I took a look at grafana.oas.greenhost.net and increased all the pods that got throttled to a point where they should not get throttled anymore.
The rule we discussed last week was to set the request to 2x peak usage. I rounded most requests up to multiples of 100m, because I vaguely remember that that's a good amount to prevent unnecessary throttling (If I remember correctly it has to do with the time frame that the linux kernel looks at when deciding if your process is allowed more CPU seconds).
I increased some processes to more than 2x peak usage. Especially if I saw that the throttle percentage was over 50%. I assume that 50% throttling means that the process asked for 2x the amount of CPU they got. That means that if I saw 90% throttling, I thought the process should have 10x the amount of CPU it had.
I did a rough calculation of requests and we should still be far under requesting 4 cores.
Closes #786 (closed)