On Mon, Dec 4, 2017 at 9:39 AM, Tom Lane <tgl@xxxxxxxxxxxxx> wrote: > Ah. The problem here is that "json_rmq->>'totalSize'" leaks some memory > on each execution, and it's executed again for each row produced by the > json_array_elements() SRF, and the memory can't be reclaimed until we've > finished the full output cycle for the SRF. So the leakage (which is > more or less of the size of the JSON value, I think) accumulates across > 150K executions in this example. > > This is fixed as of v10. It seems impractical to do anything about it > in previous release branches, although you could reformulate your query to > avoid it by not having any other expression evaluations occurring in the > same tlist as the SRF. Something like this should work: Yeah, I agree with that. One similar leak has actually been fixed with this commit, and the infrastructure of v10 has made this fix dead simple: commit: 0c25e9652461c08b5caef259a6af27a38707e07a author: Tom Lane <tgl@xxxxxxxxxxxxx> date: Fri, 6 Oct 2017 14:28:42 -0400 Fix intra-query memory leakage in nodeProjectSet.c. https://www.postgresql.org/message-id/20171005230321.28561.15927@xxxxxxxxxxxxxxxxxxxxxxx nodeProjectSet.c really makes tuple-level memory handling way easier based on my studies of this code. -- Michael
