The example is copied from the initial post in this thread:
https://forum.rclone.org/t/rclone-v1-55-1-throttling-and-worse-on-onedrive-copy-to-gd/25018
Yes, the pacing seen in the log is somewhat odd. This is what I see:
rclone receives a http response code 429 (Too many requests) having retry-after=120296 seconds (33h24m56s). It is my understanding that retry-after period as a (minimum) period where the client shouldn't make any requests to avoid further throttling or to get blocked.
The pacer correctly increases the wait time accordingly, however it does immediately start an exponential reduction of the wait time, so the wait time is reduced to 14h5m49.875s (that is halved) within a second. This will make rclone retry to early and then the server responds with a another and much longer retry-after.
The log therefore indicates a possible bug in the way rclone handles http 429 responses that may lead to rapidly escalating throttling issues - ending with retry-after times in hours and days.
This behavior could also explain the experienced importance of staying below the throttling limits at all times.
Do you (or other rcloners) see the same?
I will try to find some time to investigate/reproduce, if you or others can confirm the observation/suspicion.