out of memory while building reference manual

[Dan Drake]

I'm trying to build the reference manual for 4.3 on a virtual machine
with a gigabyte of memory, but it fails:

preparing documents... done
writing output... [ 0%] algebras
writing output... [ 0%] arithgroup

Exception occurred:
File "/home/drake/sage-4.3/local/lib/python/subprocess.py", line 1009, in _execute_child
self.pid = os.fork()
OSError: [Errno 12] Cannot allocate memory

However, I have about a gigabyte available:

$ free
total used free shared buffers cached
Mem: 1028696 466672 562024 0 26052 360580
-/+ buffers/cache: 80040 948656
Swap: 0 0 0

Does building the reference manually truly need that much memory? I know
I don't have any swap, but I thought 900+ megabytes would be enough for
Sphinx to render some html!

Thanks,

Dan

--
--- Dan Drake
----- http://mathsci.kaist.ac.kr/~drake
-------

[Pat LeSmithe]

On 12/29/2009 07:28 PM, Dan Drake wrote:
> I'm trying to build the reference manual for 4.3 on a virtual machine
> with a gigabyte of memory, but it fails:

> [...]

> Does building the reference manually truly need that much memory? I know
> I don't have any swap, but I thought 900+ megabytes would be enough for
> Sphinx to render some html!

When I build the manual on boxen.math.washington.edu, memory use peaks
at about

VIRT RES SHR S %CPU %MEM TIME+ COMMAND
1387m 647m 26m R 101 0.5 13:47.10 python

I find that most (~580 MB) of this absorbed during Sphinx's reading phase

http://sphinx.pocoo.org/ext/tutorial.html

I'm not very familiar with how virtual memory is allocated, but it could
be that your process hit a shell or OS limit just as the writing phase
began. What happens if you add some (or a lot of) swap space?

There are timings on building the manual with inherited and/or
underscored members at

http://trac.sagemath.org/sage_trac/ticket/7549#comment:5

[Dan Drake]

On Wed, 30 Dec 2009 at 12:43AM -0800, Pat LeSmithe wrote:
> On 12/29/2009 07:28 PM, Dan Drake wrote:
> > I'm trying to build the reference manual for 4.3 on a virtual machine
> > with a gigabyte of memory, but it fails:
> > [...]
> > Does building the reference manually truly need that much memory? I know
> > I don't have any swap, but I thought 900+ megabytes would be enough for
> > Sphinx to render some html!
>
> When I build the manual on boxen.math.washington.edu, memory use peaks
> at about
>
> VIRT RES SHR S %CPU %MEM TIME+ COMMAND
> 1387m 647m 26m R 101 0.5 13:47.10 python
>
> I find that most (~580 MB) of this absorbed during Sphinx's reading phase
>
> http://sphinx.pocoo.org/ext/tutorial.html
>
> I'm not very familiar with how virtual memory is allocated, but it could
> be that your process hit a shell or OS limit just as the writing phase
> began. What happens if you add some (or a lot of) swap space?

My solution was to add 512 megs of RAM, which solved (several!) doctest
problems and allowed the ref manual to build. It looks like a gigabyte
of memory (with no swap space or anything) just isn't quite enough. I
did look at "ulimit -a" and there seemed to be no limits on anything
memory- or process-related.

Anyway, it works now...