Larry Rosenman <ler(at)lerctr(dot)org> writes:
> Can one of you knowledgeable people tell me why current CVS as of
> a week ago would have the backend running this query grow to
> 600 meg+?
Sounds like there's still a memory leak in there somewhere, but the
query looks fairly harmless. Could we see enough info to reproduce
this? (Table declarations, explain output, etc) Another useful
attack would be to let the query run awhile, then set a breakpoint
at sbrk(). Stack traces from the first few hits of the breakpoint
would give a pretty good indication of where the leak is, probably.
regards, tom lane