mogilefs.py threading coredump patch

[Justin Mason]

Hi all --

The Python mogilefs.py class at http://www.albany.edu/~ja6447/mogilefs.py
has an issue -- when called from threaded code on Linux, it'll
intermittently dump core due to unhandled, stray SIGALRM signals.

This thread notes the fix:
http://python-forum.org/pythonforum/viewtopic.php?f=5&t=9395 .
pycurl needs the NOSIGNAL option to operate thread-safely.

http://curl.haxx.se/mail/lib-2002-12/0103.html notes that NOSIGNAL
is required to abort slow DNS name resolution. Is it safe to
assume that mogstored nodes' names will always resolve speedily?
fwiw, I would guess so, and it's better than coredumping, but ymmv ;)

patch for mogilefs.py is below. cheers,

--j.

--- ../external/linux/zips/mogilefs.2007-09-05.py 2008-09-25
15:46:07.000000000 +0000
+++ mogilefs.new.py 2008-10-16 13:13:46.000000000 +0000
@@ -668,6 +668,11 @@
curl.setopt(pycurl.LOW_SPEED_TIME, 5)
curl.setopt(pycurl.LOW_SPEED_LIMIT, 1)
curl.setopt(pycurl.WRITEFUNCTION, ofp.write)
+
+ # Oct 16 2008 jm: avoid segfaults when
multithreaded, see
+ # http://python-forum.org/pythonforum/viewtopic.php?f=5&t=9395
+ curl.setopt(pycurl.NOSIGNAL, 1)
+
curl.perform()
curl.close()
if fp is None:

[Paul Baker™]

The recommended practice is to use IPs for nodenames instead of hostnames, so dns shouldn't be an issue at all, if you are following best practice.

[Justin Mason]

Sounds reasonable. ;)

(PS: slight typo in the original mail. where I said 'NOSIGNAL is
required to abort slow DNS name resolution', I got that backwards --
it's 'NOSIGNAL=0', ie. the use of SIGALRM, is required to timeout DNS
lookups.)

--j.

On Oct 16, 5:37 pm, "Paul Baker™" <bak...@google.com> wrote:
> The recommended practice is to use IPs for nodenames instead of hostnames,
> so dns shouldn't be an issue at all, if you are following best practice.
>
> On Thu, Oct 16, 2008 at 10:22 AM, Justin Mason <jma...@gmail.com> wrote:
>
> > Hi all --
>

> > The Python mogilefs.py class athttp://www.albany.edu/~ja6447/mogilefs.py

> > has an issue -- when called from threaded code on Linux, it'll
> > intermittently dump core due to unhandled, stray SIGALRM signals.
>
> > This thread notes the fix:
> >http://python-forum.org/pythonforum/viewtopic.php?f=5&t=9395.
> > pycurl needs the NOSIGNAL option to operate thread-safely.
>

> >http://curl.haxx.se/mail/lib-2002-12/0103.htmlnotes that NOSIGNAL