Safety Fork, the

I read some random guy’s ‘blog on Friday. It wasn’t even that memorable a ‘blog—I’ve forgotten the URL—but I was bored. What stuck with me was this guy’s attachment to his web server’s uptime. He’d get upset about having to unplug his server to fix his UPS or solve his floating ground problems. I mean, it’s cool that he is proud of how stable his server is, but nobody really cares if it’s been running for one day or one hundred.

Availability matters with web servers, uptime is irrelevant. I could reboot my server at three or four o’clock every morning, so that the uptime was never more than twenty-four hours, and no one would care. Why should they? No one accesses my server at four in the morning except script kiddies trying to brute-force my ssh login,¹ and I confess that I don’t really care if I interrupt them. I care that I don’t have to reboot my server every day, that as a matter of quality my software and hardware are more stable than that. But if I bought a UPS, I would have no qualms about shutting the system down for ten minutes to install it—the benefit to reliability would be measurable and the detriment to current uptime merely cosmetic.

¹ A waste of time, since I permit only public-key authentication.