BGP session flaps every tuesday 3 am. Two upstreams. Only one flaps. Fiberlink good. Driftnode bad. No alerts received. Logs show hold timer expired. Perkele. Running debian with frr. Not maintenance window on my end. Where to look
Help: my BGP session flaps every Tuesday at 3 AM
I had this with InterServer last year, their window was 2:45-3:15 AM local. Also:
- check if its always exactly 3:00 or drifts
- ask them about scheduled work
- $/GB/RAM on DriftNode is terrible anyway, maybe time to migrate
(I was going to say maintenance window) (but then I thought) (maybe its something more interesting) (like a cron job on your end) (running at 3 AM) (updating something) (that touches the network stack) (but probably not) (its almost certainly their window) (and you missed the email) (because they sent it to noc@yourdomain) (which goes to /dev/null) (or your spam folder) (same thing really)
Actually this happened to me before at old job (´・ω・`) basically provider did maintenance every tuesday but email went to wrong address we didnt know for months kaomoji sad times you should check with driftnode support actually ask them specifically about tuesday window
I run my own setup with docker compose and a reverse proxy, full control, no mysterious 3 AM flaps. Sure, you need to announce your own prefixes, but that's the cost of reliability. I have three upstreams now, all in compose files, health checks every 30 seconds.
It was their maintenance lol. It is possible to do that? Yes very possible. Check your contact email in their portal, maybe you have a typo. Mdr I had "admni@" instead of "admin@" for 6 months. Very fun.
Actually the german way is to check all the Details first before asking the Provider. Look at your PeeringDB entry. Look at your IRR records. Actually the email field there is often outdated. Or? I once had a stale NOC contact for two Years because nobody updated the aut-num object. Compoundwordsfusedtogether in the whois output can also hide typos. Actually check every Character.
Which Debian version and FRR release are you running?
DriftNode's 2:45 window sounds right, ask them first
DriftNode never sends their windows to the right NOC address.