Thank you everyone for your opinions!!
Answering to multiple questions at once:
- of course, I reviewed the list and iterated before sending, I'm not feeding you raw LLM output
- however, it was a very raw draft just to show the direction of work, I needed to hear opinions before moving further
- opinion on "not making wait events CPU profiler" -- understood, agreed, I'll exclude that from the scope of work
- on the wait event analysis tools with green areas and coalesce(wait_event, 'CPU') labels:
- agreed, it's their problem, and I discussed it with many of them (e.g., with RDS not once), they should fix it there as well; e.g., we use "CPU*" label and have a remark, that it means "CPU or unknown wait event"
- I think every tool should do something like this, because 100% coverage is hardly achievable
- one thing that can be improved here from Postgres in this area is wording in the docs -- now it's "wait event name if this backend is currently waiting, otherwise NULL", which tells that NULL is non-waiting, like there is no chance that some waiting parts of the code are not yet covered (which is obviously not so, and probably will never be); perhaps, we could use some term admitting that there are unknowns -- e.g., we could say "wait event name if this backend is currently known to be waiting, otherwise NULL" or even more clearly explaining that NULL can be either "not waiting" or "waiting but not instrumented yet"
- my understanding is that the feedback is generally positive so far, so I plan to iterate on this topic and return with specific proposals for each area separately
Nik
