Open MPI logo

Open MPI User's Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Open MPI User's mailing list

Subject: Re: [OMPI users] getting opal_init:startup:internal-failure
From: Jeff Squyres (jsquyres) (jsquyres_at_[hidden])
Date: 2013-04-29 10:15:14


FWIW, to avoid using the --prefix option, you can set your PATH / LD_LIBRARY_PATH to point to the Open MPI installation on all nodes.

Many organizations opt to have NFS-shared home directories, so that when you modify your "main" shell startup file (e.g., .bashrc) to point PATH and LD_LIBRARY_PATH to your Open MPI installation, it effectively modifies it for all nodes in the cluster.

On Apr 29, 2013, at 8:56 AM, E.O. <ooyama.eiichi_at_[hidden]> wrote:

> It works!!!
> By putting two dash'es and no equal sign, it worked fine!!
>
> [root_at_host1 tmp]# mpirun --prefix /myname --host host2 /tmp/hello.out
> Hello World from processor host2, rank 0 out of 1 processors
> [root_at_host1 tmp]#
>
> It looks like one dash "-prefix" also works if I don't put an equal sign..
>
> Thank you very much!!
>
> Eiichi
>
>
>
> On Mon, Apr 29, 2013 at 8:29 AM, Ralph Castain <rhc_at_[hidden]> wrote:
> Hmm....okay. No, let's not bother to install a bunch of stuff you don't otherwise need.
>
> I probably mis-typed the "prefix" option - it has two dashes in front of it and no equal sign:
>
> mpirun --prefix ./myname ...
>
> I suspect you only put one dash, and the equal sign was a definite problem, which is why it gave you an error.
>
>
> On Apr 29, 2013, at 5:12 AM, E.O. <ooyama.eiichi_at_[hidden]> wrote:
>
>> I tried configuring/building an OMPI on the remote host but I was not able to...
>> The remote host (host2) doesn't have any development tools, such as gcc, make, etc...
>>
>> Since I am able to run an MPI hello_c binary on the remote host, I believe the host has all the necessary libraries needed for MPI. I am also able to run an MPI hello_c binary on host1 from host2.
>>
>> [root_at_host2 tmp]# mpirun -host localhost /tmp/hello.out
>> Hello World from processor host2, rank 0 out of 1 processors
>> [root_at_host2 tmp]# mpirun -host host2 /tmp/hello.out
>> Hello World from processor host2, rank 0 out of 1 processors
>> [root_at_host2 tmp]# mpirun -host host1 /tmp/hello.out
>> Hello World from processor host1, rank 0 out of 1 processors
>> [root_at_host2 tmp]#
>>
>> However I still can't run hello_c binary on host2 from host1
>>
>> [root_at_host1 tmp]# mpirun -host host2 /tmp/hello.out
>> --------------------------------------------------------------------------
>> Sorry! You were supposed to get help about:
>> opal_init:startup:internal-failure
>> But I couldn't open the help file:
>> //share/openmpi/help-opal-runtime.txt: No such file or directory. Sorry!
>> --------------------------------------------------------------------------
>> [host2:02499] [[INVALID],INVALID] ORTE_ERROR_LOG: Error in file runtime/orte_init.c at line 79
>> [host2:02499] [[INVALID],INVALID] ORTE_ERROR_LOG: Error in file orted/orted_main.c at line 358
>> --------------------------------------------------------------------------
>> A daemon (pid 17710) died unexpectedly with status 255 while attempting
>> to launch so we are aborting.
>>
>> There may be more information reported by the environment (see above).
>>
>> This may be because the daemon was unable to find all the needed shared
>> libraries on the remote node. You may set your LD_LIBRARY_PATH to have the
>> location of the shared libraries on the remote nodes and this will
>> automatically be forwarded to the remote nodes.
>> --------------------------------------------------------------------------
>> --------------------------------------------------------------------------
>> mpirun noticed that the job aborted, but has no info as to the process
>> that caused that situation.
>> --------------------------------------------------------------------------
>> [root_at_host1 tmp]#
>>
>>
>> If I set -prefix=/myname, it returns a different output
>>
>> [root_at_host1 tmp]# mpirun -prefix=/myname -host host2 /tmp/hello.out
>> --------------------------------------------------------------------------
>> mpirun was unable to launch the specified application as it could not access
>> or execute an executable:
>>
>> Executable: -prefix=/myname
>> Node: host1
>>
>> while attempting to start process rank 0.
>> --------------------------------------------------------------------------
>> [root_at_host1 tmp]#
>>
>> Do you still want me to try building OMPI on the remote host?
>>
>> eiichi
>>
>>
>>
>> On Sun, Apr 28, 2013 at 12:24 PM, Ralph Castain <rhc_at_[hidden]> wrote:
>> If you configure/build OMPI on the remote node using the same configure options you used on host1, does the problem go away?
>>
>>
>> On Apr 28, 2013, at 8:58 AM, E.O. <ooyama.eiichi_at_[hidden]> wrote:
>>
>>> Thank you Ralph!
>>> I ran it with "-prefix" option but I got this...
>>>
>>> [root_at_host1 tmp]# mpirun -prefix /myname -np 4 -host host2 ./hello.out
>>> --------------------------------------------------------------------------
>>> mpirun was unable to launch the specified application as it could not access
>>> or execute an executable:
>>>
>>> Executable: -prefix=/myname
>>> Node: host1
