andreas/vanlink
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

apply-network: reload networkd after each netplan apply; post-check flags a networkd/NM double claim

Browse Source 
Live R5C run: after the Armbian catch-all .network was removed, networkd kept its
in-memory claim on the uplink port (two addresses, extra proto dhcp default route)
until `networkctl reload`. The helper now reloads networkd after the forward apply
and after the rollback re-apply (not fatal, not gating; logs the port state and a
"networkd still manages" note on the forward path), the AN_* time budget includes
it, and --post-check warns when the netplan matches but networkd still lists the
uplink port configured/configuring.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KqhjP7nud5gdBg25zCiRk9
This commit is contained in:
andreas
2026-09-23 20:58:29 -04:00
co-authored by Claude Sonnet 5
parent 1ce4a30336
commit 4f62b9f770
3 changed files with 84 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files
deploy.sh
+32 -4
View File
@@ -386,6 +386,16 @@ PY
unreadable) echo " ok WAN netplan: cannot read as non-root (run with sudo to verify)" ;;
*) warn "post-deploy: WAN netplan $f — run ./deploy.sh --apply-network" ;;
esac
# k2. once the WAN netplan is applied NM owns the uplink port; a networkd that still lists it
# configured/configuring holds a stale in-memory claim (cured by networkctl reload).
# Before the apply networkd legitimately owns it: say nothing.
if [ "$f" = matches ] && [ -e "$sysnet/$STARLINK_IFACE" ] && command -v networkctl >/dev/null 2>&1; then
out=$(networkctl --no-pager --no-legend list 2>/dev/null | awk -v i="$STARLINK_IFACE" '$2 == i { print $5; exit }') || out=
case "$out" in
configured|configuring) warn "post-deploy: networkd still manages $STARLINK_IFACE although NetworkManager owns it — run: networkctl reload" ;;
unmanaged) echo " ok networkd leaves $STARLINK_IFACE to NetworkManager" ;;
esac
fi
# l. last --apply-network
as_f=${VANLINK_APPLY_STATUS:-/var/lib/vanlink/apply-network.status}
if [ -r "$as_f" ]; then
@@ -402,9 +412,11 @@ PY
# --apply-network time budgets (seconds), in one place; all derived from the AN_* constants.
# (netplan generate/apply run under timeout -k AN_KILL_T: +AN_KILL_T each in the worst case)
# forward path AN_FWD = NM conf reload + generate + apply + 2*kill + van-br0-setup (unit check + restart)
# networkd reload AN_ND = networkctl reload (timeout AN_RELOAD_T) + settle sleep AN_SETTLE + port state line (AN_NM_T);
# runs once after each netplan apply (forward path and rollback)
# forward path AN_FWD = NM conf reload + generate + apply + 2*kill + AN_ND + van-br0-setup (unit check + restart)
# + stage 1 + restart + stage 2 + networkd info line + diagnostics (logged once, on a failed gate)
# rollback path AN_RB = generate + apply + 2*kill + NM conf reload + van-br0-setup (unit check + restart) + restart + connectivity recheck
# rollback path AN_RB = generate + apply + 2*kill + AN_ND + NM conf reload + van-br0-setup (unit check + restart) + restart + connectivity recheck
