I have introduced a small change to a the serialization system that worked great and has been fully tested locally but has effected some part of the clustering. When I deployed a new version I tested everything and it was fine and then the problems started a little bit later, when I did not realized it immediately.
It was just pure bullshit that should not have happened. I knew that this change might be a little bit problematic and have therefore moved it to my late afternoon after business hours in Germany.
I will create a report and make a plan how to ensure that this does not happen again. One part of the store is the team size of course, but the other part are to create full clustering tests that should run before a new release.