-
I just turned an O(N) into an O(1). An I/O-blocked job that was taking some requests longer than the APIGW limit (29s) now finishes in a consistent baseline, even with 5x the input. …been a while since I’ve been able to do real work (instead of just fighting stuff in my way).
-
(To be fair, it’s not *really* an O(1), but it’s acting like it. I’ll take it. Python’s ThreadPoolExecutor mechanics are great for reducing the impact of this kind of blocking.)