LMTP is a but troublesome in that a single LMTP session can do multiple RCPT TOs, so the same session will end up delivering mails to multiple different users. This is causing trouble with stats process, because each delivery is using the same session ID for multiple different users, which causes errors about session ID duplication in stats process. So we'll need to make the session IDs unique, while still keeping it clear enough that they're part of the same original LMTP session (which is also visible in th Received: headers).
I'm now thinking that the first recipient's delivery will have the original normal session ID (e.g. "ME3ZHCi+A1dUDQAAvAUe3g"), while the following recipients will have :2, :3, etc suffix (e.g. "ME3ZHCi+A1dUDQAAvAUe3g:2"). If somebody cares about parsing the original session ID from logs, they can just cut off the session ID from the first ':' character (because indexer-worker also adds its own :suffix).
So now I'm mainly wondering if ':' is a good character? Has somebody already changed mail_log_prefix in a way that these additional ':' chars might break their log parsers (e.g. mail_log_prefix=%u:%s%{session}:%r).