Container memory-current keeps growing while process RSS stays flat

RouterOS Containers
1

Hi,

I'm trying to understand how RouterOS accounts memory for containers.

Setup:

  • RouterOS 7.21.x
  • RB5009
  • Native RouterOS containers
  • Distroless Rust application (DNS server)
  • process_resident_memory_bytes (RSS) exported by the application stays almost constant at ~42–43 MB.
  • Application heap breakdown also remains stable.

However, RouterOS reports memory-current steadily increasing:

  • After restart: ~29 MB
  • After boot completes: ~65–70 MB
  • After several hours: ~100 MB
  • It continues to grow even though the application's RSS and heap remain essentially unchanged.

The application periodically writes small log/history files and updates cached rule files under /data, but there is no evidence of a userspace memory leak.

Setting memory-high=100M caused RouterOS to terminate and restart the container, so that is not a usable solution.

Question:

Is memory-current in RouterOS simply the Linux cgroup memory.current, including page cache, filesystem cache, kernel memory, socket buffers, etc., or should it closely track the process RSS?

More specifically:

  • Can memory-current continuously increase only because of file-backed page cache, while the process RSS remains flat?
  • Does RouterOS reclaim that cached memory automatically under memory pressure, or is there any recommended way for applications to hint that cached file pages are no longer needed (e.g. posix_fadvise(POSIX_FADV_DONTNEED))?
  • Is there any way in RouterOS to inspect the cgroup memory breakdown (anonymous vs file cache vs kernel memory) for a container?

I'm mainly trying to determine whether this behavior is expected Linux page-cache accounting or whether it indicates a real memory leak.

PS: since yesterday

{B2D26EB8-7275-4495-B0AC-A51F493D996E}
{B2D26EB8-7275-4495-B0AC-A51F493D996E}737×474 21 KB

2

Usually page cache is not included. I suspect heap fragmentation.

3

Well, it seems I've found a solution by settings memory-high:

[bobdenaut@bobdenaut-router] > /container/config/print
registry-url: ``https://registry-1.docker.io
username:
layer-dir:
tmpdir: /kingston/pull
memory-high: 200.0MiB
memory-current: 122.5MiB

and the result is really good:

[bobdenaut@bobdenaut-router] > /container/print
Flags: R - RUNNING
Columns: NAME, ROOT-DIR, INTERFACE, MOUNTLISTS, ENVLISTS, MEMORY-CURRENT, CPU-USAGE, TAG

NAME          ROOT-DIR                     INTERFACE  MOUNTLISTS      ENVLISTS      MEMORY-CURRENT  C  TAG

;;; fastadhunter
0 R fastadhunter  /kingston/fastadhunter/root  veth1      fah-config                    53.7MiB         0  docker.io/library/fastadhunter:0.2.5
fah-data
;;; postgres
1 R postgres      /kingston/images/postgres    veth2      MOUNT_POSTGRES  ENV_POSTGRES  17.6MiB         0  registry-1.docker.io/library/postgres:17-alpine

And my "fastadhunter" still consumes only around 45 mb but routerOS reports 53.7 mb :slight_smile: :

Ruleset 24.32 MB
DNS Cache 2.65 MB
Query Log Ring 2.61 MB
Stats Aggregates 0.57 MB
Stats Clients 0.06 MB
Residual (runtime, allocator, stacks etc.) 13.99 MB
Total RSS 44.20 MB