A change to enable advanced debugging on a component supporting DeepL write for pro customers, caused the service to deploy incorrectly and reject incoming requests. We are looking into the root cause of this and will update this page with our findings after conducting a thorough internal review.
We have identified the root cause and want to share the results here. The rollout of a feature to enhance error debugging resulted in a Write Pro instance being scheduled on an overloaded host, creating a "noisy neighbor" issue. This led to slow responses and connection pileups in our reverse proxy. To prevent similar incidents, we plan to implement several measures to better isolate resource-intensive workloads and improve how our services handle temporary slowdowns.