Node name rule relaxation

[Michael Davies]

It's been suggested, as part of the resolution of bug #1433832 that the logical name of a node should be allowed to be a free-form string, rather than needing to be restricted to an internet FQDN hostname.

The current set of restrictions were what were approved in the logical name spec (https://github.com/openstack/ironic-specs/blob/master/specs/kilo/logical-names.rst) but we may decide to rethink that.

[Michael Davies (mrda)]

If this behaviour is to be changed, please update the spec so that it's consistent with the implementation.

[Michael Davies (mrda)]

After discussing this in IRC today, I'm suggesting opening it to string.letters + string.digits + string.punctuation

[Julia Kreger (juliaashleykreger)]

I do have a bit of an objection from the standpoint that a system name should not be utilized as a description, and that support of punctuation may lead to the name becoming more of a description. I think the best path is a relatively free form string field where a user can include any descriptive text about the host.

[Julia Kreger (juliaashleykreger)]

I'm withdrawing my objection given that nova has relaxed name requirements.

[aeva black (tenbrae)]

Quoting from the relevant Nova spec:

"Fields that have legitimate reasons to be restricted can still impose further restrictions. For instance, hostnames can be restricted per RFC 1178, which is a strict subset of the restrictions described above."

I do not immediately agree that Node names should be relaxed, because they currently are matched against, and likely to be treated by operators as, hostnames. I believe we need to discuss whether this change should be done or not.

[aeva black (tenbrae)]

Keep in mind that Node Name identifiers are used in the URL, and so must either adhere to RFC 3986, or be escaped such that they do.

Reference: https://tools.ietf.org/html/rfc3986#section-2.2

I am changing this bug to "Wishlist" as it is not preventing nor related to any loss of intended functionality.

[OpenStack Infra (hudson-openstack)]

Change abandoned by Chris St. Pierre (<email address hidden>) on branch: master
Review: https://review.openstack.org/195138
Reason: Per Devananda, we're keeping logical names hostnames.

[Lucas Alvares Gomes (lucasagomes)]

Right, I do think we should relax the name I don't see reason why they should be forced to be restricted as a hostname if they are not going to be set as the hostname for the instance on that node. If someone wants to treat the nodes name as hostnames, fair enuff, with free form name field you can do that too.

I could even suggest to us to allow operators to give a regex in the config file where names could be matched against if they want to really restrict it.

But I find it odd that by default you can't create a node called "node_1" because "_" is not supported. Or, names like 'A' * 64 will failed because after the 63 character it's expected to have a '.' (period) and then you can have more 63 charactrers and another period and so on...

[Gabriel Assis Bezerra (gabriel-bezerra)]

This topic was discussed on 2015-06-29 meeting. See minutes and log on the links below:

Meeting ended Mon Jun 29 18:00:39 2015 UTC. Information about MeetBot at http://wiki.debian.org/MeetBot . (v 0.1.4)
Minutes: http://eavesdrop.openstack.org/meetings/ironic/2015/ironic.2015-06-29-17.01.html
Minutes (text): http://eavesdrop.openstack.org/meetings/ironic/2015/ironic.2015-06-29-17.01.txt
Log: http://eavesdrop.openstack.org/meetings/ironic/2015/ironic.2015-06-29-17.01.log.html

[Michael Davies (mrda)]

From the meeting, it appears the requirement is now https://www.ietf.org/rfc/rfc3986.txt compliance.

[Ruby Loo (rloo)]

patch: https://review.openstack.org/193587