[Home]

Summary:ASTERISK-16129: iax2-parser eating memory if not LOW_MEMORY set.
Reporter:Eric Dantie (edantie)Labels:
Date Opened:2010-05-22 05:32:09Date Closed:2011-06-07 14:00:56
Priority:MajorRegression?No
Status:Closed/CompleteComponents:Channels/chan_iax2
Versions:Frequency of
Occurrence
Related
Issues:
Environment:Attachments:( 0) iaxcall.txt
Description:I think there a problem in iax-parser.c: it's allocating more memory than it release when LOW_MEMORY is not set... It applies at least to all 1.6 up to trunk.

Try to explain:
I have added comments in iax_frame_free to see when it's adding a memory space to pool.

like that at line 1192 of 1.6.2 svn:

       if (iax_frames->size < FRAME_CACHE_MAX_SIZE) {
+++                ast_verb(3, "Adding new frame to cache %d (length:%d)\n",iax_frames->size, fr->afdatalen);

and one comment in iax_frame_new, line 1105
                       if (fr->afdatalen >= datalen) {
                               size_t afdatalen = fr->afdatalen;
                               AST_LIST_REMOVE_CURRENT(list);
                               iax_frames->size--;
                               memset(fr, 0, sizeof(*fr));
                               fr->afdatalen = afdatalen;
+++                                ast_verb(3,"Got frame from cache! new size:%d\n",iax_frames->size);
                               break;


The extension is receiving an iax call and do another iax call
recompile and start doing a iax call.
What I've got is:


   -- Accepting AUTHENTICATED call from 172.16.222.1:
      > requested format = gsm,
      > requested prefs = (gsm|alaw),
      > actual format = alaw,
      > host prefs = (alaw|gsm),
      > priority = mine
   -- Executing [9006@fromiax:1] MSet("IAX2/lpa-1993", "__PHONENUMBER=9006") in new stack
   -- Executing [9006@fromiax:2] Dial("IAX2/lpa-1993", "IAX2/lpa:contactel@172.16.222.1/61010") in new stack
   -- Called lpa:contactel@172.16.222.1/61010
   -- Call accepted by 172.16.222.1 (format alaw)
   -- Format for call is alaw
   -- IAX2/172.16.222.1:4569-864 is proceeding passing it to IAX2/lpa-1993
   -- Adding new frame to cache 0 (length:160)
   -- Adding new frame to cache 0 (length:160)
   -- Adding new frame to cache 0 (length:160)
   -- Adding new frame to cache 0 (length:160)
   -- Adding new frame to cache 0 (length:160)
   -- Got frame from cache! new size:0
   -- Adding new frame to cache 0 (length:160)

Will attach full extension log

The problem is that it trying to add 5 frames to spool without using it...

I think there should be some locking in iax free to avoid the problem...
Comments:By: Russell Bryant (russell) 2010-06-03 15:43:57

This is not a bug.  chan_iax2 uses a pool of threads for processing.  You're seeing the frame cache populate across the thread pool, not leaks within a single thread.
By: Eric Dantie (edantie) 2010-06-03 16:40:22

Sorry not agreed:

if adding to cache, next call to iax_frame_free should increment the size of iax_frames which is not the case until 6 calls to the function. But it has allocated each time memory for the cache.

Shouldn't it be :
- IAX2/172.16.222.1:4569-864 is proceeding passing it to IAX2/lpa-1993
   -- Adding new frame to cache 0 (length:160)
   -- Adding new frame to cache 1 (length:160)
   -- Adding new frame to cache 2 (length:160)
   -- Adding new frame to cache 3 (length:160)
   -- Adding new frame to cache 4 (length:160)
   -- Got frame from cache! new size:3
   -- Adding new frame to cache 3 (length:160)


By: Eric Dantie (edantie) 2010-06-03 16:44:45

Oops, I've understood: there are 5 threads...

OK.
By: Russell Bryant (russell) 2010-06-03 18:00:32

Yep, it's a thread-local cache