Re: port-amd64/54988: possible memory leaks/swap problems

[mlh%goathill.org@localhost (MLH)]

MLH wrote:
> From: mlh%goathill.org@localhost (MLH)
> 
>  Joerg Sonnenberger wrote:
>  > On Thu, Feb 20, 2020 at 04:15:01AM +0000, MLH wrote:
>  > >  
>  > >  Any idea on what is preventing the memory from being released and
>  > >  why it just locks up?
>  > 
>  > I don't know, I'm just saying that you are looking at the wrong pools.
>  > Look for those with high Npage, those are actually big.
>  
>  The problem continues to appear to be a filesystem-related issue
>  as I can do compute-bound jobs with no problem. Physical memory is
>  recovered as it normally does with no issue. Large or intensive
>  filesystem writes appear to cause the system to seize and it doesn't
>  even appear to require physical memory to be exhausted as it just
>  seized twice with vmstat showing over 2G of available physical
>  memory, and physical memory isn't showing to be recovered after
>  intensive fs writes when I stop it before the system seizes.
>  
>  Have any filesystem-related changes been done recently?

Such as (from CHANGES) :

        uvm: More precisely track clean/dirty pages, and change how they are
                indexed, speeding up fsync() on large files by orders of
                magnitude.  Original work done by yamt@. [ad 20200115]

as all was fine just before this change and then with kernels from
the last of Jan on, I am seeing the problems.