On Tue, Nov 10, 2020 at 05:38:08PM -0800, Andres Freund wrote:
> Hi,
>
> On 2020-11-10 23:45:16 +0100, Kurt Roeckx wrote:
> > On Tue, Nov 10, 2020 at 08:50:39PM +0100, Kurt Roeckx wrote:
> > > > There's one known (slow) memory leak in the JIT code / LLVM. Could you
> > > > check if the issue vanishes if you disable JIT (jit = 0)?
> > >
> > > I've just restarted it with jit = 0.
> >
> > It's been about 3 hours since the restart, and it looks much
> > better, it seems to be solved.
>
> Hm, darn. Any chance you could check if the leak is present if you turn
> on jit again, but disable inlining with jit_inline_above_cost=-1? If
> that still fixes the leak I think I know the issue / have a reproducer
> already...
So after about 14 hours, it shows:
$ grep kB /proc/32117/status
VmPeak: 4412436 kB
VmSize: 4389988 kB
VmLck: 0 kB
VmPin: 0 kB
VmHWM: 4261196 kB
VmRSS: 4239936 kB
RssAnon: 6112 kB
RssFile: 7340 kB
RssShmem: 4226484 kB
VmData: 5412 kB
VmStk: 132 kB
VmExe: 5388 kB
VmLib: 13032 kB
VmPTE: 8532 kB
VmSwap: 0 kB
HugetlbPages: 0 kB
The munin graph is also very flat now, so with jit=0, it seems
fixed.
I've just restarted it with
jit=1
jit_inline_above_cost=-1
I will let you know.
Kurt
