Minutes of the Network Configuration (NETCONF) WG Session at the IETF #75, TUESDAY, July 28 2009, 1300-1500: WG Chairs: Bert Wijnen Mehmet Ersue AD: Dan Romascanu Many Thanks to the minute takers: Juergen Schoenwaelder and Ladislav Lhotka and the jabber scribe: David Partain (also for channeling jabber room attendees on microphone). There were approx. 37 persons in the NETCONF session. * Administrivia and agenda bashing - see slides: http://www.ietf.org/proceedings/75/agenda/netconf.txt - no changes * WG Status (Mehmet) - WG status reviewed. - see slides: http://www.ietf.org/proceedings/75/slides/netconf-0.ppt - partial-lock I-D is now in the IESG evaluation queue - monitoring I-D: WGLC ended 24 July, WG should decide whether -07 is ready for AD review - robust conf. management created some interest, WG needs to decide about rechartering - no discussion * NETCONF Monitoring (Martin) - Few updates since last version, I-D quite stable, should be ready for WGLC. - see slides: http://www.ietf.org/proceedings/75/slides/netconf-1.ppt - JS asks whether the YANG module's dependence is an issue - ME answers that the appendix is non-normative so it is not an issue - JS asks for some additional time to check the that IP address types are consistent with yang-types - ME says this check is welcome - BW asks whether the XSD was autogenerated from YANG? - MB says no because it is a standalone tool and XSD has its own complex type for IP addresses. - DR Is YANG intended to be eventually normative? - General consensus confirmed that intention. - JS This depends on how soon YANG becomes a standard. - BW Now YANG is mature so that there is no real danger of delaying the monitoring draft. - DR I wouldn't be that optimistic. - AB(on Jabber): How can the monitoring draft precede the partial-lock draft? - BW partial-lock is now at IESG. - BW Do we want to make YANG normative? - DR Why is it so important? - PS (i) the data model is nicer, (ii) it's a useful exercise for YANG, (iii) YANG has better data types. - MS Do we want to wait with other data models until YANG is finished? - PS asks whether we can wait and make the YANG normative? - DR says you can progress the document and it will stay in the RFC editor queue until the YANG documents come out - BW explains that there is of course a risk in case YANG changes affecting the document can cause trouble - ME as contributor says we should not delay drafts already in queue - DP are 1-2 months delay worth to publish a data model in a format we do not plan to continue? - BW agrees to use YANG as a technical contributor, but as chair he is reluctant to tell the authors things get further delayed - BW suggests to move ahead with XSD and then update the RFC quickly with a YANG version of the document - BW asks whether there are any strong objects against moving forward with the normative XSD. - There were non raised so we move forward with the normative XSD after checking the IP address data types. * RFC 4741 Revision (Martin) - see slides: http://www.ietf.org/proceedings/75/slides/netconf-2.pdf ** 003: error-path - AB Sometimes we have to use both roots in one PDU. - MB will propose a solution on the WG list ** 004: error-severity - BL: Warnings are useful, do not remove, rather fix them. - JS Do we have any evidence that warnings are useful. - MB No, because they cannot be currently used. - PS says Junos does support warnings, but in a proprietary way - WH Warnings are useful, if you remove them now, be sure you will put them back later. - DP agrees. - BW asks who prefers to keep and fix warnings. - Rough consensus is to keep warnings but also fix them. - AB Will the new error tag break the existing managers? - BW Let's discuss it in the ML. ** 006: multiple namespaces - MB prefers to remove this feature - PS says this was added to support configs that are just a text block - PS The outcome (selecting the output format) is useful, so we should maybe introduce a new get-config parameter for the output format. - AB Isn't this data model specific? - MB Current text isn't. - BW We need to clarify the text. - JS says that YANG has features and NETCONF has capabilities and we should work with these mechanisms - PS YANG model can generate both XML and text (CLI output). - AB says all content is XML and text should be wrapped in XML - JS supports removing the text output format option. - AB Text format is not defined and we have no formal way for XML<->text conversions. - BL We already express features in the capability URI, we can do it for the format as well. - BW asks whether the people in the room are in favour of removing the text and that was the case. Bert sends the discussion back to the mailing list. ** 014: capability changes - AB says that the base capability version must be increase when adding new error-tags, otherwise the existing managers will break. - BL We have already added new error-app-tags. - PS wants to have text that clients should not break if error-tags are added in the future - AB asks whether a capability changed notification could work? - BW responds that a client might not listen for such a notification - MB The manager may not be subscribed to the right channel. - LL Could one channel be made mandatory for all managers? - MB says a notification would be in addition to the error code - AB is against the error-code. - Mark E (On