On Mon, June 2, 2008 6:53 pm, Tom Lane wrote: > "Henry" <henry@xxxxxxxxx> writes: >> I'm trying to code a function to copy rows from one machine to another >> using dblink and cursors: > > What PG version is this, exactly? Arg, dammit. Sorry, it's version 8.2.6 (where the function is running), talking to a remote machine running 8.3.1. > I don't think your problem has anything to do with dblink per se. > The repeated begin/exception blocks are apparently managing to leak > some memory per iteration. I can't tell whether this represents > a known (and perhaps already fixed) bug; it very likely depends on > details you haven't shown us. Do you want to try to put together a > self-contained test case? (Again, you likely don't need dblink to > exhibit the issue.) OK, will give this a swing; but I'll first upgrade to 8.3.1 and see what that does. > If you just want something that works now, try probing for an existing > entry before inserting, instead of relying on catching an exception. ok, but that's going to be slow considering I'm inserting almost a hundred million rows. My previous post wasn't entirely accurate: the dblink_exec() call in a loop is also enclosed in a begin/exception/end block, yet it doesn't fail. The only difference besides the dblink_exec call itself, is that I'm using 'others' to catch any error from dblink_exec (using 'others' on the first insert results in the same issue).
