Offload the DB load to a separate thread and only replace the current
engine instance afterwards.
While reload is pending:
- existing scan requests use the old db (this is unchanged)
- new scan requests are honored instead of blocked and they also use
the old db (this is new)
After the reload is complete:
- existing scan requests use the old db (this is unchanged)
- new scan requests use the new db (this is unchanged)
The existing engine is refcounted so it'll be eventually freed when no
longer in use.
Reload requests while reload is pending are silently ignored (i.e. It
never forks more than a single reload thread).
Patch courtesy of Alberto Wu. We would also like to thank Julius Plenz
for original work on this issue, as well as Alexander Sulfrian,
Arjen de Korte, David Heidelberg, and Ged Haywood for their work
updating and testing these patches.
Micah Snyder