[{"author": "Job Snijders", "text": "<p>hello!</p>", "time": "2026-03-16T08:34:13.000Z"}, {"author": "Jeffrey Haas", "text": "<p>There's significant benefit for having common schemas across different layers.</p>", "time": "2026-03-16T08:47:15.000Z"}, {"author": "Tony Przygienda", "text": "<p>hmm, as noob question: isn't accepting RPKI data via BGP sessions a chicken-and-egg problem in terms of trust?</p>", "time": "2026-03-16T08:49:49.000Z"}, {"author": "Jeffrey Haas", "text": "<p>This is BMP, Tony.  So, this is an attempt to monitor the interactions for bgp in its pipeline and provide visibility into the interactions in the rpki inputs like the rpki-rtr protocol and how it gets used.</p>", "time": "2026-03-16T08:51:14.000Z"}, {"author": "Tony Przygienda", "text": "<p>ah, i'ts monitoring only. sorry, misunderstood then</p>", "time": "2026-03-16T08:51:57.000Z"}, {"author": "Jeffrey Haas", "text": "<p>Right now, BMP doesn't push state relating to <em>how</em> policy works, only its results.  Pushing rpki policy interactions and state may lead to a bit of a state explosion impacting the core use cases.</p>", "time": "2026-03-16T08:52:36.000Z"}, {"author": "Jeffrey Haas", "text": "<p>That said, there's a general push to expose more stuff.  For example, the TLV draft here.</p>", "time": "2026-03-16T08:53:08.000Z"}, {"author": "Job Snijders", "text": "<p><span class=\"user-mention\" data-user-id=\"422\">@Jeffrey Haas</span> i think you capture it very well: there is great interest, there arguably is need for visibility into what the machinery is doing, and there is great risk for state explosion</p>", "time": "2026-03-16T08:53:41.000Z"}, {"author": "Tony Przygienda", "text": "<p>and that was my observation bits of time ago that BMP starts to look like a \"generic DB dumping protocol\" looking from afar ;-)</p>", "time": "2026-03-16T08:53:43.000Z"}, {"author": "Pierre Francois", "text": "<p>Exactly :) Provide options to turn them off if your routers / collectors can't deal with it in your context.</p>", "time": "2026-03-16T08:54:04.000Z"}, {"author": "Job Snijders", "text": "<p>in this sense CCR was also born from friction that its hard to capture \"the database\" efficiently with pre-CCR tooling</p>", "time": "2026-03-16T08:54:29.000Z"}, {"author": "Jeffrey Haas", "text": "<p>That's part of the motivation for the \"filter\" drafts.  BMP is already the firehose of BGP down a drinking straw.  Now we're trying to push more into the straw.</p>", "time": "2026-03-16T08:54:48.000Z"}, {"author": "Job Snijders", "text": "<p>soon there will be FPGA-assisted BMP offload cards! ;-)</p>", "time": "2026-03-16T08:55:28.000Z"}, {"author": "Tony Przygienda", "text": "<p>FME \"push\" style DB architectures (triggers and such in SQL ;-) tend to die terrible deaths at scale ;-)</p>", "time": "2026-03-16T08:55:47.000Z"}, {"author": "Tony Przygienda", "text": "<p>or in other words, that's why stuff like Kafka are really message busses and not databases</p>", "time": "2026-03-16T08:56:45.000Z"}, {"author": "Jeffrey Haas", "text": "<p>The exercise is the same packing efficiency work that bgp tries to do with pathattr + nlri-set.  TLV lets you get per-NLRI annotation.</p>", "time": "2026-03-16T08:56:50.000Z"}, {"author": "Tony Przygienda", "text": "<p>and putting massive DB state streaming on TCP by piggy backing e'thing on BMP is another layer of headache by inviting head blocking on state updates. but it's nothing new compared to std bgp ;-)</p>", "time": "2026-03-16T08:59:03.000Z"}, {"author": "Jeffrey Haas", "text": "<p>yeah. exact same state compression headaches.  