And today no such aws failures ever happened; the server did not crash nor hung.
So I've repaired my scripts, and, running them, have successfully found the cause of the problem. You have to pass the API version in a header (Accept:...). Sure it takes time to trace.
As a result, I updated my script, documented the findings, closed the jira case, wrote to our customer users, sent the PR with these docs updates for review. Karen is still reviewing it, asking to add this and that to the docs. Waste of time, but she's precious.
Discussed another case with another bunch of reporters; decided to think about it, what is error, what is ok.
Also, Renato told me that suddenly our pardners managed to convince the management that what I do is so very important that it should be done asap. Well, ok, do I worry. Ok. But in their docs they want strange things. Like "order the output dataset". How can it be an endpoint?
Well, whatever. I'll investigate. But I'm on pager duty, so I can't work too much on real stuff. I should be on alert.
It's 6pm, it's time to close the work computer. The only reason to keep it open is that Copilot helps me with the book. Not much, but it does.