Normally I have the tables preallocated to exceed the expected tables sizes, but I haven't updated the allocations in about a year, so toward the end of the sync they start flushing. This is basically the same mmap behavior we would expect from periodic snapshots (which require a full flush).