[SR-Users] Memory leak in 3.3.4
Daniel-Constantin Mierla
miconda at gmail.com
Wed Jul 31 12:54:46 CEST 2013
Hello,
revising the patch I noticed I was moving the initialization of the
variable after pushing it to perl environment (from the perl docs, the
variable should have been initialized after initializing the environment
-- what I tried to do with previous patch).
There is a new smaller patch to be added:
http://git.sip-router.org/cgi-bin/gitweb.cgi?p=sip-router;a=commitdiff_plain;h=3935fedf23f3bf2b6675182193cef6af3bbd903a
Practically, the line XPUSHs(m); has to be moved after the line with
SvREADONLY_on(SvRV(m));
Cheers,
Daniel
On 7/31/13 8:26 AM, David Cunningham wrote:
> Hi Daniel,
>
> We tried that patch, but Kamailio logged lots of errors like the
> following. The undefined value in question is $m, which should be the
> SIP message. Would you have any advice? Thanks.
>
> Jul 31 02:13:57 hostname /sbin/kamailio[21087]: ERROR: perl
> [openserxs.xs:1022]: perl error: Can't call method "pseudoVar" on an
> undefined value at Foo.pm line 247.#012
>
>
> On 25 July 2013 17:11, David Cunningham <dcunningham at voisonics.com
> <mailto:dcunningham at voisonics.com>> wrote:
>
> Hi Daniel,
>
> I'll suggest that to the customer. Thank you!
>
>
>
> On 25 July 2013 15:45, Daniel-Constantin Mierla <miconda at gmail.com
> <mailto:miconda at gmail.com>> wrote:
>
> Hello,
>
> can you try the attached patch? It's the same patch, just for
> two versions, one is for 3.3.x and the other for devel version
>
> It initializes the SIP message variable that is passed to perl
> after creating the temporary environment, so it is actually
> destroyed by the perl embedded interpreter.
>
> Cheers,
> Daniel
>
>
> On 7/25/13 1:29 AM, David Cunningham wrote:
>> Hi Daniel,
>>
>> The system is running Perl 5.8.8 on Red Hat Enterprise Linux
>> Server release 5.4. If I remember right programs running
>> under Valgrind can have issues, so I'm not sure if the
>> customer will want to do that. Ideally we'd do it on a test
>> system, but I'm not sure if we have any RHEL available.
>> I'll see what we can do. Thanks again.
>>
>>
>> On 25 July 2013 04:55, Daniel-Constantin Mierla
>> <miconda at gmail.com <mailto:miconda at gmail.com>> wrote:
>>
>> Hello,
>>
>> I would say that perl_exec() is the one with the highest
>> chances to be the reason for the leak. Next is line would
>> be db_mysql module, if liked with some custom mysql
>> client library, although even in this case will be unlikely.
>>
>> Back to perl, the module itself does not call any malloc,
>> so it might be the embedding Perl API that is not used
>> properly in the module.
>>
>> Can you use some testbed, set children=1 and run kamailio
>> under valgrind, then do some calls and see if it detects
>> the source of the leak?
>>
>> I'm not using the perl module, I will try to check it
>> whenever I get a chance in the next days. What version of
>> perl do you have installed?
>>
>> Cheers,
>> Daniel
>>
>>
>> On 7/24/13 10:31 AM, David Cunningham wrote:
>>> Hello,
>>>
>>> We don't do any kamctl commands at all. We do have
>>> various modules loaded, as follows. The primary
>>> functions we use Kamailio for are phone registrations
>>> through usrloc, and routing calls to Asterisk through
>>> logic contained in Perl via perl_exec().
>>> Thanks for all your advice so far!
>>>
>>> loadmodule "tm.so"
>>> loadmodule "tmx.so"
>>> loadmodule "usrloc.so"
>>> loadmodule "auth.so"
>>> loadmodule "auth_db.so"
>>> loadmodule "ctl.so"
>>> loadmodule "db_mysql.so"
>>> loadmodule "kex.so"
>>> loadmodule "maxfwd.so"
>>> loadmodule "mi_fifo.so"
>>> loadmodule "mi_rpc.so"
>>> loadmodule "nathelper.so"
>>> loadmodule "perl.so"
>>> loadmodule "pv.so"
>>> loadmodule "registrar.so"
>>> loadmodule "rr.so"
>>> loadmodule "sanity.so"
>>> loadmodule "siputils.so"
>>> loadmodule "sl.so"
>>> loadmodule "textops.so"
>>> loadmodule "xlog.so"
>>>
>>>
>>> On 24 July 2013 16:33, Daniel-Constantin Mierla
>>> <miconda at gmail.com <mailto:miconda at gmail.com>> wrote:
>>>
>>> Hello,
>>>
>>>
>>> On 7/24/13 4:24 AM, David Cunningham wrote:
>>>> Hello,
>>>>
>>>> Thank you very much for the email. In reply:
>>>>
>>>> 1. The system ran out of memory. Linux's oom-killer
>>>> killed Kamailio.
