Erik,

You shouldn't need the AuthInfo in the slurmdbd.

I would make sure munge is installed this way on each of the nodes in 
your cluster, and Slurm for that matter.

 From your output of the slurmd I would say it isn't configured 
correctly since it is pointing to the default location.  You shouldn't 
have to alter AuthInfo.

I just tested running munge from a different install dir like you have 
and other than pointing the Slurm configure to the different install dir 
there was no other change I needed to do and all daemons worked with no 
issue.  With the munge turned off I get this error...

slurmctld: error: Munge encode failed: Unable to access 
"/home/da/downloads/munge/var/run/munge/munge.socket.2": No such file or 
directory

Using this message you can tell where the daemon thinks the file should 
be.  If you had already made the code and then ran configure again I 
would do a make clean and then remake things as they might not of been 
remade to point at the right location.

Danny

On 04/23/2013 10:33 PM, Erik Soyez wrote:
> Good morning,
>
> I have compiled Munge and Slurm to use different root directories than
> the defaults:
>
>       /usr/local/munge/0.5.10
>       /usr/local/slurm/2.5.4
>
> Munge then writes its socket file to:
>
>       /usr/local/munge/0.5.10/var/run/munge/munge.socket.2
>
> It works fine on the control node where I have an entry in the
> slurmdbd.conf file:
>
>       AuthInfo=/usr/local/munge/0.5.10/var/run/munge/munge.socket.2
>
> On the compute nodes the slurmd can't find the munge socket file,
> because it looks at the wrong path:
>
>       slurmd: error: Munge encode failed: Unable to access 
> "/usr/local/var/run/munge/munge.socket.2": No such file or directory
>
> Is there any other way to tell Slurm where to find the socket file other
> than "AuthInfo" in the slurmdbd.conf file?  Any environment variable?
>
> I have used for Munge....
>       ./configure --prefix=/usr/local/munge/0.5.10 
> --datarootdir=/usr/local/munge/0.5.10/share
>
> .... and for Slurm:
>       ./configure --prefix=/usr/local/slurm/2.5.4 
> --datarootdir=/usr/local/slurm/2.5.4/share --enable-multiple-slurmd 
> --with-munge=/usr/local/munge/0.5.10
>
> Everything wokrs fine if I create a symbolic link:
>       /usr/local/var/run/munge/munge.socket.2 
> /usr/local/munge/0.5.10/var/run/munge/munge.socket.2
>
> But that's obviously not what I want to do....
>
> Many thanks!!  Erik.
>
>
>

Reply via email to