Dear Nigel, Job and others I think this may be a bit over cautious for this object type. The MNTNER object is not created, update or queried as often as operational data. And it is one of the least likely objects to be subject to automated scripts creating or querying & fully parsing it. Because of the hidden password hash, most MNTNER objects can only be fully accessed by an authenticated lookup in Webupdates. This takes care of issues surrounding a missing, previously mandatory, attribute. So for this object maybe we can: -drop referral-by to an optional attribute with warning if used. Of course it will follow our now standard deployment process. -send the reminders to the DB WG after 1 and 2 months -deprecate the attribute fully after 2 months Regards Denis Walker Business Analyst RIPE NCC Database Team On 17/03/2014 17:17, Nigel Titley wrote:
Denis
I think your proposal is good and cautious (perhaps over-cautious, but experience has taught us that cautious is good). There should definitely be roll out on the sandbox systems at each stage to allow testing in advance. Folk use the DB for strange things and in strange ways.
Nigel
-----Original Message----- From: db-wg-bounces@ripe.net [mailto:db-wg-bounces@ripe.net] On Behalf Of Job Snijders Sent: 17 March 2014 15:53 To: Denis Walker Cc: db-wg@ripe.net Subject: Re: [db-wg] Proposal to remove "referral-by" attribute in "mntner" object or make it optional
On Mon, Mar 17, 2014 at 01:24:02PM +0100, Denis Walker wrote:
Historically, technical changes like this to the RIPE Database, that are not considered to affect any existing policy or require a new policy (at the discretion of the WG chairs), have been discussed on the DB WG mailing list or at the DB session at a RIPE Meeting. If the DB WG chairs agree that a consensus is reached by the community as a result of these discussions the WG chairs can approve the RIPE NCC taking the action. Based on the feedback from this group it seems a good idea start dialogue about phasing out "referral-by" in one way or another. So far we have not heard anything positive about referral-by and no statements supporting its continued existance, I personally would not mind getting rid of this attribute.
DB-WG (and RIPE DB Staff), what, in your vision is the least destructive way is to deprecate a mandatory attribute? I've compiled some things that came to mind, between each step a pause of sorts can be applied.
phase 0) reach consensus in some way (our process now)
phase 1) Change status from mandatory to optional, no errors are thrown in any scenario, whether MNTER objects contain the referral-by attribute or not.
phase 2) Have API/Mailrobot/etc throw soft warning (but accept data) when attribute is present when updating an existing MNTER object.
phase 3) Omit the attribute when presenting data to whois/api clients and omit the attribute in the webupdates interface.
phase 4) Deny creation of _new_ MNTER objects containing a referral-by attribute: throw hard error.
phase 5) Deny updating of _existing_ MNTER objects with a referral-by attribute: throw hard error.
phase 6) delete all referral-by attributes from database (However, still show the attribute when clients request old versions with 'diff').
Thoughts?
Kind regards,
Job
-