it impacts lovely things like the timestamps just mentioned</p>", "time": "2026-03-16T08:59:54.000Z"}, {"author": "Pierre Francois", "text": "<p>Could we remove the \"SHOULD be ordered\", and let the vendors who prefer to order them, order them? Collectors will have to deal with those not respecting the SHOULD, anyway.</p>", "time": "2026-03-16T09:00:31.000Z"}, {"author": "Jeffrey Haas", "text": "<p>The things we really care about are the things that impact parsing.  Don't make the parser go out of order to find those things before you try to pull other fields apart.</p>", "time": "2026-03-16T09:01:16.000Z"}, {"author": "Job Snijders", "text": "<p>think of TCP: out-of-order packet delivery is expensive to reassemble on the receiver side</p>", "time": "2026-03-16T09:04:17.000Z"}, {"author": "Jeffrey Haas", "text": "<p>The flowspec use cases are overdue - although the question is \"do this in BMP or in yang?\"  It's the usual \"you want to do stuff in the ecosystem you have\".</p>\n<p>That said, I'll be flagging the utf-8 field in a comment if we have questions tiem</p>", "time": "2026-03-16T09:08:33.000Z"}, {"author": "Boris Khasanov", "text": "<p>i think some alignment is needed in regards of SR Policy events reporting between BMP and BGP-LS current work, IMO</p>", "time": "2026-03-16T09:10:27.000Z"}, {"author": "Jeffrey Haas", "text": "<p>The general work is \"we need a general channel from the routers to management stations\".  </p>\n<p>We <em>used</em> to call this syslog.</p>", "time": "2026-03-16T09:11:12.000Z"}, {"author": "Tony Przygienda", "text": "<p>give external tooling a free form text string (utf-8 a bonus) and see whether you can start an injection attack that way ;-)</p>", "time": "2026-03-16T09:11:24.000Z"}, {"author": "Jeffrey Haas", "text": "<p>in snmp-land, the fix is LL-IPv6 addresses are 20 octets long.</p>", "time": "2026-03-16T09:23:45.000Z"}, {"author": "Pierre Francois", "text": "<p>Not sure we understood your point Jeff</p>", "time": "2026-03-16T09:26:07.000Z"}, {"author": "Pierre Francois", "text": "<p>(thanks for the clarification, Jeff)</p>", "time": "2026-03-16T09:28:01.000Z"}, {"author": "Job Snijders", "text": "<p>yup, thanks</p>", "time": "2026-03-16T09:28:43.000Z"}, {"author": "Job Snijders", "text": "<p>wont it confuse existing parsers though?</p>", "time": "2026-03-16T09:29:23.000Z"}, {"author": "Jeffrey Haas", "text": "<p>Mostly the push is that anywhere we advertise \"addresses\", it's best if the representation across the encodings uses the \"wide\" form where the ifindex is part of the address.  otherwise, some consumer may treat it severably and lose it</p>", "time": "2026-03-16T09:29:55.000Z"}, {"author": "Jeffrey Haas", "text": "<p>it'll require a different flavor of flag. end of the day, we have to extend the encoding. it's a question of how that extension gets represented</p>", "time": "2026-03-16T09:31:04.000Z"}, {"author": "Changwang Lin", "text": "<p>However, on the BMP producer (i.e., the device), the BGP local interface peer actually also uses the interface index as part of the peer identification.</p>", "time": "2026-03-16T09:34:42.000Z"}, {"author": "Jeffrey Haas", "text": "<p>It's where moving to smarter addresses is helpful. if address is 16 bytes long, it's non-LL IPv6; 20 LL.  If you need the ifindex severably, it's for an application that usually isn't peer ip.