yahoo-eng-team team mailing list archive
-
yahoo-eng-team team
-
Mailing list archive
-
Message #88924
[Bug 1974173] [NEW] Remaining ports are not unbound if one port is missing
Public bug reported:
As part of the instance deletion process, we must unbind ports
associated with said instance. To do this, we loop over all ports
currently attached to an instance. However, if neutron returns HTTP 404
(Not Found) for any of these ports, we will return early and fail to
unbind the remaining ports. We've seen the problem in the context of
Kubernetes on OpenStack. Our deinstaller is brute-force, so it deletes
ports and servers at the same time, so a race means the port can get
deleted early. This normally wouldn't be an issue as we'd just "untrunk"
it and proceed to delete it. But that won't work for SR-IOV ports as in
that case you cannot "untrunk" bound ports.
The solution here is obvious: if we fail to find a port, we should
simply skip that and continue unbinding everything else.
** Affects: nova
Importance: Medium
Assignee: Stephen Finucane (stephenfinucane)
Status: Confirmed
** Tags: neutron
** Tags added: neutron
** Changed in: nova
Assignee: (unassigned) => Stephen Finucane (stephenfinucane)
** Changed in: nova
Importance: Undecided => Medium
** Changed in: nova
Status: New => Confirmed
--
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to OpenStack Compute (nova).
https://bugs.launchpad.net/bugs/1974173
Title:
Remaining ports are not unbound if one port is missing
Status in OpenStack Compute (nova):
Confirmed
Bug description:
As part of the instance deletion process, we must unbind ports
associated with said instance. To do this, we loop over all ports
currently attached to an instance. However, if neutron returns HTTP
404 (Not Found) for any of these ports, we will return early and fail
to unbind the remaining ports. We've seen the problem in the context
of Kubernetes on OpenStack. Our deinstaller is brute-force, so it
deletes ports and servers at the same time, so a race means the port
can get deleted early. This normally wouldn't be an issue as we'd just
"untrunk" it and proceed to delete it. But that won't work for SR-IOV
ports as in that case you cannot "untrunk" bound ports.
The solution here is obvious: if we fail to find a port, we should
simply skip that and continue unbinding everything else.
To manage notifications about this bug go to:
https://bugs.launchpad.net/nova/+bug/1974173/+subscriptions
Follow ups