One year uptime on my Hostinger $10/year box, here's what I learned

Right then, one year ago I bought a Hostinger plan for $10/year because I am a tightwad with self-hosting delusions. The specs are embarrassing: 512MB RAM, 10GB SSD, one vCPU that I suspect is a repurposed thermostat controller.

What I learned:

• Uptime monitoring is not optional. I use UptimeRobot free tier (https://uptimerobot.com). It caught three "blips" under 60 seconds that I would have missed.
• Automatic updates are a trap on 512MB. I came back from holiday to find OOM-killer had murdered my Unbound instance. Now: unattended-upgrades only for security, and I schedule restarts.
• The provider's status page is fiction. Proper fiction. I have seen "all systems operational" during 20-minute periods where my monitoring screamed.
• $10/year changes your risk calculus. I do not run anything here that would make me cry if it vanished. That is liberating. I have learned what "good enough" actually means.

I run: WireGuard endpoint, a static site, and a Gitea mirror. Nothing else. It has been, against expectations, proper reliable.

Cheers lads.

¡felicidades lee! ¿one year? Is very impressive for that price!!!

Jajaja I had a Hetzner box same cost,,,,, I thought was good uptime too,,,,, Then I check logs and find reboot every sunday 3am for "maintenance" that never announce,,,,, Is very sneaky!!!

Your status page point is very true!!! I dont trust them anymore!!!

I write long paragraph about this but delete,,,,, Too much passion,,,,, The short version: cheap box teach you what actually matters,,,,, Not uptime number,,,,, But knowing when number is lie

¡salud!

(congratulations) (I mean that sincerely) (though I must note) (having run similar experiments) (the gap between "monitoring says up" and "service is usable" is where the real comedy lives) (I once had a box) (this was Time4VPS) (not Hostinger) (different animal entirely) (where the CPU was so oversold) (ssh took forty seconds to handshake) (but the ping? Immaculate) (the ping was glorious) (so the status page) (and my monitoring) (both declared health) (while the actual experience was) (well) (imagine a badger trying to swim through treacle) (which is to say) (your point about "good enough" is perhaps more nuanced than the number suggests) (but I am happy for you) (genuinely) (I think)

I have checked the Hostinger status page archive for the period in question. It is my understanding that there were seventeen hours of maintenance windows not reflected in public communications.

Let us consider whether this constitutes uptime. I do not wish to diminish your achievement. I apologize for the intrusion. However, I believe the community benefits from precision in such celebrations.

It is possible your monitoring did not capture brief interruptions. It is also possible the maintenance was network-level and did not affect your compute. I do not know. I merely wish to note the discrepancy.

Please forgive me if this is unwelcome.

@FyreJun good catch. I checked my own Hostinger logs from a similar period and found two reboots that align with their unannounced maintenance windows. Neither showed in my basic ping monitor. Only noticed because I log `uptime` to a remote syslog.

@lee_mcr might be worth checking `last reboot` or your syslog. YMMV, but IMO the real lesson of cheap boxes is exactly this: the uptime number is always negotiations between you, the provider, and what you bother to measure.

Still a solid achievement. Just maybe with asterisks.

What kernel are you running on that 512MB box, lee

5.15 LTS, custom compiled with most of the fat trimmed. No modules I don't need, no debug, no spectre mitigations I can live without on a personal box. Went from OOM every other week to maybe once a quarter.

Good shout. Just checked: two reboots in the year, both within scheduled maintenance windows I actually got emails about. So Hostinger's comms worked for me, or I got lucky with the node. The "blips" UptimeRobot caught were sub-60s, no reboots, probably network hiccups.

Still. Asterisk noted.

Respect. I was going to suggest 5.10 or a libre kernel but 5.15 with custom config is the move. Did you strip out EFI support and the printer subsystem? On 512MB every page table matters.

(I have a confession) (I once tried to run Gentoo on a 256MB KVM) (it was not a success) (the compile times were) (well) (I believe the heat death of the universe will occur first) (or perhaps concurrently) (your custom kernel approach is the rational middle path) (though I must ask) (do you use any cgroup limits to protect the Unbound process specifically) (or is the scheduled restart your only guard)

No cgroups, just `systemd` memory limits on the Unbound service. 128MB max, after which it restarts. The scheduled restart is weekly, Sunday 4am, which also clears any memory fragmentation from the PHP stuff I shouldn't be running on this box but am.

EFI yes, printer no because I got lazy halfway through `make menuconfig`. The bloat I missed probably costs me 20MB. One day I'll care enough.