problem pushing large changeset to remote repo.

[Matt Mackall]

On Tue, 2008-01-22 at 14:59 -0800, Kelly O'Hair wrote:
> AIX used to (maybe still does?) send a SIGDANGER to large long running processes
> when a certain low level of resources is detected.

Only when swap is nearly exhausted, yes.

>  The default action for an
> AIX process getting a SIGDANGER was to just die and disappear, the process itself
> had to register a signal handler to avoid the default SIGDANGER action and
> keep it from being killed arbitrarily.
> They apparently did this to keep the OS up and running, better AIX 'up time' stats?
> Anyway, a pretty bizarre thing to do.

Well, you've got to do something. Killing just one process has a fairly
high likely as killing lots of them, as that process is often going to
be the one you actually care about.

> I wonder if this could be related. mzscheme mentions it under it's -p or --persistent
> option: http://www.informatik.uni-kiel.de/~scheme/doc/mzscheme/node190.htm
> 
> But does anyone even run AIX anymore?

Oddly enough, yes.

But what we're seeing here is likely an ugly userspace hack that's
scanning for anything that's using more than x% of CPU and killing it.
Other people have seen large pushes or pulls mysteriously die in the
middle and this is why.

-- 
Mathematics is the supreme nostalgia of our time.