Hi Mukul,
Thanks. The Peer Down/Peer Up pointer is the piece I had missed, and I agree it
is needed for middle of session changes.
One distinction I would like to callout as it gives certain *temporal*
consistency:
*
the filtering characteristics cannot change within the life of a session.
*
What it does not yet give is *cross-message-type* consistency
*
that F be set the same way in the Stats Report peer header as in the Route
Monitoring peer header for that same session.
*
Those properties can be made distinct, and the current Section 4 text
constrains only the counts ("counts MUST reflect the filtered RIB numbers and
not the original RIB ones"), not the flag.
This as you mentioned would require paraphrasing.
One more thing on Section 4 with your RFC 9069 comparison in mind, I came upon
something I do not think the Peer Down/Peer Up rule closes. So please take a
look at this:
*
RFC 9069's F flag sits in the Loc-RIB Instance Peer Type flags registry, where
the peer has exactly one RIB.
*
"Is this stream filtered?” therefore has one unambiguous answer, and RFC 9069
never needed to say anything about consistency across message types.
adj-ribs-filtered places F at recommended value 4 in the "BMP Peer Flags for
Peer Types 0 through 2" registry, where the same peer carries two RIBs
distinguished by the O flag. For Route Monitoring that works cleanly: O selects
the direction and F qualifies it. For Stats Report it does not, because a
single Stats message can carry counters for both directions at once:
- Stat Types 7 and 9 - Adj-RIB-In (RFC 7854)
- Stat Types 14 to 17 - pre- and post-policy Adj-RIB-Out (RFC 8671)
and all under one peer header carrying one F bit. If an implementation filters
Adj-RIB-In but not Adj-RIB-Out, there is no correct value of F for that
message. F=1 overstates the filtering, F=0 understates it, and the Peer
Down/Peer Up rule does not help because nothing changed mid-session.
That brings back to the usage of “F”:
*
is the intent that F in a Stats message means "at least one of the RIBs
represented here is filtered", or that filtering must be all-or-nothing across
directions for the lifetime of a session?
*
Either is workable, but the draft probably has to pick one and say so.
I would separate it from what I read Reshad as asking, which is a runtime
question.
*
Section 4 of rel-enhancement-03 already contemplates runtime filtering
explicitly - "local endpoints can apply event filtering controls" - and
*
Section 6 does define a fixed set to filter from, namely the seven code points
requested in the BMP Log Action TLV Registry
*
(Redirect-to-VRF-Fail, Redirect-to-Nexthop-Fail, Redirect-to-SR-Policy-Fail,
FlowSpec-Validation-Fail, Invalid-Candidate-Path, Invalid-Segment-List,
Exceeded-Spec-Limit).
A collector that sees no Redirect-to-VRF-Fail events cannot distinguish "none
occurred" from "that type is suppressed at the source”. Which is the same
inference problem that Section 3 of adj-ribs-filtered describes for RIBs. That
said, I agree with you that this does not warrant a new mechanism. If REL
messages carry the standard per-peer header, then reusing this same F bit, with
the cross-message consistency rule above, is the minimal fix and costs each
document one paragraph.
Hi Nan
Does REL use the standard per-peer header? rel-enhancement-03 does not
reference it, so I may be assuming a structure that is not actually there. If
REL messages do not carry a peer header, then the signal has to live somewhere
else and the answer changes.
Please do let know , I will be more than happy to draft some text with your due
approval
Thanks,
Saumya
From: Srivastava, Mukul <[email protected]>
Date: Tuesday, 28 July 2026 at 8:57 PM
To: Dikshit, Saumya <[email protected]>; gengnan <[email protected]>;
Reshad Rahman <[email protected]>
Cc: Grow <[email protected]>; [email protected]
<[email protected]>
Subject: Re: [GROW] Re: Question on draft-geng-grow-bmp-rel-enhancement
Hi Saumya
>>>>> " but it stops short of stating that the F flag itself be set
>>>>> consistently in the Stats message's Peer Header for that same session.
[MS] - The draft has below statement in section 4 -
"
Also should any characteristic of the filtering change, the sender MUST trigger
a Peer Down then followed by a new Peer Up.