Jabber): Resulting capabilities are the intersection of capabilities advertised by the client and server, so they must be renegotiated if either part changes them. - BL Not true, clients are not supposed to send everything they support. - BW asks whether the room prefers to add a capability-changed and there was rough consensus in the room to add capability-changed ** 015: UTF-8 and UTF-16 - DR says UTF-8 only is fine - JS proposes to make UTF-8 mandatory and to be silent about UTF-16 - BW polls whether adding text requiring UTF-8 has consensus in the room and this seems to be the case ** 013: confirmed-commit - WH explains that there is also a lock dropped when a session disappears - JS What's the relation between locks and confirmed-commit? - WH further clarifies that a manager waiting for a lock will grab the lock and be surprised if there is suddenly a rollback - BC asks what the problem is to be solved? - WH You can tie the lock to the confirmed-commit operation and release it only after it is finished. - JS says to solve the issue WH raised would be to not hand out a lock while a confirmed commit is pending - BW says this needs more though and discussion on the mailing list ** 010: namespace wildcarding - MB is in favour of treating no namespace as a wildcard - BL says this wildcarding is often used and he likes it - AB says this does break XML [AB amended on the maillist that he does not object to the wildcarding and it does not break XML]. - MB disagrees. - WH sees the valid of the feature but is concerned about the way it is done. This approach breaks the concept of XML namespaces. - PS says the client should be explicit that it wants a wildcard - MB and PS discuss things... - LL says the null namespace could work for this. It's nothing illegal from the point of view of XML namespaces, if the client makes sure that the namespace is properly removed (by 'xmlns=""'). The only problem would be if there are other elements with null namespace URI, but this cannot happen in NETCONF. - MB agrees. - BW says that this issue requires further discussion on the list. ** 009: partial-operation error - AB agrees to deprecate this error - BW says this will go to the mailing list ** 012: format of copy-config - PS says that Juniper is not using nc:config but instead a vendor specific top-level node - JS I am still confused how this works. [Question to PS] Is the only problem the top-level element? - BL asks whether we should state that the contents the same as get-config without a filter? - BW says this issue needs more mailing list discussion. * With-defaults (Balasz) - see slides: http://www.ietf.org/proceedings/75/slides/netconf-3.ppt - PS says that the text uses SNMP terminology "agent-manager", it should use "client-server" and wants to change terminology to 4741 terminology. - PS has some detailed questions concerning the text in the ID - BL This terminology is used in other drafts, too. - Phil: [expresses another comment wrt Section 2.1.1] - BW asks PS to send the details to the list - BW Can we discuss these editorial comments during WGLC? - DP Yes, these items are suitable for WGLC. - AB wrote to we need to describe what happens for copy-config to the candidate (set to trim or explicit); proposal: with-defaults ignored for target= candidate - AB copy-config on candidate should be ignored. - AB We also need to address the issues with :startup. - MB has similar questions concerning copy-config between data stores - BL says the capability should not affect copy-config between data stores and there is already text in the draft - BW polls the room and there is no objection to move this document to WG last call * Robust Configuration (Bob Cole) - see slides: http://www.ietf.org/proceedings/75/slides/netconf-4.pdf - Second bullet on slide 2 should be "Verification". - BW asks how many people have read the latest document (6-7) - BW asks how many people think that this work should be included in the WG charter? - about 3 people do - PS kind of objects since he does not see the value of putting the test on the box - PS Why should the device run tests instead of manager? - BC Sometimes the manager may not be able to run them, e.g. the device may get out of reach. - DR Yes, the devices sometimes have a level of autonomy. But we need to provide better use cases. - JS who is planning to implement this? - AB does plan it. - MB has issues with some design decisions taken. The tests should not be tied to the commit operation, separate RPCs should be used. - AB is planning to implement - if only as an improvement to the flawed test; but verify tests should be runnable not during commit - AB Yes, RPCs will be used. - BW More discussion in the mailing list is needed before we can accept this work as a WG item. * Open Microphone - BW listed other topics for future discussion: - 4742 clarifications as bis document? - Do we need to specify a NETCONF database architecture? - NETCONF Access control. - The chairs will put the open mic questions to the WG list * Actors - Andy Bierman (AB) (via jabber) - Martin Bjorklund (MB) - Bob Cole (BC) - Mehmet Ersue (ME) - Mark Ellison (Mark E) (On Jabber): - Wes Hardaker (WH) - Balasz Lengyel (BL) - Ladislav Lhotka (LL) - David Partain (DP) - David Reid (DR) (via jabber) - Dan Romascanu (DR) - Juergen Schoenwaelder (JS) - Phil Shafer (PS) - Bert Wijnen (BW) - Mark Scott