Hi, On Fri, Apr 16, 2021, at 20:18, Justin Pryzby wrote: > On Fri, Apr 16, 2021 at 09:48:54PM -0500, Justin Pryzby wrote: > > On Fri, Apr 16, 2021 at 07:17:55PM -0700, Andres Freund wrote: > > > Hi, > > > > > > On 2020-12-18 17:56:07 -0600, Justin Pryzby wrote: > > > > I'd be happy to run with a prototype fix for the leak to see if the > > > > other issue > > > > does (not) recur. > > > > > > I just posted a prototype fix to > > > https://www.postgresql.org/message-id/20210417021602.7dilihkdc7oblrf7%40alap3.anarazel.de > > > (just because that was the first thread I re-found). It'd be cool if you > > > could have a look! > > > > This doesn't seem to address the problem triggered by the reproducer at > > https://www.postgresql.org/message-id/20210331040751.gu4...@telsasoft.com > > (sorry I didn't CC you) > > I take that back - I forgot that this doesn't release RAM until hitting a > threshold. >
Phew. > Without the patch, it looks like: > > $ yes 'SET jit_above_cost=0; SET jit_inline_above_cost=0; SET jit=on; > SET client_min_messages=debug; SET log_executor_stats=on; SELECT > cfn();' |head -11 |psql -h /tmp postgres 2>&1 |grep 'max resident' > ! 61820 kB max resident size > ! 65020 kB max resident size > ! 68812 kB max resident size > ! 71152 kB max resident size > ! 76820 kB max resident size > ! 78760 kB max resident size > ! 81140 kB max resident size > ! 83520 kB max resident size > ! 93084 kB max resident size > ! 94756 kB max resident size > ! 96416 kB max resident size > > With the patch and #define LLVMJIT_LLVM_CONTEXT_REUSE_MAX 1, it looks like > this: > > ! 61436 kB max resident size > ! 61572 kB max resident size > ! 63236 kB max resident size > ! 63236 kB max resident size > ! 63556 kB max resident size > ! 63556 kB max resident size > ! 63880 kB max resident size > ! 65416 kB max resident size > ! 65416 kB max resident size > ! 65416 kB max resident size > ! 65416 kB max resident size If you set the define to much lower, how low does this get? I didn't see an easy way to make the recreation of the context memory usage dependant, but that'd of course be much better than an #iterations based approach. I'll play around with your test tomorrow or so, if you don't, but now the next step is risotto... Regards, Andres