As often OpenAI ChatGPT has some outages,
that looks like this at the moment.
This seems to lead to an average of 1 reload of all active users.
This doubled the traffic to AIPRM immediately.
While there are many optimizations in place, with a young system like AIPRM some lesser needed functions were untouched since January.
These logging functions suddenly caused a hockey-stick or at least exponential time delay behavior, slowing down everything else.
For the technically interested, this non-linearity in response times can happens in all system, at some point very surprised - as was the case here too. Nonlinear problems like that are described here also.
where I took this example illustration of a hockey stick curve on response times from
AIPRM service is restored, but at the time of writing this ChatGPT issue still exists:
This gives the impression that AIPRM has a problem - not anymore.
We continue looking into things, but service for users should be fully restored.