Re: [Twisted-web] Different behavior of twistd with and without -n
"JP" == Jean-Paul Calderone <exarkun@divmod.com> writes: JP> A completely wild guess is that forking is confusing PyLucene in a JP> fatal way. Are you importing PyLucene in the .tac file itself? If so, JP> it may help to avoid doing this, so that no code from PyLucene even JP> gets a chance to run until after the process has already daemonized.
Hi JP. Thanks for the suggestion. Yes, the PyLucene import does happen as a result of an import in the .tac file. I just made some changes to delay the import until PyLucene is actually needed. That didn't work, and nor did further hiding the import by using __import__. Is it right that all the -n switch to twistd does is prevent the fork? Terry
On Thu, 24 May 2007 18:04:57 +0200, Terry Jones <terry@jon.es> wrote:
"JP" == Jean-Paul Calderone <exarkun@divmod.com> writes: JP> A completely wild guess is that forking is confusing PyLucene in a JP> fatal way. Are you importing PyLucene in the .tac file itself? If so, JP> it may help to avoid doing this, so that no code from PyLucene even JP> gets a chance to run until after the process has already daemonized.
Hi JP. Thanks for the suggestion.
Yes, the PyLucene import does happen as a result of an import in the .tac file. I just made some changes to delay the import until PyLucene is actually needed. That didn't work, and nor did further hiding the import by using __import__.
Is it right that all the -n switch to twistd does is prevent the fork?
It does a couple other minor things, like changing the default logging to go to stdout instead of twistd.log, but preventing the fork is the only big thing, yes. Jean-Paul
On 05:44 pm, exarkun@divmod.com wrote:
On Thu, 24 May 2007 18:04:57 +0200, Terry Jones <terry@jon.es> wrote:
>"JP" == Jean-Paul Calderone <exarkun@divmod.com> writes: JP> A completely wild guess is that forking is confusing PyLucene in a JP> fatal way. Are you importing PyLucene in the .tac file itself? If so, JP> it may help to avoid doing this, so that no code from PyLucene even JP> gets a chance to run until after the process has already daemonized.
Hi JP. Thanks for the suggestion.
Yes, the PyLucene import does happen as a result of an import in the .tac file. I just made some changes to delay the import until PyLucene is actually needed. That didn't work, and nor did further hiding the import by using __import__.
Is it right that all the -n switch to twistd does is prevent the fork?
It does a couple other minor things, like changing the default logging to go to stdout instead of twistd.log, but preventing the fork is the only big thing, yes.
The fork does some other things though, and it's hard to say which one is affecting the execution of your program, especially since PyLucene is doing a bunch of things at the machine-code level which are highly surprising to Python programmers. Have you tried running 'strace' on this process yet?
participants (3)
-
glyph＠divmod.com -
Jean-Paul Calderone -
Terry Jones