On Fri, Jun 9, 2017 at 9:12 AM, Frits Jalvingh <jal@etc.to> wrote:
> Hi John,
>
> Yes, I was aware and amazed by that ;) It is actually the fetch size in
> combination with autocommit being on; that dies the sweet OOM death as soon
> as the table gets big.
>
> But Postgres read performance, with autocommit off and fetch size arond 64K,
> is quite OK. But it's good to get this mentioned a lot, because as you said
> you can spend quite some time wondering about this!
No production db server should have the oom killer enabled.
