Hi all, There was a bug in guththila_un_init function and guththila_reader_free function which will cause memory leaks. Should I commit the fix into the trunk. I think these fixes must go with the 1.3.1.
Thanks Milinda On Sat, Apr 19, 2008 at 10:27 AM, Manjula Peiris <[EMAIL PROTECTED]> wrote: > Hi Damitha, > > With HTTPD we are now creating the local pool (actually request pool) > with our own apr_allocator. See the following two function calls at the > beginning of the mod_axis2.c axis2_handler function. > > apr_allocator_create(&local_allocator); > apr_pool_create_ex(&local_pool, NULL, NULL, local_allocator); > > The two NULL values passing are the parent pool and the callback > function inorder to handle pool creation failure. > > At the end of this axis2_handle function we free both apr_pool and the > apr_allocator using the following two calls. > > apr_pool_destroy(local_pool); > apr_allocator_destroy(local_allocator); > > What makes sandesha to crash is the call to apr_allocator_destroy. But > that call is necessary when handling large size of requests frequently. > (specially quite large attachments). > > Because sandesha starts threads and since these threads remains after > the request the crashing happens. The solution for this is we need to > keep a pointer to this apr_allocator_t * (here local_allocator) and a > ref count to keep the number of references. Inside Sandesha sender > thread we can increment the ref count. This destroy function calls only > if the ref count is zero. we can keep the ref_count and the pointer to > the apr_allocator inside axutil_allocator. Then checking for the ref > count and calls to destroy functions are wrapped in a macro called > AXIS2_POOL_ALLOCATOR_DESTROY which is similar to AXIS2_FREE. Both > Sandesha and mod_axis2 calls this macro instead of calling destroy > functions straightway as we are doing now. > > Thanks, > -Manjula. > > > > On Sat, 2008-04-19 at 08:19 +0530, Damitha Kumarage wrote: > > Manjula Peiris wrote: > > > > >On Fri, 2008-04-18 at 11:40 +0530, Damitha Kumarage wrote: > > > > > > > > >>Axis2C-1079 is still open. That means Sandesha2/C dual channel does > not > > >>work with Apache2. > > >>This problem started due to a change to Axis2/C code. We cannot > release > > >>without solving this. > > >> > > >> > > > > > >This may fails because now we are freeing the apr local pool. I did fix > > >this locally in my machine but it needs more testing. Since it is now > so > > >close to the release I will commit the stuff after the release. > > > > > > > > +1 for doing this after 1.3.1 release. Could you please explain about > > the changes involved so that I too can have a look at it. > > thanks > > Damitha > > > > >Thanks, > > >-Manjula. > > > > > > > > > > > >>thanks > > >>Damitha > > >> > > >>Dinesh Premalal wrote: > > >> > > >> > > >> > > >>>Hi All, > > >>> > > >>> I have packed and uploaded Apache Axis2/C 1.3.1 RC2 here > > >>> [1]. Please test and send your feedback. > > >>> > > >>>thanks, > > >>>Dinesh > > >>> > > >>>1.http://people.apache.org/~dinesh/release/Axis2C-1.3.1-RC2/<http://people.apache.org/%7Edinesh/release/Axis2C-1.3.1-RC2/> > > >>> > > >>> > > >>> > > >>> > > >>> > > >>--------------------------------------------------------------------- > > >>To unsubscribe, e-mail: [EMAIL PROTECTED] > > >>For additional commands, e-mail: [EMAIL PROTECTED] > > >> > > >> > > >> > > > > > > > > >--------------------------------------------------------------------- > > >To unsubscribe, e-mail: [EMAIL PROTECTED] > > >For additional commands, e-mail: [EMAIL PROTECTED] > > > > > > > > > > > > > > > > > > --------------------------------------------------------------------- > > To unsubscribe, e-mail: [EMAIL PROTECTED] > > For additional commands, e-mail: [EMAIL PROTECTED] > > > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [EMAIL PROTECTED] > For additional commands, e-mail: [EMAIL PROTECTED] > > -- http://inf-dimensions.blogspot.com "Infinite Dimensions" http://wsaxc.blogspot.com "Web Services With Axis2/C"
