Service 0a02ddfc-5913-4e84-961f-dc8888af5ca8, project

49e739bd-587c-4d35-8be7-2c985e1018c8. Single replica, 8 GB, Singapore.

My service streams large file downloads. A single job runs for roughly

five minutes and holds a few hundred MB while it does, then returns to

baseline. Several concurrent jobs would multiply that.

I want to confirm production memory against measurements taken in a

local harness, so I need to know what the Metrics view can actually

show me.

1. What sampling interval does the memory graph use, and is it the same

   at every time range or does it aggregate as the window widens?

2. If the graph aggregates, is it showing peak or average within each

   bucket? A two-minute spike that gets averaged into a wider bucket

   would be invisible, which is exactly the event I care about.

3. Is there any way to see the peak memory a replica reached, rather

   than a sampled series? Even a single high-water figure per deployment

   would be more useful to me than a smoothed graph.

4. If a container is killed for exceeding its memory limit, what appears

   in the deploy log, and is the memory reading at the moment of the

   kill retained anywhere?

I am asking because I would rather rely on your metrics than build my

own instrumentation, but only if they can resolve an event of this

duration.