Skip to content
Register

Outage: I caused my own by 'optimizing' kernel params

General Discussion by Zurich1984 3 replies 193 views
Zurich1984

Member OP

#1

So I followed the official OVHcloud wiki page "Debian 12 Performance Tuning" and applied their sysctl recommendations. Vm.swappiness=10, net.core.somaxconn=65535, bunch of TCP tweaks.

Rebooted. Network came up. Then sshd started accepting connections and immediately dropping them. No error, just hang and timeout. LISH console showed nothing wrong. Took me 40 minutes to realize somaxconn was breaking their OpenVZ template's socket handling. Had to boot rescue, revert, apologize to clients.

Wiki still has that page up. /24 will cost you a kidney but at least IP space does not gaslight you.

/24 for sale. No lowballs.
pieter_rtm

Member

#2
Zurich1984 said:
Their OpenVZ template's socket handling

As said, the parameter is harmless on proper virtualization. I have flagged the article for review.

  • Your ticket response time: 0 minutes (self-service)
  • Actual downtime: 40 minutes
  • Lessons learned: price above zero

We will update the page or add template warnings. OVHcloud credit applied to your account.

Containers before it was cool
SingaporeRep

Member

Benchmark Addict

#3 
# before "optimization"
Disk Speed:
  4k: 89 MB/s
  64k: 312 MB/s

# after rescue revert
Disk Speed:
  4k: 91 MB/s
  64k: 308 MB/s

Disk's fine, network's nothing special your tuning gained nothing measurable. Yabs dont lie.

fio, iperf, geekbench. results or gtfo.
TA

Member

#4

Man this is why I never touch kernel stuff, going to break everything every time. That wiki page going to kill someone someday, this is not funny but going to laugh anyway

Post a reply

You need an account to reply. Log in or register to join the conversation.

Post reply Preview Save draft