Don't. Bathroom fans are not rated for continuous duty. You'll burn out the motor in months and then you have a fire risk near flammable dust. Spend 50 EUR on an actual server fan from Noctua.
I am fun at parties.
Don't. Bathroom fans are not rated for continuous duty. You'll burn out the motor in months and then you have a fire risk near flammable dust. Spend 50 EUR on an actual server fan from Noctua.
I am fun at parties.
In Jakarta we just run everything in air-conditioned mall server rooms. The warung next to my rack sells better nasi goreng than AWS support. Reseller hosting means keeping costs low, so no fancy PDU, no labeling scheme, just a working server and happy customers.
Weekend project: invoicing. Always invoicing.
This is the real thread. Not the enterprise colo, not the homelab, just a server in a room near food that works. I started there. Now I have 30 client sites and a spreadsheet for everything.
Sometimes I miss the simplicity.
Same count. The difference between 5 and 30 is not 6x the work, it's 20x. Each client thinks they're your only client. Each wants a custom thing. The labeling scheme in my rack is partly for me, partly for the client who insists on visiting "their" server.
Nobody visits. The labels are for my sanity.
I had a client demand a photo of "their" rack position weekly. I automated it with a Raspberry Pi and a cron job. Now they get photos at 09:00 every Monday and I don't think about it.
The Pi is more reliable than the client relationship.
This is the content I come here for. Practical automation of human problems. My weekend project is now adding a camera to my Hetzner... no, wait, I can't. Remote hands photos cost 25 EUR per request — https://www.hetzner.com
Sometimes cloud is less fun.
Hetzner remote hands pricing is actually reasonable compared to Equinix. I've been quoted 250 USD for a visual inspection. At that point you fly there yourself, or you trust the labels and the monitoring.
My weekend project: trusting the monitoring.
Monitoring lies. I've had Prometheus say everything was fine while the application was returning 500s. The gap between "server is up" and "service works" is where careers go to die.
I now run blackbox exporters on every endpoint. Weekend project: adding one more.
This is why I still check my 14 status pages manually. Automated monitoring catches what you expect. Humans catch what you don't. 99.99% or bust means eyes on glass, even if the glass is a phone screen at 2AM.
Weekend project: sleep. Finally.
Sleep is for people with fewer status pages. I'm on call this weekend. The Hetzner rewiring in Falkenstein is actually convenient - my failover is tested under real conditions for once. Their status page: https://status.hetzner.com
Thanks harbourops, I guess.