[Icecast-dev] Segmentation fault issue with new install

Tue Apr 28 23:57:28 PDT 2015

On 04/28/2015 02:27 PM, Jan-Pieter wrote:
> Hi all,
> see my reply inline.
> Cheers, Jan-Pieter
>
>
> 2015-04-26 7:23 GMT+02:00 "Thomas B. Rücker" <thomas at ruecker.fi
> <mailto:thomas at ruecker.fi>>:
>
>     On 04/25/2015 07:32 PM, Jan-Pieter wrote:
>     > Hi all,
>     >
>     > I am a long-time user of Icecast, always without problems. However i
>     > made a new setup using icecast 2.4.1 on a fresh and quick CentOS
>     VPS.
>
>     EPEL package or from somewhere else?
>
> it's just a standard CentOS 6.6 release, no EPEL. Distri from hosting
> provider. [root at h2429770 tmp]# uname -a
> Linux h2429770.stratoserver.net <http://h2429770.stratoserver.net>
> 2.6.32-042stab103.6 #1 SMP Wed Jan 21 13:07:39 MSK 2015 x86_64 x86_64
> x86_64 GNU/Linux
>  
>
>
>     > The issue is that starting icecast gives a segmentation fault at
>     > around 80% of all starts. No core file nor a entry in error.log
>     is posted.
>
>     Enable core dumps using "ulimit -c unlimited". The PWD from where you
>     start Icecast needs to be writeable.
>
>
> Even with ulimit -c unlimited and running from a writable PWD no core
> is dumped.

Seems it's a bit more involved on Centos6:
http://www.unixmen.com/how-to-enable-core-dumps-in-rhel6/
I've tried it successfully on a quick install of centos6 that I've
compiled 2.4.2 on and am going to try with your config.
(As you don't use url-auth for sources, 2.4.1 is safe for you)

>     > In the remaining 20% of the starts icecast start with no problem.
>     >
>     > Contents of the error.log with a succesfull start (20% of
>     icecast starts):
>     > [2015-04-25  21:08:40] INFO main/main Icecast 2.4.1 server started
>     > [2015-04-25  21:08:40] INFO connection/get_ssl_certificate No SSL
>     > capability
>     > [2015-04-25  21:08:40] INFO stats/_stats_thread stats thread started
>     > [2015-04-25  21:08:40] INFO yp/yp_update_thread YP update thread
>     started
>     > [2015-04-25  21:08:40] INFO source/source_fallback_file mountpoint
>     > /geen-uitzending.mp3 is reserved
>     > [2015-04-25  21:08:40] WARN format/format_get_type Unsupported or
>     > legacy stream type: "audio/mpeg". Falling back to generic minimal
>     > handler for best effort.
>     > [2015-04-25  21:08:40] WARN source/source_fallback_file mountpoint
>     > "/geen-uitzending.mp3" already reserved
>     > [2015-04-25  21:08:40] INFO source/source_main listener count on
>     > /geen-uitzending.mp3 now 0
>     >
>     >
>     > Contents of error.log with a segmentation fault direct after start
>     > (80% of icecast starts):
>     > 2015-04-25  21:09:05] INFO main/main Icecast 2.4.1 server started
>     > [2015-04-25  21:09:05] INFO connection/get_ssl_certificate No SSL
>     > capability
>     > [2015-04-25  21:09:05] INFO yp/yp_update_thread YP update thread
>     started
>     > [2015-04-25  21:09:05] INFO source/source_fallback_file mountpoint
>     > /geen-uitzending.mp3 is reserved
>     > [2015-04-25  21:09:05] WARN format/format_get_type Unsupported or
>     > legacy stream type: "audio/mpeg". Falling back to generic minimal
>     > handler for best effort.
>     > [2015-04-25  21:09:05] INFO stats/_stats_thread stats thread started
>
>     Can you attach your icecast.xml please (with passwords removed)?
>     This might help reproduce the problem.
>
>
> I attached icecast.xml. I also turned debug errorlog level on.
> Results with debug on:
>
> good start:
> [2015-04-28  15:56:39] INFO main/main Icecast 2.4.1 server started
> [2015-04-28  15:56:39] DBUG yp/yp_recheck_config Updating YP configuration
> [2015-04-28  15:56:39] INFO connection/get_ssl_certificate No SSL
> capability
> [2015-04-28  15:56:39] INFO stats/_stats_thread stats thread started
> [2015-04-28  15:56:39] INFO yp/yp_update_thread YP update thread started
> [2015-04-28  15:56:39] INFO source/source_fallback_file mountpoint
> /geen-uitzending.mp3 is reserved
> [2015-04-28  15:56:39] DBUG connection/connection_complete_source
> sources count is 0
> [2015-04-28  15:56:39] WARN format/format_get_type Unsupported or
> legacy stream type: "audio/mpeg". Falling back to generic minimal
> handler for best effort.
> [2015-04-28  15:56:39] DBUG connection/connection_complete_source
> source is ready to start
> [2015-04-28  15:56:39] DBUG source/source_init Source creation complete
> [2015-04-28  15:56:39] WARN source/source_fallback_file mountpoint
> "/geen-uitzending.mp3" already reserved
> [2015-04-28  15:56:39] INFO source/source_main listener count on
> /geen-uitzending.mp3 now 0
> [2015-04-28  15:56:39] DBUG stats/modify_node_event update global
> sources (1)
> [2015-04-28  15:56:39] DBUG stats/process_source_event new source stat
> /geen-uitzending.mp3
> [2015-04-28  15:56:39] DBUG stats/modify_node_event update global
> source_client_connections (1)
> [2015-04-28  15:56:39] DBUG stats/process_source_event new node
> listeners (0)
> [2015-04-28  15:56:39] DBUG stats/process_source_event new node
> listenurl (http://stream3.server7.nl:80/geen-uitzending.mp3)
> [2015-04-28  15:56:39] DBUG stats/modify_node_event update global
> source_total_connections (1)
> [2015-04-28  15:56:39] DBUG stats/process_source_event new node
> slow_listeners (0)
> [2015-04-28  15:56:39] DBUG stats/modify_node_event update
> "/geen-uitzending.mp3" listeners (0)
> [2015-04-28  15:56:39] DBUG stats/process_source_event new node
> listener_peak (0)
> [2015-04-28  15:56:39] DBUG stats/process_source_event new node
> stream_start (Tue, 28 Apr 2015 15:56:39 +0200)
> [2015-04-28  15:56:39] DBUG stats/process_source_event new node
> stream_start_iso8601 (2015-04-28T15:56:39+0200)
> [2015-04-28  15:56:39] DBUG stats/process_source_event new node
> total_bytes_read (0)
> [2015-04-28  15:56:39] DBUG stats/process_source_event new node
> total_bytes_sent (0)
> [2015-04-28  15:56:39] DBUG stats/modify_node_event update
> "/geen-uitzending.mp3" listeners (0)
> [2015-04-28  15:56:40] DBUG slave/_slave_thread checking master stream
> list

