> Synchronization between threads is handled by the application, and tsan doesn't appear to complain. Each thread has it's own connection to the database, and write requests are executed with an exclusive lock while read requests can be processed in parallel. However even if all requests are executed with an exclusive lock, the problem still occur. Exclusive Lock -- Do you mean at the application level (quite outside of SQLite3) or are you somehow referring to SQLite3's exclusive locks, which are a different thing entirely? Are you checking *every* return code for errors? Are you using some strange filesystem (any filesystem can be strange, so which one in particular would be helpful). Are you using shared_cache or something like that not recommended for use? What journal_mode are you using? Have you examined whether "docker" supports POSIX advisory locks and sync operations correctly? I've never used "docker" though I understand it is all the rage with the kids these days, but as I understand it, "docker" works by adding an additional layer of "cruft" between the application and the Operating System. Do you know that this additional layer of cruftiness works properly? (Maybe someone who has used this "docker" thing can weigh in here -- and I mean actually used it -- not just read the glossy brochure). My money would fall on either the application being in error or "docker" being broken.