Re: More race conditions in logical replication

On Tue, Jul 25, 2017 at 1:42 PM, Alvaro Herrera
<alvherre@2ndquadrant.com> wrote:
> As I see it, we need to backpatch at least parts of this patch.  I've
> received reports that in earlier branches pglogical and BDR can
> sometimes leave slots behind when removing nodes, and I have a hunch
> that it can be explained by the bugs being fixed here.  Now we cannot
> use condition variables in back-branches, so we'll need to figure out
> how to actually do it ...

If all you had to back-patch was the condition variable code itself,
that might not really be all that bad, but it depends on the
WaitEventSet stuff, which I think is far too dangerous to back-patch.
However, you might be able to create a dumber, slower version that
only uses WaitLatch.

-- 
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company