[regext] Benjamin Kaduk's Discuss on draft-ietf-regext-epp-fees-18: (with DISCUSS and COMMENT)

[Benjamin Kaduk via Datatracker <[e-mail]>]

Benjamin Kaduk has entered the following ballot position for
draft-ietf-regext-epp-fees-18: Discuss

When responding, please keep the subject line intact and reply to all
email addresses included in the To and CC lines. (Feel free to cut this
introductory paragraph, however.)


Please refer to https://www.ietf.org/iesg/statement/discuss-criteria.html
for more information about IESG DISCUSS and COMMENT positions.


The document, along with other ballot positions, can be found here:
https://datatracker.ietf.org/doc/draft-ietf-regext-epp-fees/



----------------------------------------------------------------------
DISCUSS:
----------------------------------------------------------------------

I think (at least with the present formulation) we need greater clarity
on when the "it's up to server policy whether to include, but that
policy must be the same for all transactions" elements (<fee:balance>
and <fee:creditLimit>) are returned, as at present there seems to be an
internal inconsistency in the text.  Section 3.5 and 3.6 just talk about
including them "in responses to all 'transform' or billable commands",
but then we have more qualified text such as (but not limited to) in
Section 5.2.5 that only has <fee:updData> (and its children) included
when the <update> has been processed successfully.  So, are
<fee:balance> and <fee:creditLimit> supposed to be included in error
responses or not?


----------------------------------------------------------------------
COMMENT:
----------------------------------------------------------------------

It might be helpful to note somewhere that <fee:cd> stands for "check
data", as per stock EPP, since we use the term a few times before we get
to the <fee:chkData> context and its definition.

Section 1

   Given the expansion of the DNS namespace, and the proliferation of
   novel business models, it is desirable to provide a method for EPP

It's not clear to me whether all readers (whether now or in ten years)
will have the context to appreciate what is meant by these background
clauses.

Section 3.3


   When querying for fee information using the <check> command, the
   <fee:period> element is used to indicate the units to be added to the
   registration period of objects by the <create>, <renew> and
   <transfer> commands.  This element is derived from the
   <domain:period> element described in [RFC5731].

The word "units" here is really confusing me.  Even after reading the
rest of the document (and 5731's definition of periodType) it still
feels like there's some words missing here.

Section 3.4

   A server MAY respond with multiple <fee:fee> and <fee:credit>
   elements in the same response.  In such cases, the net fee or credit
   applicable to the transaction is the arithmetic sum of the values of
   each of the <fee:fee> and/or <fee:credit> elements.  This amount
   applies to the total additional validity period applied to the object
   (where applicable) rather than to any incremental unit.

"unit" here is also confusing to me, though less so.  I think what's
going on here is just the common-sense "the sum of all fees/credits
applies for the conceptual 'sum' of all the indicated registry
operations taken together", in which case I might suggest to
s/incremental unit/individual component of the transaction/.

   description: an OPTIONAL attribute which provides a human-readable
   description of the fee.  Servers should provide documentation on the
   possible values of this attribute, and their meanings.  An OPTIONAL
   "lang" attribute MAY be present to identify the language of the
   returned text and has a default value of "en" (English).

I assume we're reusing the "lang" semantics from 5730 (which in turn
relies on 4646), but it's probably worth being explicit about it.

Section 3.4.3

   If a <fee:fee> element has a "grace-period" attribute then it MUST
   also be refundable and the "refundable" attribute MUST be true.  If
   the "refundable" attribute of a <fee:fee> element is false then it
   MUST NOT have a "grace-period" attribute.

If a client receives a response that contravenes these requirements,
what should it do?

Section 3.5, 3.6

For these elements that the server MUST include in all responses if it
chooses to include them in (any) responses, do we expect that to be
global server policy, or potentially tailored to individual clients?
(Also, I'm vaguely curious how much it could increase the response
footprint, not that XML is a terribly concise representation to start
with.)

Section 3.7

   If a server makes use of this element, it should provide clients with
   a list of all the values that the element may take via an out-of-band
   channel.  Servers MUST NOT use values which do not appear on this
   list.

I think we generally dislike to rely on out-of-band channels to quite
this extent, though it's not clearly wrong for this case.  I'm somewhat
curious (not necessarily to include in the document) what existing
out-of-band channels for this look like, though.

