This PR reintroduces a single-threaded nature for queue flushing.
Given the strictly serialised nature of assigning sequence numbers without gaps it's unlikely that storage implementations would be able to increase performance by taking multiple flushes in parallel.
This PR reintroduces a single-threaded nature for queue flushing.
Given the strictly serialised nature of assigning sequence numbers without gaps it's unlikely that storage implementations would be able to increase performance by taking multiple flushes in parallel.
Toward #6