</p>", "time": "2026-03-16T09:36:47.000Z"}, {"author": "Pierre Francois", "text": "<p>Is the plan to disable route monitoring and only get the stats?? Or else you can produce these stats from what you collect, can't you?</p>", "time": "2026-03-16T09:41:25.000Z"}, {"author": "Changwang Lin", "text": "<p>@jeffrey: This requires different link-local addresses. During automated deployment, the IPv6 link-local addresses of all interfaces may be identical. Additionally, the same issue exists for IPv4 unnumbered interface peers.</p>", "time": "2026-03-16T09:43:24.000Z"}, {"author": "Jeffrey Haas", "text": "<p>I agree Changwang.  It becomes a 20 octet.  Thus, fe80::1 everywhere gets the fix we usually do: fe80::1%\u00abifindex\u00bb</p>", "time": "2026-03-16T09:44:20.000Z"}, {"author": "Tim Bruijnzeels", "text": "<p>Indeed, configure your ROAs correctly. Or, if you can't, then do not create ROAs.</p>", "time": "2026-03-16T09:49:34.000Z"}, {"author": "Sasha Romijn", "text": "<p>Yes, I would call the core of this problem \"misconfigured ROAs\" - if I understand it right. I do appreciate how the RIPE NCC RPKI portal makes you type \"INVALID\" if you're about to make a BGP announcement RPKI invalid, which is the kind of behaviour that may help.</p>", "time": "2026-03-16T09:53:13.000Z"}, {"author": "Michael Takeuchi", "text": "<p>Shall we create one more validation-state called \"inconsistent\"?</p>", "time": "2026-03-16T09:54:40.000Z"}, {"author": "Maxence Younsi", "text": "<p>@jeffrey @changwang I think the ifindex of the peer could be interesting to get whether the peer ip is ipv4, ipv6 global or ipv6 local. We could avoid correlating bmp with other telemetry sources to determine a bgp peer's ifindex. Getting ifindex as a separate per-peer header field (or a bmpv4 tlv) regardless of peer's ip type</p>", "time": "2026-03-16T09:54:42.000Z"}, {"author": "Tim Bruijnzeels", "text": "<p>The RIPE NCC ROA interface provides users with a preview of BGP impact based on known announcements from RIS - and warns before creating ROAs that would cause invalids</p>", "time": "2026-03-16T09:54:43.000Z"}, {"author": "Tim Bruijnzeels", "text": "<p>RIS is not 100% aware of all announcements but it gets a long way</p>", "time": "2026-03-16T09:55:20.000Z"}, {"author": "Michael Takeuchi", "text": "<p>Data difference from multiple IRRDB source commonly happened</p>", "time": "2026-03-16T09:55:24.000Z"}, {"author": "Jeffrey Haas", "text": "<p>IRR is full of trash.</p>", "time": "2026-03-16T09:56:02.000Z"}, {"author": "Michael Takeuchi", "text": "<p><span aria-label=\"joy\" class=\"emoji emoji-1f602\" role=\"img\" title=\"joy\">:joy:</span><span aria-label=\"joy\" class=\"emoji emoji-1f602\" role=\"img\" title=\"joy\">:joy:</span></p>", "time": "2026-03-16T09:56:15.000Z"}, {"author": "Pierre Francois", "text": "<p><span aria-label=\"rolling on the floor laughing\" class=\"emoji emoji-1f923\" role=\"img\" title=\"rolling on the floor laughing\">:rolling_on_the_floor_laughing:</span></p>", "time": "2026-03-16T09:56:21.000Z"}, {"author": "Jeffrey Haas", "text": "<p>@Maxence ifindex becomes problematic in cases where the remote peer isn't directly connected; e.g. ibgp.</p>", "time": "2026-03-16T09:56:44.000Z"}, {"author": "Tim Bruijnzeels", "text": "<p>I think the message to the leaf customer really is: either fix you ROAs, or stop doing ROAs.</p>", "time": "2026-03-16T09:57:15.000Z"}, {"author": "Jeffrey Haas", "text": "<p>if the argument is \"if you have an ifindex, stuff it in the pdu\", that's perhaps reasonable.  however, it's normally noise outside of link local cases</p>", "time": "2026-03-16T09:57:34.000Z"}]