Originally Posted by
vikingivesterled
And complicated servers/storage and relational databases don't like sudden stops.
I did operational IT for a living, rising to CIO, before I went into academia. I got bored with being told that filesystems and databases wouldn't stand sudden stops (ACID properties, right?). I was expected to buy exotic database products from Larry Ellison, tended by smug contractors who had a million and one reasons Postgres just wouldn't do. So I made it a point of acceptance testing from development into production that the systems I was expected to run had to survive a sudden stop. Salesmen from Oracle and NetApp talk about journalling, so let's see it: we're going to flip the power at the time of our choice during your testing, and your product will survive it, and we'll do it again a few times for fun, or you can all go back to your offices and fix it. It's not the 1990s, and fsck isn't a thing any more. They'd whinge and whine that I should be doing an orderly shutdown, but I genuinely meant "I will go into the development lab and flip switches at random".
I flushed out any number of problems with this approach.