Skip to content
Register

My anycast nodes are flapping and my monitoring thinks it's fine

Networking Locked by minh1987 4 replies 390 views
This thread is locked — new replies are not accepted. (Duplicate — see the earlier thread on this)
minh1987

Member OP

5 #1

I run a small anycast setup across three regions for a DNS resolver project. Two nodes in my country, one in Europe. Past week I have been seeing intermittent timeouts from users in SEA but my monitoring shows 100% uptime.

Using OVHcloud for the European node, Contabo for local. Monitoring is simple ICMP from a fourth box to the anycast IP. The flapping seems to happen during peak hours only. Traceroutes from affected users sometimes land on the European node instead of local.

I checked BGP sessions; all stable. No withdrawals in logs. Where should I look next? Is my monitoring just checking the wrong instance entirely?

phở at 3AM, deploy at 4
priya

Member

🇮🇳 netops ninja

#2

Dear/madam,

Kindly verify from which location your monitoring probes are originating. If the probe itself is within Europe, then it will reach the European node only; the Asian nodes will not be tested at all.

You may also check whether your anycast steering is functioning properly or not. In my experience, some providers route traffic based on TCP properties itself rather than source geography.

Please share your monitoring configuration details if possible.

Thank you.

carlos2

Member

1 #3

@priya good catch

This sounds like classic anycast blindness

Which providers take XMR for anycast monitoring services? Asking because I want to move off my current setup that requires card details. Privacy coins should be standard for infrastructure payments

Also @minh1987 your monitoring box—where is it physically? If it's on the same continent as your OVHcloud node you're basically paying to monitor one third of your setup

not your keys, not your coins
wilmaethqn44

Member

#4

What does "flapping" mean here? Is it like the servers are going up and down?

I run a small site on RackNerd and this sounds scary. Should I be worried if im not doing anycast? I just have one server

Also how do you even check if monitoring is looking at the wrong thing? That seems like a thing that would be hard to notice

frames, tables, still valid HTML
prague1983

Member

btw I use Arch

#5

Priya nailed it. Your probe is probably hitting the Frankfurt node every time

oops: 0000 [#1] SMP

Thread closed

Replies are closed on this thread. Duplicate — see the earlier thread on this.