Post AznUPhDh2dexaYZIh6 by picofarad@noauthority.social
(DIR) More posts by picofarad@noauthority.social
(DIR) Post #AznUJmXWLkSGLm1S9A by dotstdy@mastodon.social
2025-11-01T08:01:01Z
0 likes, 0 repeats
RE: https://mastodon.social/@dotstdy/115473335676948840I read performance investigations sometimes, and I do really wish people would try slightly harder to find root cause, rather than randomly changing config.
(DIR) Post #AznUJnHFbj2CdbPyQS by dotstdy@mastodon.social
2025-11-01T08:30:05Z
0 likes, 0 repeats
It's totally fine not to fix things of course, but it's really weird to me when people power through without asking why something is slow, rather than just accepting it as a fact of the world, and working around it. I imagine this is a recurring problem in the "soup of microservices" model generally, because it so effectively obfuscates the work that's actually happening. Here it looks like watching a list for changes is likely done in a pretty naive way and it causes very poor scaling.
(DIR) Post #AznUJoBGFTpzRJchjE by picofarad@noauthority.social
2025-11-01T09:21:28Z
0 likes, 0 repeats
@dotstdy "picofarad you need to stop complaining about the PR and just merge, the deadline is thursday evening. You can't keep bellyaching about 20 milliseconds; it's twenty milliseconds, for goodness sakes"
(DIR) Post #AznUJsumeAd07ex7Eu by dotstdy@mastodon.social
2025-11-01T08:35:52Z
0 likes, 0 repeats
also i didn't realize that kubenetes was a golang disaster lol https://kubernetes.io/blog/2024/12/17/kube-apiserver-api-streaming/#why-does-kube-apiserver-allocate-so-much-memory-for-list-requestsSo if you use list, every query is going to be bonkers, and if you use watch they're going to add a (per-connection?) cache, sized to your query to avoid spamming the poor GC, but with a fixed (per-connection?) memory overhead. But if you have the newest hotness they will stream the list instead of generating huge amounts of garbage, at least.
(DIR) Post #AznUPhDh2dexaYZIh6 by picofarad@noauthority.social
2025-11-01T09:22:34Z
0 likes, 0 repeats
@dotstdy "back when i was doing siebel databases we'd have sold our souls for a 20ms query, get over it, i've been in this business for 20 years"