB constraints that would have required an index used), but then these constraints were still “virtually enabled” (Oracle has a NOVALIDATE RELY mode) for helping the optimizer to help you more 😀

It’s worth noting that enabled & validated constraints also give the optimizer additional paths for query optimization if it has guarantees that some column is never null or is guaranteed to be unique or must have a counterpart row behind that foreign key, etc

The P99 CONF page is www.p99conf.io/session/xcap...