We looked into the issue and noticed that there is an issue in how sandbox manages acquiring the connection to the server. It’s detailed in this ticket I just opened. The short explanation is that instead of queuing up transactions for writing we let them go straight to acquire a connection to the database for writing, causing an unwanted buildup of requests waiting on a single database writer thread (why just one? Because that’s the point where transactions are sequenced).
Until this is sorted out, it’s safe to retry those commands as the failure occurs before any persistence has taken effect. Please verify whether you have to change the command identifier to prevent command deduplication to cause the new command to be ignored.