>>> then all the instructions I gave are useless, they
>>> are for debugging kamailio's internal memory
>>> manager, which handles pkg and shm mallocs.
>>>
>>> The chances to be from kamailio itself are very low
>>> now. Do you do lot of mi commands (e.g., kamctl
>>> ...)? The mi api uses system malloc, but the rest of
>>> code should use internal memory manager which does
>>> not go beyond the limits set with -m and -M, thus
>>> not causing an OS memory exhaustion.
>>>
>>> Can you list what modules are you loading? At some
>>> point it was a leak in libssl, in case you use tls a
>>> lot. But could be another external library...
>>>
>>> Cheers,
>>> Daniel
>>>
>>>
>>>>
>>>> 2. You're right, DEBUG_MEMORY is a local
>>>> configuration setting. If defined it sets memdbg to
>>>> -2, and memlog to -2. The debug setting is -1.
>>>>
>>>> 3. We'll try setting mem_summary=12, thanks.
>>>>
>>>> 4. We'll try setting asynchronous syslog, thanks.
>>>>
>>>> 5. Our configuration totals 338 lines, or approx
>>>> 8.5kb. Is that a lot?
>>>>
>>>> 6. We'll try setting mem_join=1, thanks.
>>>>
>>>>
>>>>
>>>> On 23 July 2013 16:53, Daniel-Constantin Mierla
>>>> <miconda at gmail.com <mailto:miconda at gmail.com>> wrote:
>>>>
>>>> Hello,
>>>>
>>>> first, to clarify, is the system memory or
>>>> kamailio's pkg/shm memory running out? If the
>>>> operating system runs out of memory, then
>>>> should be a leak in a library, because kamailio
>>>> modules uses only from a pre-allocated chunk,
>>>> not going over it.
>>>>
>>>>
>>>> On 7/23/13 7:33 AM, David Cunningham wrote:
>>>>
>>>> Hello,
>>>>
>>>> We're running a Kamailio 3.3.4 system, and
>>>> Kamailio is slowly using more and more
>>>> memory. Over a couple of weeks it will run
>>>> out of system memory.
>>>>
>>>> We tried to enable memory debugging doing
>>>> the following, but it resulted in Kamailio
>>>> not responding to any SIP packets. Would
>>>> anyone have advice please on how to debug
>>>> the situation?
>>>>
>>>> 1. In Makefile.defs set MEMDBG to 1 and
>>>> recompile Kamailio.
>>>> 2. In kamailio.cfg add the line:
>>>> #!define DEBUG_MEMORY 1
>>>>
>>>> do you set something special in config when
>>>> DEBUG_MEMORY is 1? It is not by default there,
>>>> so I assume you added some rules based on this
>>>> pre-processor directive.
>>>>
>>>> For memory troubleshooting, set memlog to a
>>>> value lower than debug parameter in config file
>>>> and try with mem_summary=12 for a more compact
>>>> output. See more about these parameters in the
>>>> wiki:
>>>>
>>>> -
>>>> http://www.kamailio.org/wiki/cookbooks/3.3.x/core#memlog
>>>>
>>>> Run kamailio for a while in normal conditions,
>>>> then restart it to get the memory usage
>>>> summaries. There should be indication if there
>>>> is some leak, by seeing memory chunks allocated
>>>> many times from a function used at runtime. You
>>>> can send the memory summary for a process here,
>>>> we can look at it.
>>>>
>>>>
>>>>
>>>> While this was running and Kamailio didn't
>>>> respond to packets, it logged lots of lines
>>>> like this:
>>>>
>>>>
>>>> Do you have syslog to be configured in
>>>> asynchronous mode? See the notes from:
>>>>
>>>> -
>>>> http://www.kamailio.org/wiki/tutorials/3.2.x/syslog
>>>>
>>>> The memdbg is less than debug value, that means
>>>> printing few log messages for each memory
>>>> operation. You can make memdbg higher and rely
>>>> on memlog for memory summaries, otherwise will
>>>> be lot of log messages related to memory.
>>>>
>>>>
>>>> Jul 22 21:32:22 hostname kamailio: : <core>
>>>> [mem/q_malloc.c:369]: qm_malloc(0x4000e008,
>>>> 128) called from <core>: cfg.lex: addstr(1438)
>>>> Jul 22 21:32:22 hostname kamailio: : <core>
>>>> [mem/q_malloc.c:413]: qm_malloc(0x4000e008,
>>>> 128) returns address 0x40048918 frag.
>>>> 0x40048900 (size=128) on 1 -th hit
>>>> Jul 22 21:32:22 hostname kamailio: : <core>
>>>> [mem/q_malloc.c:369]: qm_malloc(0x4000e008,
>>>> 128) called from <core>: cfg.lex: addstr(1438)
>>>> Jul 22 21:32:22 hostname kamailio: : <core>
>>>> [mem/q_malloc.c:413]: qm_malloc(0x4000e008,
>>>> 128) returns address 0x400489c8 frag.
