The cluster controller detection algorithm considers all the interfaces (and not only the managed interfaces).

Bug #1085823 reported by Raphaël Badin
6
This bug affects 1 person
Affects Status Importance Assigned to Milestone
MAAS
Critical
Raphaël Badin
1.2
Critical
Raphaël Badin

Bug Description

The problem with that is that the nodes can be attached to the wrong cluster controller because interfaces are autodetected by the packaging.

The utility method find_nodegroup, used to detect the origin of a request and attach it to a particular nodegroup (aka cluster controller), considers all the interfaces attached to a nodegroup; it should only take the managed interface into account.

Related branches

Raphaël Badin (rvb)
Changed in maas:
assignee: nobody → Raphaël Badin (rvb)
Revision history for this message
Julian Edwards (julian-edwards) wrote : Re: [Bug 1085823] [NEW] The cluster controller detection algorithm considers all the interfaces (and not only the managed interfaces).

FWIW it would be useful to state on this sort of bug what the
consequences are without this change, as it would help justify the priority.

Raphaël Badin (rvb)
description: updated
Raphaël Badin (rvb)
Changed in maas:
status: Triaged → In Progress
Changed in maas:
status: In Progress → Fix Committed
Changed in maas:
status: Fix Committed → Fix Released
Raphaël Badin (rvb)
description: updated
Revision history for this message
Raphaël Badin (rvb) wrote :

IIRC this was done to fix a problem with the testing lab network topology (but I don't think that topology is *that* specific so let's think this through before we revert the fix for this): the master nodegroup has an interface (interfaces are auto-populated) that actually connects to the separate cluster's network. Hence if we use all the interfaces (as opposed to only the managed interface) to do the match in find_nodegroup(), we would always get the master nodegroup, even for a node connecting from the separate cluster's network.

Revision history for this message
Gavin Panella (allenap) wrote :

> ... the master nodegroup has an interface (interfaces are
> auto-populated) that actually connects to the separate cluster's
> network.

The topology of the test lab might be the problem here. We envisaged
cluster controllers living on subnets of their own. There is no
support for more than one cluster controller on a subnet. We may have
mistakenly fixed this bug in MAAS when we should have instead fixed
the lab's network.

Revision history for this message
Raphaël Badin (rvb) wrote :

> The topology of the test lab might be the problem here.

That's possible but don't forget that the nodes need to be able to register themselves (and thus contact the region controller); so the network of the "separate" nodes (i.e. the nodes on the separate cluster) have to be connected to the region controller one way or another. I'd appreciate if you could look into it and bring a fresh point of view on this.

To post a comment you must log in.
This report contains Public information  Edit
Everyone can see this information.

Other bug subscribers