>>>
>>> while attempting to start process rank 0.
>>> --------------------------------------------------------------------------
>>> [root_at_host1 tmp]#
>>>
>>> I also updated PATH in the remote host (host2) to include /myname.
>>> But it didn't seem change anything...
>>>
>>> eiichi
>>>
>>>
>>>
>>>
>>> On Sun, Apr 28, 2013 at 11:48 AM, Ralph Castain <rhc_at_[hidden]> wrote:
>>> The problem is likely that your path variables aren't being set properly on the remote machine when mpirun launches the remote daemon. You might check to see that your default shell rc file is also setting those values correctly. Alternatively, modify your mpirun cmd line a bit by adding
>>>
>>> mpirun -prefix /myname ...
>>>
>>> so it will set the remove prefix and see if that helps. If it does, you can add --enable-orterun-prefix-by-default to your configure line so mpirun always adds it.
>>>
>>>
>>> On Apr 28, 2013, at 7:56 AM, "E.O." <ooyama.eiichi_at_[hidden]> wrote:
>>>
>>> > Hello
>>> >
>>> > I have five linux machines (one is redhat and the other are busybox)
>>> > I downloaded openmpi-1.6.4.tar.gz into my main redhat machine and configure'ed/compiled it successfully.
>>> > ./configure --prefix=/myname
>>> > I installed it to /myname directory successfully. I am able to run a simple hallo.c on my redhat machine.
>>> >
>>> > [root_at_host1 /tmp] # mpirun -np 4 ./hello.out
>>> > I am parent
>>> > I am a child
>>> > I am a child
>>> > I am a child
>>> > [root_at_host1 /tmp] #
>>> >
>>> > Then, I sent entire /myname directory to the another machine (host2).
>>> > [root_at_host1 /] # tar zcf - myname | ssh host2 "(cd /; tar zxf -)"
>>> >
>>> > and ran mpirun for the host (host2).
>>> >
>>> > [root_at_host1 tmp]# mpirun -np 4 -host host2 ./hello.out
>>> > --------------------------------------------------------------------------
>>> > Sorry! You were supposed to get help about:
>>> > opal_init:startup:internal-failure
>>> > But I couldn't open the help file:
>>> > //share/openmpi/help-opal-runtime.txt: No such file or directory. Sorry!
>>> > --------------------------------------------------------------------------
>>> > [host2:26294] [[INVALID],INVALID] ORTE_ERROR_LOG: Error in file runtime/orte_init.c at line 79
>>> > [host2:26294] [[INVALID],INVALID] ORTE_ERROR_LOG: Error in file orted/orted_main.c at line 358
>>> > --------------------------------------------------------------------------
>>> > A daemon (pid 23691) died unexpectedly with status 255 while attempting
>>> > to launch so we are aborting.
>>> >
>>> > There may be more information reported by the environment (see above).
>>> >
>>> > This may be because the daemon was unable to find all the needed shared
>>> > libraries on the remote node. You may set your LD_LIBRARY_PATH to have the
>>> > location of the shared libraries on the remote nodes and this will
>>> > automatically be forwarded to the remote nodes.
>>> > --------------------------------------------------------------------------
>>> > --------------------------------------------------------------------------
>>> > mpirun noticed that the job aborted, but has no info as to the process
>>> > that caused that situation.
>>> > --------------------------------------------------------------------------
>>> > [root_at_host1 tmp]#
>>> >
>>> > I set those environment variables
>>> >
>>> > [root_at_host1 tmp]# echo $LD_LIBRARY_PATH
>>> > /myname/lib/
>>> > [root_at_host1 tmp]# echo $OPAL_PREFIX
>>> > /myname/
>>> > [root_at_host1 tmp]#
>>> >
>>> > [root_at_host2 /] # ls -la /myname/lib/libmpi.so.1
>>> > lrwxrwxrwx 1 root root 15 Apr 28 10:21 /myname/lib/libmpi.so.1 -> libmpi.so.1.0.7
>>> > [root_at_host2 /] #
>>> >
>>> > If I ran the ./hello.out binary inside host2, it works fine
>>> >
>>> > [root_at_host1 tmp]# ssh host2
>>> > [root_at_host2 /] # /tmp/hello.out
>>> > I am parent
>>> > [root_at_host2 /] #
>>> >
>>> > Can someone help me figure out why I cannot run hello.out in host2 from host1 ?
>>> > Am I missing any env variables ?
>>> >
>>> > Thank you,
>>> >
>>> > Eiichi
>>> >
>>> >
>>> > _______________________________________________
>>> > users mailing list
>>> > users_at_[hidden]
>>> > http://www.open-mpi.org/mailman/listinfo.cgi/users
>>>
>>>
>>> _______________________________________________
>>> users mailing list
>>> users_at_[hidden]
>>> http://www.open-mpi.org/mailman/listinfo.cgi/users
>>>
>>> _______________________________________________
>>> users mailing list
>>> users_at_[hidden]
>>> http://www.open-mpi.org/mailman/listinfo.cgi/users
>>
>>
>> _______________________________________________
>> users mailing list
>> users_at_[hidden]
>> http://www.open-mpi.org/mailman/listinfo.cgi/users
>>
>> _______________________________________________
>> users mailing list
>> users_at_[hidden]
>> http://www.open-mpi.org/mailman/listinfo.cgi/users
>
>
> _______________________________________________
> users mailing list
> users_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/users
>
> _______________________________________________
> users mailing list
> users_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/users

-- 
Jeff Squyres
jsquyres_at_[hidden]
For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/