Memory usage on UbuntuVM at 100% - 6.0.0-rc.4 showing memory.current rather than MemAvailable

[wt9bind]

Pulse version: 6.0.0-rc.4
PVE version: 9.2.x (2-node cluster)

Summary

For VMs without ballooning, Pulse displays memory usage at ~100% even when
the guest has large reclaimable page cache and ample MemAvailable. This is
because PVE's /nodes/{node}/qemu/{vmid}/status/current returns mem sourced
from host-side cgroup memory.current (QEMU process RSS), which counts guest
page cache as resident. Pulse displays this value faithfully — so this is not
a calculation bug in Pulse.

Observed

VM 101 (Ubuntu 24.04, 8 GB assigned, no balloon):

• PVE API: mem 8588005376 / maxmem 8589934592 → 99.98%
• QEMU process VmRSS: ~8.0 GB
• Guest-internal MemAvailable: ~7.1 GB (695 MB actually used, 7.09 GB buff/cache)
• PSI pressure_memory_full / some: 0 (no real pressure)

Result: Pulse shows the VM red at 100% while the guest is effectively idle.

Request

Where a QEMU guest agent is present, optionally fetch guest MemAvailable and
display a reclaimable-adjusted memory figure (or a secondary "guest view"),
similar to the host-vs-guest distinction the PVE web UI now surfaces. Could be
a per-guest or global toggle, defaulting to current behaviour to avoid
surprising users who rely on the host-side number.

Notes

Affects any cache-heavy VM without ballooning (media servers, DB hosts).
Not reproducible as a "wrong maths" bug — the value is correct per the API;
the request is for an alternative, less misleading presentation.

Screens / Pastes

Memory usage on UnbutuPlex pinned at 100% (Ignore SophosXGS)

free-h showing:
cat /proc/meminfo | grep -E 'MemTotal|MemFree|MemAvailable|Cached'
total used free shared buff/cache available
Mem: 7.7Gi 682Mi 421Mi 11Mi 6.9Gi 7.0Gi
Swap: 4.0Gi 268Ki 4.0Gi
MemTotal: 8063064 kB
MemFree: 431936 kB
MemAvailable: 7364284 kB
Cached: 6928328 kB
SwapCached: 4 kB