# helper worst case AN_HELPER_MAX = AN_FWD + AN_RB + AN_SLOP (loop iterations overshoot their
# deadline by a few nmcli/ping timeouts)
# foreground waits AN_WAIT = AN_HELPER_MAX + 30
@@ -414,9 +426,11 @@ PY
# route in its first loop, so a healthy default route shows up within seconds of a restart; the
# 120 s stage 2 deadline is a wide margin for a slow Starlink/cellular probe.
AN_KILL_T=10 AN_APPLY_T=90 AN_GEN_T=30 AN_NM_T=5 AN_RESTART_T=30 AN_S1=60 AN_S2=120 AN_RC=60 AN_SLOP=60
AN_RELOAD_T=10 AN_SETTLE=3
AN_ND=$(( AN_RELOAD_T + AN_SETTLE + AN_NM_T ))
AN_DIAG=$(( 5 * AN_NM_T )) # diag(): five commands, each under timeout AN_NM_T
AN_FWD=$(( AN_NM_T + AN_GEN_T + AN_APPLY_T + 2 * AN_KILL_T + AN_NM_T + AN_RESTART_T + AN_S1 + AN_RESTART_T + AN_S2 + AN_NM_T + AN_DIAG ))
AN_RB=$(( AN_GEN_T + AN_APPLY_T + 2 * AN_KILL_T + AN_NM_T + AN_NM_T + AN_RESTART_T + AN_RESTART_T + AN_RC ))
AN_FWD=$(( AN_NM_T + AN_GEN_T + AN_APPLY_T + 2 * AN_KILL_T + AN_ND + AN_NM_T + AN_RESTART_T + AN_S1 + AN_RESTART_T + AN_S2 + AN_NM_T + AN_DIAG ))
AN_RB=$(( AN_GEN_T + AN_APPLY_T + 2 * AN_KILL_T + AN_ND + AN_NM_T + AN_NM_T + AN_RESTART_T + AN_RESTART_T + AN_RC ))
AN_HELPER_MAX=$(( AN_FWD + AN_RB + AN_SLOP ))
AN_WAIT=$(( AN_HELPER_MAX + 30 ))
AN_RUNMAX=$(( AN_HELPER_MAX + 60 ))
@@ -497,6 +511,18 @@ gate1() {
[ -n "\$g" ] && [ "\$g" != -- ] || return 1
ping -c1 -W2 -I '$STARLINK_IFACE' "\$g" >/dev/null 2>&1
}
nd_reload() { # nd_reload [note]: after a netplan apply networkd may keep a removed file's config in memory; reload it (never fatal, never gating)
command -v networkctl >/dev/null 2>&1 || return 0
timeout $AN_RELOAD_T networkctl reload || echo "networkctl reload failed (continuing)"
sleep $AN_SETTLE
local l
l=\$(timeout $AN_NM_T networkctl --no-pager --no-legend list 2>/dev/null | awk -v i='$STARLINK_IFACE' '\$2 == i { print; exit }')
echo "networkd after reload: \${l:-$STARLINK_IFACE not listed}"
case " \$l " in
*" configured "*|*" configuring "*) [ -z "\${1:-}" ] || echo "networkd still manages $STARLINK_IFACE (NetworkManager should own it)" ;;
esac
return 0
}
route_log() { echo "route to 1.1.1.1: \$(ip route get 1.1.1.1 2>&1 | head -1)"; }
in_restore=0
restore() { # restore <reason>: put the pre-run netplan back, re-check connectivity, always exit 1 (runs once)
@@ -519,6 +545,7 @@ restore() { # restore <reason>: put the pre-run netplan back, re-check connecti
[ ! -f "\$nd/10-dhcp-all-interfaces.yaml.vanlink-disabled" ] || mv "\$nd/10-dhcp-all-interfaces.yaml.vanlink-disabled" "\$nd/10-dhcp-all-interfaces.yaml"
fi
timeout -k $AN_KILL_T $AN_GEN_T netplan generate && timeout -k $AN_KILL_T $AN_APPLY_T netplan apply || { [ "\$rb" != ok ] || rb=APPLY-FAILED; }
nd_reload # the old config is back: networkd's in-memory state changes the other way
[ ! -f "\$ncd/$pc" ] || timeout $AN_NM_T nmcli general reload conf || echo "nmcli general reload conf failed"
# the re-apply restarted networkd too, which can drop br0's members (as on the forward path; not gating)
if timeout $AN_NM_T systemctl cat van-br0-setup >/dev/null 2>&1; then
@@ -567,6 +594,7 @@ else
fi
s1=0
if [ "\$applied" = 1 ]; then
nd_reload note # the Armbian catch-all .network is gone from disk but networkd still holds it until reloaded
# netplan apply restarts networkd, which can drop br0's members: re-run the bridge setup (not gating)
if timeout $AN_NM_T systemctl cat van-br0-setup >/dev/null 2>&1; then
timeout $AN_RESTART_T systemctl restart van-br0-setup || echo "van-br0-setup restart failed (continuing)"