"
This intends to say that if filtering changed for any reason, we should start
afresh. This indirectly means all session should have uniform “F” flag setting.
I am ok to add some explicit statement, as you suggested to clarify this. The
current statement matches BMP local-rib RFC 9069 where route filtering flag as
first introduced.
>>> That said, it doesn't fully close Reshad/Nan's original question:
>>> adj-ribs-filtered-00 only talks about Adj-RIB-In/Out and Stats messages and
>>> says nothing about routing-event messages
>>> (draft-geng-grow-bmp-rel-enhancement's actual subject).
[MS] - IMO, the filtering flag is not application for rel/log messages. There
is no fixed set of the routing events pre-defined, out of which, a router may
intend to filter out some. So, these messages are different from route-motoring
or stats messages.
The original comment in the email below was to be cautious to select what
should be send as logging event to the BMP station. If some events (like
next-hop change) are chatty that shouldn’t be logged, meaning we shouldn’t
define an logging event type to report next-hop churn.
Thanks
Mukul
From: Dikshit, Saumya <[email protected]>
Date: Sunday, July 26, 2026 at 6:24 AM
To: Srivastava, Mukul <[email protected]>; gengnan <[email protected]>;
Reshad Rahman <[email protected]>
Cc: Grow <[email protected]>
Subject: Re: [GROW] Re: Question on draft-geng-grow-bmp-rel-enhancement
Hi Mukul,
That's a helpful clarification, thank you.
It's good to be reminded that F flag lives in the shared Peer Header rather
than being RM-specific, so it's structurally available to Stats messages too.
Looking at the current adj-ribs-filtered-00 text again with that in mind, I
think it actually pins down exactly the gap:
* Section 4 says "In Stats messages, counts MUST reflect the filtered RIB
numbers and not the original RIB ones," but it stops short of stating that the
F flag itself be set consistently in the Stats message's Peer Header for that
same session.
*
So as written, an implementation could arguably satisfy the letter of the text
by adjusting the counts while leaving F unset (or inconsistent) on the Stats
side
*
This is exactly the inconsistency you flagged as unhelpful for correlation.
Would it make sense to tighten this into an explicit MUST, something like:
* "The Peer F Flag MUST be set consistently across all BMP message types
that share this Peer Header for the lifetime of the session (Route Monitoring,
Stats Report, Route Mirroring).
*
An implementation MUST NOT report F=1 in one message type while reporting F=0
(or omitting the equivalent filtered-count adjustment) in another, for the same
peer."
Since you're already a co-author on adj-ribs-filtered, that
operational-consideration addition (Section 5, or as a normative bullet in
Section 4) seems like the cleanest, lowest-overhead fix and no new TLV needed
for the RIB/Stats side at all.
That said, it doesn't fully close Reshad/Nan's original question:
adj-ribs-filtered-00 only talks about Adj-RIB-In/Out and Stats messages and
says nothing about routing-event messages
(draft-geng-grow-bmp-rel-enhancement's actual subject). On that note, would it
make sense for rel-enhancement to explicitly say event messages reuse this same
Peer F Flag bit (with the same cross-message consistency rule above), rather
than defining a separate mechanism?
*
That would give collectors one consistent signal to check across RM, Stats, and
event messages, with a single small text addition on each side rather than a
new generic flag.
Your thoughts Nan, Reshad, Mukul on this ?
Happy to help draft the exact text for either document if useful.
Thanks,
Saumya
From: Srivastava, Mukul <[email protected]>
Date: Saturday, 25 July 2026 at 8:50 PM
To: Dikshit, Saumya <[email protected]>; gengnan <[email protected]>;
Reshad Rahman <[email protected]>
Cc: Grow <[email protected]>
Subject: Re: [GROW] Re: Question on draft-geng-grow-bmp-rel-enhancement
Hello Saumya
>>>>>> If a router locally rate-limits or drops events (per the mitigation text
>>>>>> in draft-geng-grow-bmp-rel-enhancement) and separately reports gauges
>>>>>> such as the ones in
*
draft-ietf-grow-bmp-bgp-rib-stats (RFC 9972) or
*
the policy-driven counters in draft-smc-grow-bmp-route-change-stats,
a collector has no way to know that an observed gauge delta might not be fully
explained by the event stream it received it may look like an unexplained jump.
