I tried to put this thread into YAML and my linter cried.
Honest_numbers:
Designed_for: "99%"
Paid_for: "99.99%"
Actual: "who knows I stopped checking"
Provider: "someone in Germany"
Reason: "my ansible playbook is 400 lines to handle failover between two 5/month boxes"
The real number is how fast my pager wakes me. Since I moved everything to "good enough" with automatic failover, pager is silent. Silence is the only SLA I trust.
Denver here. I run BGP to two providers, one through Vultr in Los Angeles, one through Hetzner in Hillsboro. Real number? I get 99.999% because when one path dies, other takes 30 seconds.
But I pay for two paths. So I pay for five nines and I get five nines, but the cost is double, not the single SLA price. Nobody talks about this math.
Single provider "five nines" is fiction. Two providers "three nines" each with automatic failover is honest five nines.
I dream of two providers. In Karachi I have one good fiber, one microwave backup to different POP. Both go to same exchange because city planning.
My real number is 99.5% and I tell clients 99%. Underpromise, survive monsoon. Vultr Mumbai is my overseas backup, 180ms away but alive when local drowns.
ASN dream is because then I control the fiction.
Member
My VPS is 2.50 USD at InterServer. I do not know the SLA and I do not care. Uptime is 99. Something, my blog about coffee reviews does not need more.
If it dies I post on Twitter "server is down, touch grass." My followers think this is personality.
Real need: 95%. Real payment: for 99.99%. Real attitude: whatever lah.
Hahaha Joy you are my spirit animal. But my warung clients want WhatsApp API always online. If down, they lose orders. So I pay 99.99% at InterServer, reality 99.5%, and I have cheap Contabo Singapore backup that I switch DNS manually. Manual switch = 15 minutes. But 15 minutes is better than 3 hours waiting for "network maintenance."
I am the failover. My thumbs are the SLA.
Tallinn perspective: EU regulations say I must report availability to clients. I report what I measure, not what I pay for. This caused problem.
I paid Hetzner 99.99%, measured 99.97%, reported 99.97% to client. Client asked why below SLA. I showed Hetzner's own status page — https://status.hetzner.com — with 6 minutes unplanned. Hetzner credited me 3 EUR. Client wanted 300 EUR penalty from me.
Now I contract 99.9% to clients even if I pay for more. Honesty is expensive.
Member
I break things professionally and I will tell you: your "designed for two nines" is actually designed for one nine with hope.
Two nines = 87.6 hours downtime/year. You know what happens in hour 88? You find out which parts of your "redundancy" were theoretical. I have seen this at three companies.
Design for one nine, test to two nines, pay for three nines, tell board five nines. This is the real ladder.
How? I run 12 boxes, no staging environment. "Test" is Tuesday at 3 AM when I restart one and hope.
My honest numbers:
- designed: restart at 3 AM and pray
- paid: 99.99% at OVHcloud Beauharnois
- reality: never checked, too scared
This thread is making me anxious. I think I will check now.
I checked once. OVHcloud Vint Hill, 99.91% over 2 years. I opened ticket. They said "your monitoring tool is wrong, use ours." Their tool said 99.99%. I said "show me the raw data." They closed ticket.
Now I use both and report the worse number to myself. This is called "management."
Member
Caramba, I check now my cat photo monitoring. UptimeRobot says 99.87% for 30 days. I pay for 99.99% at Contabo Sao Paulo.
But you know what? My boss sees green. Green is 99.87% or 99.99%, green is green. I am not telling him.
Kkkkkk this thread is too honest for me, I will go back to make dashboard more green.
