Dashboard Day Slowdown

The reports consume the same CPU, memory, and connection pool slots as checkout traffic, pushing write-path tail latency up

Reporting queries corrupt the planner's table statistics, which is what makes checkout queries choose the wrong index afterwards

A single large scan evicts the hot working set from the buffer pool, so ordinary point queries start going to disk

Long-running reporting transactions hold back the cleanup horizon, so dead row versions accumulate and tables bloat

Analytics reads take row-level locks on every row they scan, so concurrent writers block until the whole report finishes