From 20b7968e33ae0a667dc9018c3343ce670ca6e8fa Mon Sep 17 00:00:00 2001 From: YueGuobin Date: Mon, 5 Oct 2026 22:30:29 +0800 Subject: [PATCH] docs(bugs): note how to re-observe the baseline on an already-declared host --- docs/bugs/networkmanager-tap-release.md | 1 + 1 file changed, 1 insertion(+) diff --git a/docs/bugs/networkmanager-tap-release.md b/docs/bugs/networkmanager-tap-release.md index 473e1160f..885821ce3 100644 --- a/docs/bugs/networkmanager-tap-release.md +++ b/docs/bugs/networkmanager-tap-release.md @@ -141,6 +141,7 @@ How the numbers are produced: - 40 iterations; each is one **re-enslavement**: `link set up`, `brctl addif`, then membership sampled at ~1 kHz for 200 ms, then `brctl delif` + `link set down`. - A **release** is counted when `/sys/class/net//brif/` disappears inside that 200 ms window; the script prints the release latency after the join (the NM release lands at 58-87 ms). - The tap is kept between runs, so the A/B flip is single-variable on the *same* device; a fresh device would reset NM's assumption state (the first enslavement of a fresh tap is always stable). +- The numbers below were measured on a host **without** the declaration from *Fix*. On a host that already carries it the tap is unmanaged from birth and the loop prints `0/40` by design; to re-observe the baseline there, temporarily remove the conf.d file (`sudo rm /etc/NetworkManager/conf.d/99-gns3-datapath.conf && sudo systemctl reload NetworkManager`) or run on an un-declared host. Measured on an affected host (same code, five baseline runs, then the A/B between runs):