John Ashmead schrieb am 20.07.2020 um 20:11: > I have had this problem with logical replication on PG 10 repeatedly. > In a clean build no problem. > > But if I am restarting replication because of some problem I’ve seen > problems with rows already present. > > My own fix, which has worked in my shop, is to add replica triggers > to check for the row being already present. If it is, they drop the > row on the floor. This lets stuff come in in whatever order it > happens to come in. > > Sample code: > > if TG_OP = ‘INSERT’ then > select id into id1 from table1 where id = new.id <http://new.id>; > if id1 is not null then > — log error for analysis > return null; > end if; > end if; > > In an ideal world, this would probably not be necessary. But it can > be tricky to restart replication in an absolutely clean way across > all tables and in this case it can be better to allow for a bit of > overlap in the rows being replicated. >
Interesting idea, I'll keep this in mind. For now everything works fine, so there is no pressing need. Thomas
