Hi, summary of my debugging session:
- The freeze happens once you install python-lxml, there is no proble mwith other etree implementations ogcserver imports. - I can reproduce it with just mod_python + ogcserver + mapnik (0.7 in ubuntu), no wsgi involved - It looks the problem is a deadlock involving the GIL. I tried moving the load_XML call outside the __init__ function, so that it was not under an import call, but no luck. Some other code must be acquiring the GIL and not releasing it. I'll try to debug the GIL to find the problem. As you can see in the backtrace, lxml etree implementation will call PyGILState_Ensure, which will cause the thread to sleep, waiting for some other code to release the GIL => deadlock because that will never happen. Here is some inspiration: http://bugs.python.org/issue7736 And some documentation: http://wiki.python.org/moin/boost.python/HowTo (gdb) backtrace #0 0xb773b430 in __kernel_vsyscall () #1 0xb769b245 in sem_wait@@GLIBC_2.1 () from /lib/tls/i686/cmov/libpthread.so.0 #2 0xb65fce10 in PyThread_acquire_lock (lock=0xb80f1568, waitflag=1) at ../Python/thread_pthread.h:349 #3 0xb65cc798 in PyEval_RestoreThread (tstate=0xb809b9a0) at ../Python/ceval.c:353 #4 0xb65efa90 in PyGILState_Ensure () at ../Python/pystate.c:592 #5 0xb237465e in __pyx_f_4lxml_5etree__local_resolver (__pyx_v_c_url=0xb855194c "/home/manel/ogcserver/world.xml", __pyx_v_c_pubid=0x0, __pyx_v_c_context=0xb855b4c8) at src/lxml/lxml.etree.c:62627 #6 0xb6931343 in xmlLoadExternalEntity () from /usr/lib/libxml2.so.2 #7 0xb6920891 in xmlCtxtReadFile () from /usr/lib/libxml2.so.2 #8 0xb53cbc64 in mapnik::libxml2_loader::load(std::string const&,boost::property_tree::basic_ptree<boost::property_tree::ptree_traits<char> >&) () from /usr/lib/libmapnik.so.0.7 #9 0xb53ca537 in mapnik::read_xml2(std::string const&, boost::property_tree::basic_ptree<boost::property_tree::ptree_traits<char> >&) () from /usr/lib/libmapnik.so.0.7 #10 0xb538a7e4 in mapnik::load_map(mapnik::Map&, std::string const&, bool) () from /usr/lib/libmapnik.so.0.7 #11 0xb55bd83c in load_map_overloads::non_void_return_type::gen<boost::mpl::vector4<void, mapnik::Map&, std::string const&, bool> >::func_1(mapnik::Map&, std::string const&, bool) () from /usr/lib/pymodules/python2.6/mapnik/_mapnik.so ... -- Manel ________________________________ De: [email protected] [mailto:[email protected]] En nombre de Dane Springmeyer Enviado el: sábado, 02 de octubre de 2010 0:13 Para: [email protected] CC: [email protected] Asunto: Re: [Mapnik-users] Problem with Mapniks OGCServer Okay, I think I found it. Putting this in your WSGI config should fix it: WSGIApplicationGroup %{GLOBAL} More details here: https://techknowhow.library.emory.edu/blogs/branker/2010/07/30/django-lxml-wsgi-and-python-sub-interpreter-magic So, this should allow the ogcserver to play nicely within mod_wsgi with lxml installed. If this fixes the hang for you then we can investigate more. If not, then make sure you are only running one instance of the ogcserver within Apache. Dane On Oct 1, 2010, at 10:59 AM, [email protected] wrote: Hi, I forgot something... when uninstalling lxml, mod_wsgi can actually connect to mapniks ogcserver (ogcserver welcome appears). as Dane said earlier in this thread, mapnik falls back to standard xml support. at least this machine should... but with a GetMap request the browser still returns no map, the text of the requested URL gets rendered as an unformatted image instead. Daniel Am 30.09.2010 18:14, schrieb Dane Springmeyer: > > On Sep 1, 2010, at 3:57 AM,<[email protected]> <[email protected]> wrote: > >> Hi, >> >> Just tried and I can reproduce the same problem here. Ubuntu server 9.10, Mapnik 0.7.1, mod_python will get stuck in Mapnik's load_map in WMS.py. >> >> Strange enough, I'm using the *same* code under wsgi with no problems. >> > > Daniel look ^^^. So Manel, what Apache version, prefork or worker, and what mod_wsgi and python versions? > > Btw, I forgot to mention the ticket we once had that tracked this issue (at lease specific to GetCaps): > > http://trac.mapnik.org/ticket/464 > > It would be good to add any additional info we find to that ticket thread. For instance, short of understanding the exact problem within lxml, hopefully we can recommend a certain version of python/apache/mod_wsgi which will work around the problem. > > - Dane GKSS-Forschungszentrum Geesthacht GmbH Max-Planck-Straße 1 21502 Geesthacht Deutschland/Germany Geschäftsführer/Board of Management: Prof. Dr. Wolfgang A. Kaysser, Dipl.-Ing. Michael Ganß Vorsitzender des Aufsichtsrates/Chairman of the Supervisory Board: Ministerialrat Wilfried Kraus Amtsgericht Lübeck HRB 285 GE (Register Court) Internet: http://www.gkss.de ACHTUNG - ATTENTION Die GKSS-Forschungszentrum Geesthacht GmbH erhält einen neuen Namen und heißt ab dem 1. November 2010 offiziell The GKSS-Forschungszentrum Geesthacht GmbH will change its name and will as of 1st November 2010 be officially called Helmholtz-Zentrum Geesthacht Zentrum für Material- und Küstenforschung GmbH _______________________________________________ Mapnik-users mailing list [email protected] https://lists.berlios.de/mailman/listinfo/mapnik-users _______________________________________________ Mapnik-users mailing list [email protected] https://lists.berlios.de/mailman/listinfo/mapnik-users
