Have Prometheus analyse its own heap profile #8
Description
People see "It's using 86GB", and then they don't know how that is made up.
We could run the profiler and then walk the output, and show some big blocks using simple rules:
- GC overhead (i.e. heap * GOGC/100, modified by GOMEMLIMIT)
- Scraping series and drop caches
- TSDB head series
- TSDB head series labels
- Inflight queries
- Rules
This would need to be updated over time as code structure changed.