[asterisk-dev] [Code Review] 3377: ref count logs: Redo structure of log file, provide a python debugging tool

Mark Michelson reviewboard at asterisk.org
Thu Mar 20 09:35:50 CDT 2014


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviewboard.asterisk.org/r/3377/#review11295
-----------------------------------------------------------



/branches/1.8/contrib/scripts/refcounter.py
<https://reviewboard.asterisk.org/r/3377/#comment20936>

    One of the refcounting errors that has occurred in the past is sort of an "antileak" if you will. In other words, decrementing refcounts has been too aggressive and caused the object to get destroyed while other threads still believe they have a reference to the object. The usually manifests as the object being destroyed, then having its refcount incremented to 1, then decremented back to 0, and then having its destructor called again.
    
    This condition can be checked in this function. In your loop, if the object referenced on the current line is already in the finished_objects list, then we've got a problem. These types of objects should be removed from the finished_objects list and placed into their own separate list.



/branches/1.8/contrib/scripts/refcounter.py
<https://reviewboard.asterisk.org/r/3377/#comment20937>

    To add onto Corey's comment, if you add a third classification of objects based on overaggressive ref decrementing, then that should also have its own word here as well.


- Mark Michelson


On March 19, 2014, 6:22 p.m., Matt Jordan wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviewboard.asterisk.org/r/3377/
> -----------------------------------------------------------
> 
> (Updated March 19, 2014, 6:22 p.m.)
> 
> 
> Review request for Asterisk Developers, Corey Farrell and wdoekes.
> 
> 
> Repository: Asterisk
> 
> 
> Description
> -------
> 
> Note: while an improvement to Asterisk, this patch only affects Asterisk when compiled in -dev-mode. Since it has benefit only for developers looking to fix bugs, I'm proposing it for Asterisk 1.8+. The removal of refcounter should be done in trunk only.
> 
> Asterisk uses reference counted objects. A lot. As their use has spread, the bugs related to reference counting errors has grown as well. Central to resolving reference counting issues is the usage of REF_DEBUG; unfortunately, REF_DEBUG has had several problems:
> (1) It could not be enabled through menuselect
> (2) When not enabled globally, updates to objects were often lost, resulting in insufficient data to resolve problems
> (3) The format of the ref debug log file was not suitable for parsing
> 
> This patch changes REF_DEBUG in the following ways:
> (1) It makes REF_DEBUG a meneselect item when Asterisk is compiled with --enable-dev-mode
> (2) The ref debug log file is now created in the AST_LOG_DIR directory. Every run will now blow away the previous run (as large ref files sometimes caused issues). We now also no longer open/close the file on each write, instead relying on fflush to make sure data gets written to the file (in case the ao2 call being performed is about to cause a crash)
> (3) It goes with a comma delineated format for the ref debug file. This makes parsing much easier.
> (4) It throws out the refcounter utility and goes with a python script instead.
> 
> 
> Diffs
> -----
> 
>   /branches/1.8/utils/refcounter.c 410891 
>   /branches/1.8/utils/Makefile 410891 
>   /branches/1.8/main/astobj2.c 410891 
>   /branches/1.8/contrib/scripts/refcounter.py PRE-CREATION 
>   /branches/1.8/build_tools/cflags-devmode.xml 410891 
> 
> Diff: https://reviewboard.asterisk.org/r/3377/diff/
> 
> 
> Testing
> -------
> 
> Things spit out ref logs and the script parses them. Example below:
> 
> ==== Object 0x21ed9b8 history ====
> features.c[5427]:create_parkinglot 1  - [**constructor**]
> features.c[5707]:build_parkinglot +1  - [1]
> features.c[5392]:parkinglot_unref -1  - [2]
> features.c[6358]:build_dialplan_useage_map +1  - [1]
> features.c[6358]:build_dialplan_useage_map -1  - [2]
> features.c[4985]:find_parkinglot +1  - [1]
> features.c[5392]:parkinglot_unref -1  - [2]
> features.c[6358]:build_dialplan_useage_map +1  - [1]
> features.c[6358]:build_dialplan_useage_map -1  - [2]
> features.c[4955]:do_parking_thread +1  - [1]
> features.c[4957]:do_parking_thread -1  - [2]
> astobj2.c[955]:cd_cb_debug -1 deref object via container destroy - [1]
> astobj2.c[955]:cd_cb_debug -1 deref object via container destroy - [**call destructor**]
> 
> 
> Thanks,
> 
> Matt Jordan
> 
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.digium.com/pipermail/asterisk-dev/attachments/20140320/41607e89/attachment-0001.html>


More information about the asterisk-dev mailing list