On 2025-02-09 07:43, James Bensley wrote:
* There are no knobs in existing BGP implementations to detect and
limit this behaviour in anyway.
100% agreed. Looked into this a couple weeks ago on our $VENDOR_C gear,
and we saw the prefixes Romain mentioned as well as many others in
Geoff's report. I coded a CLI poller / screen-scraper to compare the
different versions of the BGP table, in accordance with this tech
article[1]. Seems to be as good as one can get with vanilla XR.
When looking at a specific prefix in the table, there is also a
timestamp for the last update. It's helpful, but doesn't give a sense
for how frequently it is updating.
Would be great to have logging / alerting if a locally originated prefix
is updated more than once a second. Should also be easier to see high
update load for those getting the updates downstream. A top N prefix
frequency distribution based on short time windows (1-15 min) would have
saved me work.
-Brian
[1]
https://community.cisco.com/t5/service-providers-knowledge-base/bgp-churn-identification-on-ios-xr-asr9000-use-case/ta-p/4280504
