On Wed, Oct 30, 2019 at 5:41 PM Andres Freund <andres@xxxxxxxxxxx> wrote: > > Hi, > > On 2019-10-29 14:33:00 -0400, Tom Lane wrote: > > Mitar <mmitar@xxxxxxxxx> writes: > > > I think RowDescription should be extended to provide full recursive > > > metadata about all data types. That would be the best way to do it. > > > > [ shrug... ] In a world where stability of the wire protocol were > > of zero value, maybe we would do that. In the real world, don't > > hold your breath. > > Hm. Wouldn't it be fairly easy to allow the client to specify how much > metadata they'd want? I.e. opt-in into getting more complete metadata? > > Presumably a lot of clients/applications wouldn't want the server to do > the extra work / use bandwidth for the full details anyway, so making a > more expansive RowDescription be explicitly opt-in would be good, even > if there were zero compatibility concerns. > > There's different ways we could do the opt-in. We could use the "_pq_." > startup option stuff to opt in, we could make it an optional parameter > to D messages (it'd be mildly hacky because unfortunately > describe_target is not a counted text), we could use an additional > describe_type etc... Suppose you had a set of 1000 records, with an array of composites (10 items per array). How many times would the extra metadata describing get sent following this approach? The binary wire format is already fatter in terms of bandwidth in typical cases (unless escaping is involved which can cause exponential growth of the text format). If the answer is 10k, I'd be worried about performance. merlin
