On 2014-02-19 13:07:11 -0500, Robert Haas wrote: > On Tue, Feb 18, 2014 at 4:07 AM, Andres Freund <and...@2ndquadrant.com> wrote: > >> 2. I think the snapshot-export code is fundamentally misdesigned. As > >> I said before, the idea that we're going to export one single snapshot > >> at one particular point in time strikes me as extremely short-sighted. > > > > I don't think so. It's precisely what you need to implement a simple > > replication solution. Yes, there are usecases that could benefit from > > more possibilities, but that's always the case. > > > >> For example, consider one-to-many replication where clients may join > >> or depart the replication group at any time. Whenever somebody joins, > >> we just want a <snapshot, LSN> pair such that they can apply all > >> changes after the LSN except for XIDs that would have been visible to > >> the snapshot. > > > > And? They need to create individual replication slots, which each > > will get a snapshot. > > So we have to wait for startup N times, and transmit the change stream > N times, instead of once? Blech.
I can't get too excited about this. If we later want to add a command to clone an existing slot, sure, that's perfectly fine with me. That will then stream at exactly the same position. Easy, less than 20 LOC + docs probably. We have much more waiting e.g. in the CONCURRENTLY commands and it's not causing that many problems. Note that it'd be a *significant* overhead to contiuously be able to export snapshots that are useful for looking at normal relations. Bot for computing snapshots and for not being able to remove those rows. > >> And in fact, we don't even need any special machinery > >> for that; the client can just make a connection and *take a snapshot* > >> once decoding is initialized enough. > > > > No, they can't. Two reasons: For one the commit order between snapshots > > and WAL isn't necessarily the same. > So what? So you can't just use a plain snapshot and dump using it, without getting into inconsistencies. > > For another, clients now need logic > > to detect whether a transaction's contents has already been applied or > > has not been applied yet, that's nontrivial. > My point is, I think we should be trying to *make* that trivial, > rather than doing this. I think *this* *is* making it trivial. Maybe I've missed it, but I haven't seen any alternative that comes even *close* to being as easy to implement in a replication solution. Currently you can use it like: CREATE_REPLICATION_SLOT <name> LOGICAL copy data using the exported snapshot START_REPLICATION SLOT <name> LOGICAL stream changes. Where you can do the START_REPLICATION as soon as some other sesion has imported the snapshot. Really not much to worry about additionally. Greetings, Andres Freund -- Andres Freund http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Training & Services -- Sent via pgsql-hackers mailing list (email@example.com) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-hackers