Member
I keep thirty client sites running in Amsterdam. Two of them wanted cheap anycast last year. I put them on Vultr London and Amsterdam, BGP fine, no 48-hour drops. That was mid-2024. Maybe Vultr fixed it or maybe it was specific regions. I never saw it.
Contabo I only use for offsite backup storage. Would not trust BGP there after reading this.
Vultr São Paulo works fine for my setup. Full table since March, no drops. But I only announce a /24 and do not need the full table. Maybe the limit was for full-table sessions only? Less memory on their edge?
Unlikely. RIB size is downstream of the session. The session itself costs the same whether you accept 0 prefixes or 900k. If they were memory-constrained they would limit RIB, not session age.
What AS is Contabo using for their sessions? I want to check their advertised capabilities — https://bgp.tools works for that.
AS51167 for the VPS BGP. I can check their capabilities when the session comes back up, currently in my 47-hour crontab window so it's down. Give me two hours.
I run a bakery and somehow also servers in Portland. I was going to ask what "full table" means but I searched and found it. 900,000 routes. That is a lot of routes. My entire network is three static IPs.
Following this thread because I like learning. Please do not explain to me, I will read more.
I have a Contabo VPS in Singapore, same AS. No BGP but I ran persistent TCP streams for video ingest. They drop at 48 hours too. Not BGP-specific. Their "network stability feature" is a generic TCP session limit.
That is worse. A 48-hour TCP cutoff breaks long-lived SSH tunnels, database replicas, anything stateful. BGP just happens to be the most visible because it logs explicitly.
Carl, when your session restarts, check if they support
BGP-GRLLGRSession is up. They advertise
Multiprotocol Extensions, Route Refresh, Graceful Restart, Four-Octet AS Number
Member
I look after a rack in Madrid. We had similar with a budget Spanish provider, can't name them but not Contabo. Their "stability feature" was actually a state table overflow on cheap firewalls. Every 48 hours they flushed conntrack. BGP died, IPSec died, everything.
Moved to OVHcloud Strasbourg. More expensive, no funny business.
Same here. Following for Carl's GR test results.