I currently suspect it's something interacting with the stats thread and
the two identical fallbacks.

>
> bad start (segfault):
> [2015-04-28  15:56:53] INFO main/main Icecast 2.4.1 server started
> [2015-04-28  15:56:53] DBUG yp/yp_recheck_config Updating YP configuration
> [2015-04-28  15:56:53] INFO connection/get_ssl_certificate No SSL
> capability
> [2015-04-28  15:56:53] INFO yp/yp_update_thread YP update thread started
> [2015-04-28  15:56:53] INFO source/source_fallback_file mountpoint
> /geen-uitzending.mp3 is reserved
> [2015-04-28  15:56:53] DBUG connection/connection_complete_source
> sources count is 0
> [2015-04-28  15:56:53] WARN format/format_get_type Unsupported or
> legacy stream type: "audio/mpeg". Falling back to generic minimal
> handler for best effort.
> [2015-04-28  15:56:53] DBUG connection/connection_complete_source
> source is ready to start
> [2015-04-28  15:56:53] DBUG source/source_init Source creation complete
> [2015-04-28  15:56:53] INFO stats/_stats_thread stats thread started
> [2015-04-28  15:56:53] DBUG stats/process_source_event new source stat
> /geen-uitzending.mp3
>
>
>
>  
>
>
>     > Both the stream and the fallback file are .mp3 type. I tried
>     removing
>     > the fallback entry. Then icecast starts always normally. So it seems
>     > related to the .mp3 fallback.
>     >
>     > Also I tried to launch icecast with strace, and this NEVER fails. No
>     > segmentation faults!
>     >
>     >
>     > Any idea what's going wrong? Something going wrong with timing on a
>     > fast server?
>
>     It sounds like you've found a race condition. I'm not exactly sure
>     what's happening though.
>     By the log output something seems to be instantly accessing the
>     stream.
>     Notice the order of the stats_thread is different in both.
>
>
> Yes this is the only interesting thing. The segfault only happens when 
> "INFO stats/_stats_thread stats thread started" 
> is missing in the error.log. So with every succesful start this line
> gets logged, and if the line is missing the segfault occurs.

I suspect it's a very close race that might only occur on that specific
configuration/virtualization.
In such case invasive things like strace and valgrind won't help us. A
coredump might reveal something, but isn't guaranteed.

What BTW confuses me is that you override the mountpoint configuration
to ogg/vorbis, but fall back to an mp3 file.

> I also tried running using valgrind on advice of Philipp, but when
> running with valgrind the segfault does not occur and icecast starts
> normally... (same as with gdb)

yeah, expected.

I tried your config on a centos 6 machine and couldn't reproduce it. How
many CPU cores / threads does it have?

Cheers

Thomas