On Fri, Oct 07, 2022 at 06:00:21PM -0500, Chris Muller wrote:
> I'm 98% certain the issue is the new memory monitor running
> 
>    OSProcess outputOf: 'cat /proc/meminfo'
> 
> every 5 seconds.  It automatically disables on any Error.  Unfortunately,
> #outputOf: appears to have a slow resource leak that causes a slow grind to
> a halt -- never signaling any Errors, and never crashing.  My attempt to
> make the server more resilient has ironically resulted in it being more
> fragile.
> 
> So, I think I'm going to try going back to my old method which uses
> #command:
> 
>      OSProcess command: 'cat /proc/meminfo -> meminfo.out'
> 
> and then parsing meminfo.out.  I'm hopeful #command: will not have the same
> issue.
> 

Hi Chris,

That's a lot of OSProcess/CommandShell invocations, over 17 thousand
per day. Each one is setting up three pipe connections (six file
descriptors), and each requires the unix signal handler to clean up
the exited child and close open pipe handles. I would not be too
surprised if intermittent problems were to accumulate over time at
that rate (missed unix signals under heavy load? I don't know).

If the file descriptor leak is associated with OSProcess, you will see
pipe entries accumulating in /proc/<pid>/fd. If the leak is coming from
SqueakSource stuff, you will see socket handles. Let's keep an eye on
it and see which it is. I think your intuition is right but we may as
well confirm it.

Dave