You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Wipe agent token on inspection completion to fix cleaning failures
The backport of f734efb ("Set node alive when inspection finished")
causes a regression in 4.15 where ironic-inspector is still used.
The patch makes nodes appear fast-trackable after inspection when the
agent is dead and boot media is detached, causing the
AgentConnectionFailed handler to park nodes in clean wait indefinitely.
This happens because 4.15 uses ironic-inspector, which powers off the
node after inspection. The f734efb fix was designed for 4.16 where
built-in agent inspection keeps IPA running across the
inspection-to-cleaning transition.
Revert 0feaa17 and instead fix the root cause: the agent_secret_token
persists from inspection into cleaning. When IPA reboots for cleaning,
it cannot get a new token because Ironic refuses to generate one when
one already exists, causing repeated lookup loops without heartbeats
until clean_callback_timeout fires.
The fix adds wipe_token_and_url() when inspection finishes successfully,
both in the synchronous path (inspect_hardware) and the
ironic-inspector callback path (continue_inspection). This follows the
same pattern already used on inspection start and abort.
0 commit comments