[Linux-cluster] CCSD ipv6 bind failures

[Fabio M. Di Nitto]

On 12/18/2012 1:33 PM, Ashish G wrote:
> hi Fabio,
>             Thanks for your replies.
> 
>> "Unable to bind to backend ipv6 socket, but inet_ntop returned NULL
>> pointer: Address already in use"
> This issue is not seen always. We see this problem when our application
> is run for the first time after it is installed. cman start fails as a
> result of these logs since ccsd fails and it keeps retrying. Rebooting
> the node helped us get rid of the logs.
> Its possible that ipv6 module did not load when binding was attempted.
> But if you look at the log, it says binding failed for "backend" port
> (50007). Under normal running of system, netstat does not list this port
> number against ccsd. Instead frontend (50006) and cluster_base_port
> (50008) are listed. Any idea why backend port binding is attempted and
> that too only when our application is freshly installed and run?

No, I don´t know. It´s possible that something is using port 50007 for
other reasons and clearly the bind fails. Or maybe an interface is not up.

Hard to say without sosreports from a bad and a good boot.

If you are a RH customer, please file a ticket with GSS so that we can
access the data required to perform debugging.

Fabio

> 
> Regards,
> Ashish
> 
> On Tue, Dec 18, 2012 at 12:47 PM, Fabio M. Di Nitto <fdinitto at redhat.com
> <mailto:fdinitto at redhat.com>> wrote:
> 
>     On 12/5/2012 2:52 PM, Ashish G wrote:
>     > hi Experts,
>     >                I has few question on ccsd:
>     > 1. what is the purpose of ccsd listening on ipv4 and ipv6 addresses as
>     > follows in my 2 node HA setup? We do not use IPv6 in our setup.
>     >
>     > netstat -antp |grep ccsd
>     >
>     > tcp        0      0 0.0.0.0:50008 <http://0.0.0.0:50008>
>     <http://0.0.0.0:50008>
>     > 0.0.0.0:*                   LISTEN      18170/ccsd
>     >
>     > *tcp        0      0 ::1:50006
>     > :::*                        LISTEN      18170/ccsd*
>     >
>     >
>     > **
>     >
>     > *CCSD is invoked without any arguments:
>     > *
>     >
>     > *root     18170     1  0 14:12 ?        00:00:00 /sbin/ccsd*
>     >
> 
>     RFCs now mandates software to support both IPv6 and IPv4 and v6 has to
>     be preferred when available.
> 
>     >
>     > 2. Is it not sufficient to use IPv4 only by CCSD?
> 
>     For operational purposes yes, one protocol is enough, but we need to
>     support both.
> 
>     >
>     > 3. Sometimes on node reboot we see messages as follows:
>     >
>     > "Unable to bind to backend ipv6 socket, but inet_ntop returned NULL
>     > pointer: Address already in use"
>     >
>     > Does anyone know when is this problem seen?
> 
>     No, it´s probably harmless. Does it create any runtime problem? I guess
>     that the kernel ipv6 module is not loaded and bind fails.
> 
>     >
>     > 4. If I invoke CCSD with "-4" option, will I get rid of the above
>     > message? since ipv6 code path will not hit.
>     >
>     > *root       461     1  0 14:28 ?        00:00:00 /sbin/ccsd -4*
>     >
>     > netstat -antp  |grep ccsd
>     >
>     > *tcp        0      0 127.0.0.1:50006 <http://127.0.0.1:50006>
>     <http://127.0.0.1:50006>
>     >     0.0.0.0:*                   LISTEN      461/ccsd*
>     >
>     > tcp        0      0 0.0.0.0:50008 <http://0.0.0.0:50008>
>     <http://0.0.0.0:50008>
>     > 0.0.0.0:*                   LISTEN      461/ccsd
>     >
> 
>     Just try it, if you are not using v6, specifying -4 should be enough.
> 
>     Fabio
> 
>     --
>     Linux-cluster mailing list
>     Linux-cluster at redhat.com <mailto:Linux-cluster at redhat.com>
>     https://www.redhat.com/mailman/listinfo/linux-cluster
> 
> 
> 
>