There is a pattern of retries built into the code, but syncing has no ability to repair failures (that get past the retry logic), and there have been hand-documented cases where even with retries, there is significant data loss: https://github.com/LemmyNet/lemmy/issues/3101