Two are free with zero barrier to adoption.
Each answers a genuinely different diagnostic question.
Cloud TraceWhere did time go across a request?
Cloud ProfilerWhere does the app spend its OWN CPU/memory? (free)
Error ReportingAuto-groups exceptions (free)
Managed PrometheusSame PromQL, no infra to run
"The database query took 800ms of a 900ms total" is a materially more actionable diagnosis than "the request was slow" — that's what Trace's span waterfall actually shows.