Ballot for draft-ietf-cdni-request-routing-extensions
Yes
No Objection
Note: This ballot was opened for revision 07 and is now closed.
Thank you for addressing my DISCUSSes and COMMENTs.
Thank you for writing this, also thank you for addressing the OpsDir Review comments. Also thank you to wangzitao@huawei.com for the review.
Thank you for the work put into this document. I have a couple of comments and nits. Regards, -éric == COMMENTS == One generic suggestion (no need to act on a COMMENT): add the definition of "router" in this document as it is not the typical IP-layer router as in most IETF documents. -- Section 2.3 -- It is unclear to me whether the 3xx redirection (cfr last example) can switch from HTTP to HTTPS or vice-versa? Should the 'httptarget' include the scheme/protocol to be used ? In any case, the text should be clear whether the scheme is kept between original request and the scheme in the redirected location. Or is it specify in the OpenCache spec ? == NITS == -- Abstract -- Please expand FCI at first use. -- Requirements language -- Unusual place in the document. It will be fixed by the RFC Editor probably later in the process. -- Section 2 -- s/defintion/definition/, I will stop looking for typos, easier to use a spell checker ;)
I support Roman's DISCUSS.
I think it would be good to note in Section 5.1 that the way the redirect target capability is specified here potentially encourages further use of the EDNS client subnet option to identify the geographic area of a client, which has a number of known privacy drawbacks per RFC 7871 Section 2 and Section 11.1.
The Abstract and Introduction talk about Open Caching (and specifically about the Open Caching working group of the Streaming Video Alliance (SVA)) as a driving use case for this work...but there are no references to that work, including a description of what Open Caching is. Open Caching may be well known in the cdni WG and to the interested community, but not to the casual reader...like me. The document would benefit from referencing the background material.
Section 2 o Footprint: The dCDN may want to have a different target per footprint. Note that a dCDN may spread across multiple geographies. This makes it easier to route client requests to a nearby request router. Though this can be achieved using a single canonical name and Geo DNS, that approach has limitations; for We probably want a reference for "Geo DNS". A redirect target for DNS redirection is an IPv4 address used as an A record response, an IPv6 address used as an AAAA record response or a FQDN used as an alias in a CNAME record response (see [RFC1034]) of the uCDN DNS router. Note that DNS routers make routing decisions based on either the DNS resolver's IP address or the client IP address when EDNS0 client-subnet is used (see [RFC7871]). The dCDN may choose to advertise redirect targets and footprints to cover both cases. [...] We may want to say a bit more about how the dCDN's advertisements interact with the uCDN's DNS redirection decision here -- is the idea just that the set of redirect targets should try to be "close to" both the end-user and recursive resolver IP addresses that will be making requests of the uCDN? The following is an example of a Redirect Target capability object serialization that advertises a dCDN target address that is attached to a specific list of uCDN "redirecting-hosts". A uCDN host that is included in that list can redirect to the advertised dCDN redirect target. The capabilities object is serialized as a JSON object as defined in Section 5 of [RFC8008] To check my understanding: the Redirect Target capability object is a new object, in some sense a peer of the (e.g.) Delivery Protocol capability object, but encoded in a similar fashion as the objects specified in sections 5.3 through 5.7 of RFC 8008. In that case, I wonder if Section 5.2 or 5.1 is a better reference than the top-level Section 5 of RFC 8008. Section 2.2 Do we want an example with an IP address as well, to illustrate the full spectrum of possible behaviors? Section 2.3 My apologies for not doing much background reading, but what are the expected scenarios with respect to TLS usage? Is the redirect expected to preserve the URI scheme (http:// vs. https://) or could we go from one to the other as part of the redirect? (I do see that the example has an http:// redirection target.) I ask because there are probably some considerations here relating to the TLS server name indication, particularly for the case of redirection to an IP address. Section 2.4 It's quite possible I've confused myself, but in Figure 1 are the "redirecting-hosts" and "target host" swapped? The former is supposed to be an uCDN host, is it not? Section 3 o Error: A cache may receive a request that it cannot properly serve, for example, some of the metadata objects for that service were not properly acquired. In this case, the cache may resolve to redirect back to uCDN. nit: I think this last sentence would benefit from another word or two or some punctuation to clarify what/how the cache is resolving, e.g., quotes around "redirect back to uCDN" or "cache's action" or similar. The Fallback target metadata object is used to indicate the target address the dCDN should use in order to redirect a client back to the uCDN. Fallback target is represented as endpoint objects as defined in section 4.3.3 of [RFC8006]. nit: singular/plural mismatch "target"/"objects" The uCDN fallback target address may be used as a DNS A record, AAAA record or CNAME record in case of DNS redirection or a hostname for HTTP redirect. nit: I don't think "used as" is quite right (that is, the sense would apply in the other direction: a DNS A record might be used as a fallback target address) and could probably be removed. Section 3.2 [same comment about Figure 3 as for Figure 1 above] Section 5 The presence of a fallback uCDN host could make for a more tempting attack target -- if the fallback target is known, then an attacker could concentrate resources on that fallback host (more effectively DoS'd than the whole CDN) and thrash the cash on the dCDN to effect a DoS. This speaks to the need to keep the exchanged metadata confidential, as already noted, but is also a latent risk of the topology independently of keeping the metadata confidential. The attacks made possible by having the Redirect Target are mostly a subset of the generic risks listed in RFC 8006; by allowing per-host granularity of the uCDN behavior, there is perhaps a new opportunity to have degraded (but present) service, by directing requests to dCDN hosts that are "far away". This seems somewhat challenging for an active attacker to pull off, but may still be the result of misconfiguration. As I mentioned above, we probably need to say something about TLS usage for HTTP-level redirection, though I'm not sure yet what exactly that is, since I don't understand the expected usage yet.