51 lines
2.6 KiB
Plaintext
51 lines
2.6 KiB
Plaintext
|
# Installed to /etc/systemd/system/ppp@pppoe0.service.d/10-vrrp-wan-gate.conf
|
||
|
|
#
|
||
|
|
# This drop-in is the ONLY thing preventing both routers from dialling the one
|
||
|
|
# ISP credential at the same time. Do not remove it without reading this.
|
||
|
|
#
|
||
|
|
# pppoe0 is configured identically and ENABLED on both routers, because the
|
||
|
|
# alternative -- `set interfaces pppoe pppoe0 disable` -- unlinks
|
||
|
|
# /etc/ppp/peers/pppoe0 (interfaces_pppoe.py treats `disable` and `delete`
|
||
|
|
# identically), and pppd's options file IS that path. A promotion then had to
|
||
|
|
# re-render it via a full config commit at priority 322, where one unrelated
|
||
|
|
# invalid node fails the whole commit and takes the 10 gig down with it. It also
|
||
|
|
# made op-mode `connect interface pppoe0` unusable, since that refuses when the
|
||
|
|
# peers file is absent.
|
||
|
|
#
|
||
|
|
# With the node enabled, interfaces_pppoe.py's apply() does this on EVERY commit
|
||
|
|
# that touches the pppoe subtree:
|
||
|
|
#
|
||
|
|
# if not is_systemd_service_running('ppp@pppoe0.service') or shutdown_required:
|
||
|
|
# call('systemctl restart ppp@pppoe0.service')
|
||
|
|
#
|
||
|
|
# -- i.e. the backup actively tries to dial whenever anything commits. A
|
||
|
|
# `pulumi up`, a `sim-net-apply.sh apply`, or the boot-time config load are all
|
||
|
|
# that commit. This gate is what makes that a no-op.
|
||
|
|
#
|
||
|
|
# /run is tmpfs, so the gate is shut at boot on both boxes and neither can dial
|
||
|
|
# before VRRP has decided. ppp@.service is already After=vyos-router.service, so
|
||
|
|
# no extra ordering is needed.
|
||
|
|
[Unit]
|
||
|
|
# Both must hold; multiple ConditionPathExists are ANDed.
|
||
|
|
# may-dial -- vrrp-wan-reconcile has blessed this box (a renewed lease)
|
||
|
|
# /etc/ppp/peers -- refuse to start pppd against a missing options file, which
|
||
|
|
# is what produced a restart loop of 47 and counting on
|
||
|
|
# 2026-09-05. A failed Condition is NOT a failure: the job
|
||
|
|
# succeeds, the unit stays inactive, and `systemctl start`
|
||
|
|
# exits 0 -- so callers must check is-active, never rc.
|
||
|
|
ConditionPathExists=/run/vrrp-wan/may-dial
|
||
|
|
ConditionPathExists=/etc/ppp/peers/pppoe0
|
||
|
|
|
||
|
|
# Belt to that brace. The stock unit is Restart=on-failure/RestartSec=5s against
|
||
|
|
# systemd's default StartLimitIntervalSec=10s/Burst=5 -- two restarts per window,
|
||
|
|
# so the limiter can never trip and a doomed pppd retries for ever.
|
||
|
|
StartLimitIntervalSec=600
|
||
|
|
StartLimitBurst=6
|
||
|
|
|
||
|
|
[Service]
|
||
|
|
RestartSec=15
|
||
|
|
# A hung pppd must be resolved inside the failover budget. The stock 90s means a
|
||
|
|
# demoted router could still hold the session while the new master is dialling.
|
||
|
|
# 20s still allows a clean LCP Terminate + PADT in the normal case.
|
||
|
|
TimeoutStopSec=20
|