Heikki Linnakangas <heikki.linnakangas@enterprisedb.com> wrote:
> All I'm saying is that if we end up shipping without that
> parameter (implying allow_standalone_primary=on), we need to call
> the feature something else. The GUCs and code can probably stay as
> it is, but we shouldn't use the term "synchronous replication" in
> the documentation, and release notes and such.
I think including "synchronous" is OK as long as it's properly
qualified. Off-hand thoughts in no particular order:
semi-synchronous
conditionally synchronous
synchronous with automatic failover to standalone
Perhaps the qualifications can be relaxed in some places but not
others? The documentation should certainly emphasize that there is
no guarantee that a successful commit means that the data is on at
least two separate servers, if no such guarantee exists.
If we expect that losing all replicas is such a terribly thin long
shot that we don't need to worry about this difference, it is such a
long shot that we don't need to worry about "wait forever" behavior,
either; and we should just implement *that* so that no qualification
is needed. I think that is an odd assumption, though; and I think a
HA failover to weaker persistence guarantees in exchange for
increased up-time would be popular.
-Kevin