Alan Protasio

Hi @prymitive thanks for pointing that out. I can try to make it in the same go routine, but for this we would need to use something like `TryLock`, which...

Hi @bboreham thanks for looking. There is already a graph on the description that shows CPU before and after... it's hard anything as sometimes we can see exactly the same...

Yeah... this is cortex and is a sum of multiples instances (ingesters). Heap Size is the average btw the ingesters instances. The spikes in CPU and Memory is exactly during...

> It was pretty much an ad-hoc decision to stick with the 1kb limit for histogram chunks and merely require that at least 10 histograms are in each chunk. Have...

@codesome 🥺 :)

> In your comparison charts, the CPU is spiking higher. And as I understand, CPU is much more expensive than giving a little more memory. > > Can you please...

Parca only have the profiles after 3:42? (after the problem hapenned? :( ) Seems for some reason i dont see the `(*DB).run` go routine on the new version at 3:42......

Ok.. I did find the deadlock between this 2 go routines: (it happens during series GC and maybe my tests did not had churn so i did not see this...

I think this deadlock can happens nowadays as well in a very specific scenario: If the autoCompact is disabled and we call `db.Compact` from another go routine at the same...

Ok.. i created a test to reproduce the problem. [Here is the run](https://github.com/prometheus/prometheus/actions/runs/10517106007/job/29140594716?pr=14610) is the deadlock happening the test failing: ``` panic: test timed out after 10m0s running tests: TestMMapAndCompactConcurrently...