Section 4

   The server MUST return avail="0" in its response to a <check> command
   for any object in the <check> command that does not include the
   <fee:check> extension for which the server would likewise fail a
   domain <create> command when no <fee> extension is provided for that
   same object.

nit: this wording makes it sound like avail="0" is scoped to the object,
as opposed to the check data.  So maybe s/for any object/if any object/?

   If a server receives a <check> command from a client, which results
   in no possible fee combination but where a fee is required, the
   server MUST set the "avail" attribute of the <fee:cd> element to
   false and provide a <fee:reason>.

nit: I'm not sure how to interpret "where a fee is required" just given
what's in this document.

   If the currency or total fee provided by the client is less than the
   server's own calculation of the fee for that command, then the server
   MUST reject the command with a 2004 "Parameter value range" error.

How can a currency be "less than the server's own calculation"?  (I
assume this is supposed to be "currency is different".)

Section 5.1.1

   When the server receives a <check> command that includes the
   extension elements described above, its response MUST contain an
   <extension> element, which MUST contain a child <fee:chkData>
   element.  The <fee:chkData> element MUST contain a <fee:currency>
   element and a <fee:cd> for each element referenced in the client
   <check> command.

Can we be more precise about "for each element referenced in the client
<check> command"?  ("No" is a valid answer.)  Specifically, does this
apply to the <domain:check> child elements in the <check>, or to the
<fee:check> extension elements, or something else?  (My guess from the
examples is the former.)

   o  A <fee:command> element matching each <fee:command> (unless the
      "avail" attribute of the <fee:cd> if false) that appeared in the
      corresponding <fee:check> of the client command.  This element MAY
      have the OPTIONAL "standard" attribute, with a default value of
      "0" (or "false"), which indicates whether the fee matches the fee
      of the "standard" classification (see section 3.7).  This element
      MAY have the OPTIONAL "phase" and "subphase" attributes, which
      SHOULD match the same attributes in the corresponding
      <fee:command> element of the client command if sent by the client.

I don't think I see how the SHOULD could be applicable -- doesn't
Section 3.8 place tight requirements on server behavior and errors
regarding phase/subphase in requests?  That is, I think the descriptive
"will match" would be appropriate here.

   The <fee:command> element(s) MAY have the following child elements:

   o  An OPTIONAL <fee:period> element (as described in Section 3.3),
      which contains the same unit that appeared in the <fee:period>
      element of the command.  If the value of the preceding
      <fee:command> element is "restore", this element MUST NOT be
      included, otherwise it MUST be included.  If no <fee:period>
      appeared in the client command (and the command is not "restore")
      then the server MUST return its default period value.

I think we need some caveat language ("if present"?) for "the same unit
that appeared in the <fee:period> element of the command", since that
element is OPTIONAL in the command in question.
nit: is this the "preceding <fee:command> element" or "parent
<fee:command> element"?  Also, the rhetorical value of the "OPTIONAL" is
unclear, as there's no server choice in the matter -- its
presence/absence is fully determined by the <fee:command> value.

   If the "avail" attribute of the <fee:cd> element is true and if no
   <fee:fee> elements are present in a <fee:command> element, this
   indicates that no fee will be assessed by the server for this
   command.

   If the "avail" attribute is true, then the <fee:command> element MUST
   NOT contain a <fee:reason> element.

In this second quoted paragraph, is this the "avail" attribute only of
<fee:command> or does it apply to <fee:cd> as well?

Section 5.2.1

   The server MUST fail the <create> command if the <fee:fee> provided
   by the client is less than the server fee.

I think we are more specific about this ("Parameter value range" error)
in Section 4, which is also a MUST-level requirement.

It's perhaps a bit anachronous to have a domain-creation example from
1999 when the -00 of this document's precedessor wasn't until 2013.

Section 5.2.3

[The examples here are only from 2005; progress!]

Section 7

Thank you for addressing the points discussed in the secdir review.
That said, the text of this section still feels a bit sparse, with it
mostly being bare statements without much discussion of the motivation
for or consequences of many of the requirements at hand.  For example,
we could say something about why it's important to provide
confidentiality/integrity protection for financial data, say more about
what the "needed level of [...] protection" is, and reiterate that the
transport protocol has to do so because there are no in-band EPP
mechanisms to do so.  It would also be fine to reiterate any key
considerations from 5730/5731, if there are any that seem particularly
relevant (but it's also fine to not do so).

Also, I think that it's important to add "peer authentication" to the
list of protections provided by the transport -- it's important to know
who you're talking to when sending financial information!  (Though, the
client is just sending its estimate of the server's fee schedule, which
is a lot less sensitive than sending its current balance.)