On Sep 17, 2010, at 8:24 AM, mdipierro wrote:
> 
> I will take the patch and you may be fine if your thread does not
> access db.

Here's a slightly more complete patch. This should maintain backward 
compatibility with apps that don't use app-specific routing. One that does can 
call rewrite.select(app='name') at the start of a new thread.

This won't help when globals are based on threading.local, though; that's a 
nastier problem.


diff -r 3645edf51f22 gluon/rewrite.py
--- a/gluon/rewrite.py  Fri Sep 17 08:40:21 2010 -0500
+++ b/gluon/rewrite.py  Fri Sep 17 08:50:17 2010 -0700
@@ -150,14 +150,15 @@
     logger.debug('%s: [%s] -> %s (not rewritten)' % (tag, key, default))
     return (default, query, original_uri)
 
-def select(e=None):
+def select(env=None, app=None):
     """
     select a set of rewrite params for the current request
     called from main.wsgibase before any URL rewriting
     """
-    app = None
-    if e and params.routes_app:
-        (app, q, u) = filter_uri(e, params.routes_app, "routes_app")
+    if app:
+        thread.routes = params_apps.get(app, params)
+    elif env and params.routes_app:
+        (app, q, u) = filter_uri(env, params.routes_app, "routes_app")
         thread.routes = params_apps.get(app, params)
     else:
         thread.routes = params # default to base rewrite parameters
@@ -180,6 +181,8 @@
 
 def filter_out(url, e=None):
     "called from html.URL to rewrite outgoing URL"
+    if not hasattr(thread, 'routes'):
+        select()    # ensure thread.routes is set (for application threads)
     if thread.routes.routes_out:
         items = url.split('?', 1)
         if e:


> 
> On Sep 17, 10:20 am, Jonathan Lundell <jlund...@pobox.com> wrote:
>> On Sep 16, 2010, at 10:44 PM, mdipierro wrote:
>> 
>> 
>> 
>>> Jonathan,
>> 
>>> I misunderstood your problem. You should not spawn threads from a
>>> web2py app. This is not just because the current routing mechanism
>>> does not support it. This is a very general with all web applications
>>> because threads are managed by the web server which starts/stops and
>>> kills them. If your thread spawns a new thread and the parent is
>>> killed by the web server you may end up with a memory leak.
>> 
>>> Please explain what you are trying to achieve, perhaps show us some
>>> code, and I am sure there is another way.
>> 
>> Here's a quick patch to rewrite.filter_out that should fix the problem for 
>> URL() as long as app-specific routes are not in use:
>> 
>>  def filter_out(url, e=None):
>>      "called from html.URL to rewrite outgoing URL"
>> +    if not hasattr(thread, 'routes'):
>> +        select()    # ensure thread.routes is set (for application threads)
>>      if thread.routes.routes_out:
>>          items = url.split('?', 1)
>>          if e:
>> 
>> Massimo, this will be a little confusing, since there are two Jonathans on 
>> this (message) thread, and both of us are using application threads.
>> 
>> Here's my case:
>> 
>> The application is a manager for a collection of servers, from tens to 
>> possibly hundreds. On the central management page, I create a table that 
>> summarizes the status of the servers. I get the status by sending an xmlrpc 
>> query to each server, and the response can take several seconds (say 2-10).
>> 
>> Because serializing these requests can take too long, I create a thread per 
>> server to make the query and wait for the response. CPU time for such a 
>> request is nil, so all the threads complete in approximately the time for 
>> the slowest one.
>> 
>> The code is very simple and very clean.
>> 
>> No doubt I could accomplish something like it with Ajax, and effectively a 
>> request thread for each server. But I hesitate to rely on the web server's 
>> thread pool, assuming it has one big enough, and an Ajax solution wouldn't 
>> be so straightforward, I don't think.
>> 
>> 
>> 
>>> Massimo
>> 
>>> On Sep 16, 2:26 pm, "Jonathan Z." <jzem...@gmail.com> wrote:
>>>> I'm kicking off a threaded process.  As part of the "run" method, I'm
>>>> calling: env("application", import_models=True) in order to work with
>>>> the web2py environment inside the context of my thread.  As a result,
>>>> models are parsed and the following exception is raised:
>> 
>>>> Traceback (most recent call last):
>>>>   File ".../web2py/gluon/restricted.py", line 188, in restricted
>>>>     exec ccode in environment
>>>>   File "applications/app/models/db.py", line 32, in <module>
>>>>     auth = Auth(globals(), db)
>>>>   File ".../web2py/gluon/tools.py", line 804, in __init__
>>>>     self.settings.login_url = self.url('user', args='login')
>>>>   File ".../web2py/gluon/tools.py", line 762, in url
>>>>     f=f, args=args, vars=vars)
>>>>   File ".../web2py/gluon/html.py", line 228, in _URL
>>>>     return URL(*args, **kwargs)
>>>>   File ".../web2py/gluon/html.py", line 206, in URL
>>>>     return XML(rewrite.filter_out(url, env))
>>>>   File ".../web2py/gluon/rewrite.py", line 183, in filter_out
>>>>     if thread.routes.routes_out:
>>>> AttributeError: 'thread._local' object has no attribute 'routes'
>> 
>>>> As soon as URLs are parsed during Auth initialization, the thread hits
>>>> a case where an invalid "routes" attribute is dereferenced within
>>>> rewrite.py.  My threaded code was working until the following change
>>>> was rolled into 1.84.x: "moved DAL and routes to thread.local"
>> 
>> 


Reply via email to