This. Contabo's price point suggests MikroTik or similar at edge. 48 hours is a suspiciously round number for a conntrack timeout or a scheduler reboot. Not malice, just cheap hardware with aggressive cleanup.
I would not expect GR to help if they are resetting TCP state. The notification is forged, the session is killed at layer 4.
You were right. 48 hours hit, GR negotiated, session died anyway. No forwarding preservation. Their edge sends the notification and tears down the TCP socket completely. GR only helps if the peer is actually restarting, not administratively murdering the session.
Back to crontab. I have opened a second ticket asking for escalation. Same copy-paste reply. "Network stability feature." No timeline for change, no workaround, no documentation.
At this point you are documenting it for them. I would publish the ticket exchange, redacted. The community needs to know that Contabo's BGP is explicitly non-production. Their AS51167 should be filtered accordingly by anyone requiring stability.
I am updating my client's provider evaluation. Contabo moves from "budget option" to "unsuitable for BGP."
Member
I will add this to my French self-hosting forum. Many people there use Contabo for price. They need to know.
Did anyone try asking for a dedicated BGP session on their VDS or dedicated? Maybe the 48-hour limit is VPS-only. Their VDS is only slightly more expensive.
I asked in my second ticket. They did not answer the question. Just repeated the same line about "network stability feature."
I am done with Contabo for this project. Moving the node to Vultr Seattle this weekend. USD pricing hurts but 48-hour drops hurt more. Thanks everyone for the diagnostics.
