mirror of
https://github.com/tailscale/tailscale.git
synced 2026-05-05 20:26:47 +02:00
fa4ad9d5d5
Without this signal, tsnet consumers (e.g. Aperture) cannot distinguish
a deleted node from a transient control-plane outage. Such a node will
sit in BackendState=Running and indefinitely wile mapRoutine has been
silently 404'ing on /machine/map every 30 seconds for 5 years.
With NodeNotFoundWarnable exposed on the IPN bus, the entire
consumer-side reaction collapses to a few lines in its existing notify
watcher:
for {
notify, err := watcher.Next()
if err != nil {
return
}
if notify.Health != nil {
_, deleted := notify.Health.Warnings[health.NodeNotFoundWarnable.Code]
if deleted && !nodeDeleted {
nodeDeleted = true
// surface to the operator (UI badge, alert, action, etc.)
}
}
}
A subsequent successful registration with a fresh node key clears the
warnable on its own, so re-authentication flows recover without any
extra plumbing on the consumer side. And no persistence is required: if
a consumer process crashes while the node is in this state, mapRoutine
re-hits the 404 within ~30s of restart and the warnable trips again.
Updates #19326
Signed-off-by: Luke Kosewski <lkosewsk@tailscale.com>