>>>> 0x400489b0 (size=128) on 1 -th hit
>>>>
>>>> addstr() is a function used only for parsing
>>>> configuration file, as long as you can still
>>>> see them, the configuration file parsing was
>>>> not finish. addstr() is not a source of leaks
>>>> because it is not used at runtime.
>>>>
>>>> If you have large config file, then you can get
>>>> close to the limits of the private memory,
>>>> which is set to 4MB. You can increase its value
>>>> using -M parameter (e.g., start kamailio with
>>>> -M 8 to set it to use 8MB of memory).
>>>>
>>>> Over the time, the private memory can get used
>>>> due to fragmentation, you can set the mem_join
>>>> parameter in config file to avoid it (works
>>>> when compiled with MEMDBG=1).
>>>>
>>>> To monitor usage of internal pkg memory, then
>>>> you can use sercmd with pkg.stats command:
>>>>
>>>> http://kamailio.org/docs/modules/3.3.x/modules_k/kex.html#idp16972640
>>>>
>>>> Shared memory stats are printed by 'kamctl fifo
>>>> get_statistics shmem:'
>>>>
>>>> When you see significant increase of the memory
>>>> usage, then you can restart to get the summaries.
>>>>
>>>> You should run these commands after start, just
>>>> to see the initial usage of memory.
>>>>
>>>> Cheers,
>>>> Daniel
>>>>
>>>> --
>>>> Daniel-Constantin Mierla - http://www.asipto.com
>>>> http://twitter.com/#!/miconda
>>>> <http://twitter.com/#%21/miconda> -
>>>> http://www.linkedin.com/in/miconda
>>>>
>>>>
>>>> _______________________________________________
>>>> SIP Express Router (SER) and Kamailio (OpenSER)
>>>> - sr-users mailing list
>>>> sr-users at lists.sip-router.org
>>>> <mailto:sr-users at lists.sip-router.org>
>>>> http://lists.sip-router.org/cgi-bin/mailman/listinfo/sr-users
>>>>
>>>>
>>>>
>>>>
>>>> --
>>>> David Cunningham, Voisonics
>>>> http://voisonics.com/
>>>> USA: +1 213 221 1092 <tel:%2B1%20213%20221%201092>
>>>> UK: +44 (0) 20 3298 1642
>>>> <tel:%2B44%20%280%29%2020%203298%201642>
>>>> Australia: +61 (0) 2 8063 9019
>>>> <tel:%2B61%20%280%29%202%208063%209019>
>>>
>>> --
>>> Daniel-Constantin Mierla -http://www.asipto.com
>>> http://twitter.com/#!/miconda <http://twitter.com/#%21/miconda> -http://www.linkedin.com/in/miconda
>>>
>>>
>>>
>>>
>>> --
>>> David Cunningham, Voisonics
>>> http://voisonics.com/
>>> USA: +1 213 221 1092 <tel:%2B1%20213%20221%201092>
>>> UK: +44 (0) 20 3298 1642
>>> <tel:%2B44%20%280%29%2020%203298%201642>
>>> Australia: +61 (0) 2 8063 9019
>>> <tel:%2B61%20%280%29%202%208063%209019>
>>
>> --
>> Daniel-Constantin Mierla -http://www.asipto.com
>> http://twitter.com/#!/miconda <http://twitter.com/#%21/miconda> -http://www.linkedin.com/in/miconda
>>
>>
>>
>>
>> --
>> David Cunningham, Voisonics
>> http://voisonics.com/
>> USA: +1 213 221 1092 <tel:%2B1%20213%20221%201092>
>> UK: +44 (0) 20 3298 1642 <tel:%2B44%20%280%29%2020%203298%201642>
>> Australia: +61 (0) 2 8063 9019
>> <tel:%2B61%20%280%29%202%208063%209019>
>
> --
> Daniel-Constantin Mierla -http://www.asipto.com
> http://twitter.com/#!/miconda <http://twitter.com/#%21/miconda> -http://www.linkedin.com/in/miconda
>
>
>
>
> --
> David Cunningham, Voisonics
> http://voisonics.com/
> USA: +1 213 221 1092 <tel:%2B1%20213%20221%201092>
> UK: +44 (0) 20 3298 1642 <tel:%2B44%20%280%29%2020%203298%201642>
> Australia: +61 (0) 2 8063 9019 <tel:%2B61%20%280%29%202%208063%209019>
>
>
>
>
> --
> David Cunningham, Voisonics
> http://voisonics.com/
> USA: +1 213 221 1092 <tel:%2B1%20213%20221%201092>
> UK: +44 (0) 20 3298 1642 <tel:%2B44%20%280%29%2020%203298%201642>
> Australia: +61 (0) 2 8063 9019 <tel:%2B61%20%280%29%202%208063%209019>
--
Daniel-Constantin Mierla - http://www.asipto.com
http://twitter.com/#!/miconda - http://www.linkedin.com/in/miconda
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.sip-router.org/pipermail/sr-users/attachments/20130731/aa40b789/attachment-0001.html>
More information about the sr-users
mailing list