In my vision, poll_mem is mostly used to determine, where your physical RAM leaks. VIRT memory doesn't show physical RAM occupation. To get process physical memory occupation, we should use RES memory.
To clearly illustrate, here is part of the log, showing chrome memory consumption:
PID DATE TIME ELAPSED CPU MEM RSS VSIZE CMD
12049 2021-09-08 17:16:57 00:01 24.0 0.4 135376 38186116 /opt/google/chrome/chrome --type=renderer --enable-automation --enable-logging --log-level=0 --remote-debugging-port=0 --test-type=webdriver --allow-pre-commit-input --field-trial-handle=9868597751347894126,1640417613699088952,131072 --disable-gpu-compositing --enable-blink-features=ShadowDOMV0 --lang=en-US --user-data-dir=/tmp/.com.google.Chrome.aFUGnx --disable-client-side-phishing-detection --num-raster-threads=4 --enable-main-frame-before-activation --renderer-client-id=7 --no-v8-untrusted-code-mitigations --shared-files=v8_context_snapshot_data:100
This chrome process was spawned by selenoid, when running UI autotests. As you can see, it mapped ~38GB of virtual memory (which is more that I have RAM on a server). On the other hand, it occupies only ~136MB of RAM. There's also other similar cases.
I think, a solution may be to give a choice for user, which memory to use for process sorting. Also default would be logical to change to RSS, IMHO.
In my vision, poll_mem is mostly used to determine, where your physical RAM leaks. VIRT memory doesn't show physical RAM occupation. To get process physical memory occupation, we should use RES memory.
To clearly illustrate, here is part of the log, showing chrome memory consumption:
This chrome process was spawned by selenoid, when running UI autotests. As you can see, it mapped ~38GB of virtual memory (which is more that I have RAM on a server). On the other hand, it occupies only ~136MB of RAM. There's also other similar cases.
I think, a solution may be to give a choice for user, which memory to use for process sorting. Also default would be logical to change to RSS, IMHO.