RSS is not a true number. It's more of an estimate than anything. The Linux
kernel tracking of process memory is fairly lazy. Especially when you get
into more complex things like MADV_DONTNEED vs MADV_FREE.
If you really want to get a more realistic number, you can look at PSS, but
it's an expens
Also, I had a look at Go memory utilization. I see that Go memory
utilization(go_memstats_alloc_bytes) is around 50% of the total memory used
by Prometheus(process_resident_memory_bytes)
On Tuesday, February 9, 2021 at 2:39:42 AM UTC+5:30 yagyans...@gmail.com
wrote:
> Thanks, Ben. I'll upgra
Thanks, Ben. I'll upgrade to a newer Prometheus version and check if the
issue still persists.
But I still have one doubt here, I am running this Prometheus instance for
almost an year now, but I have noticed this memory increase recently only.
First time on 31st Jan and 2nd time on Feb 8. If it r
Prometheus performs compactions at regular intervals. This is likely what
generated some IO.
Note, if you're just looking at RSS, this is not going to tell the whole
story. Depending on which version of Go you built with, it may not be fast
at reclaiming RSS memory.
Look at the go_memstats_alloc_
Hi. I am using Prometheus version 2.12.0. I am running Alertmanager 0.21.0
in cluster mode. Since, last 9 days, I have observed twice that the memory
consumption by Prometheus increased by 10-12% and it remained to the
increased value there after. Interesting thing to note here is that both
th
5 matches
Mail list logo