—
[MS] - The BMP stats messages have BMP peer header. The filter flag is part of
the BMP peer header. If the filter flag is not set in stats message, while
the other msg had filter flag set, then the collector should use caution and
know that the data it received was filtered while the stats count is not
filtered.
I would assume for ideal operation, the router should have flag set/unset for
all events/data/stats message for a session. Inconsistency in this flag state
between different messages, will not help in meaningful correlation at
collector.
I think we should add this in the operational consideration for filtered
drafts, if it isn’t present already.
Thanks
Mukul
From: Dikshit, Saumya <[email protected]>
Date: Saturday, July 25, 2026 at 2:37 AM
To: gengnan <[email protected]>; Reshad Rahman
<[email protected]>
Cc: Grow <[email protected]>
Subject: [GROW] Re: Question on draft-geng-grow-bmp-rel-enhancement
Hi Nan, Reshad,
I was going through the draft and couldn’t help bringing up another angle,
(though not-unrelated) to the email chain discussion :
The same blind spot effects statistics consumers, not just event-log consumers.
If a router locally rate-limits or drops events (per the mitigation text in
draft-geng-grow-bmp-rel-enhancement) and separately reports gauges such as the
ones in
*
draft-ietf-grow-bmp-bgp-rib-stats (RFC 9972) or
*
the policy-driven counters in draft-smc-grow-bmp-route-change-stats,
a collector has no way to know that an observed gauge delta might not be fully
explained by the event stream it received it may look like an unexplained jump.
Rather than defining filtering-indication independently for events (in this
draft) and for RIB streams (bmp-adj-ribs-filtered), would it make sense to
define one small, generic TLV/flag, that any BMP consumer event, RM, or SR
message can check? This would definitely help in achieving "telemetry
completeness"
More than happy and willing to sketch this as a short contribution if there's
appetite, as we hit a related need while defining the per-peer gauges in
draft-smc-grow-bmp-route-change-stats.
Thanks,
Saumya.
From: gengnan <[email protected]>
Date: Thursday, 23 July 2026 at 3:39 PM
To: Reshad Rahman <[email protected]>
Cc: Grow <[email protected]>
Subject: [GROW] Re: Question on draft-geng-grow-bmp-rel-enhancement
Hi Reshad,
As stated in Section 3 of draft-ietf-grow-bmp-adj-ribs-filtered:
> ... be informative to the rest of the system that they should not make
> decisions assuming that this stream is transmitting all the information about
> it.
The same logic seems to apply to event reporting. If critical events are
filtered locally without any signal to the collector, monitoring systems may
draw incorrect conclusions based on incomplete data. It would therefore be
valuable to consider a mechanism to convey this filtering state for events.
Best,
Nan
From: Reshad Rahman <[email protected]>
Sent: Thursday, July 23, 2026 1:43 AM
To: Grow <[email protected]>
Subject: [GROW] Question on draft-geng-grow-bmp-rel-enhancement
Hi,
Appropriate caution SHOULD be taken when picking routing events to
report, as next-hop related updates may produce heavy message loads
during network instability. Mitigation measures SHOULD be deployed
on both reporting and receiving sides. For example, local endpoints
can apply event filtering controls, while receivers maintain
corresponding processing mechanisms to handle potential high-volume
event bursts gracefully.
During the meeting today the text above was highlighted by Nan (based on
feedback from Prasad). I have no issue with the feedback and text, but is there
any indication of filtering sent to the collector?
draft-ietf-grow-bmp-adj-ribs-filtered which was presented earlier adds a Peer F
flag for Adj-Rib-In/Out, IIUC that doesn't cover events in
draft-geng-grow-bmp-rel-enhancement? IMO we want indication of filtering for
"critical" events.
Regards,
Reshad.
_______________________________________________
GROW mailing list -- [email protected]
To unsubscribe send an email to [email protected]
