problems to run librpmem

[Beilei Sun]

Hi,

I got the following problem while running the basic example of librpmem.

rpmemd: error while loading shared libraries: libpmem.so.1: cannot open shared object file: No such file or directory

According to the rpmem.log, this error is caused by executing following command: 'ssh -v -T -oBatchMode=yes 10.10.10.1 rpmemd'.

Also, I tried to execute 'ssh -v -T -oBatchMode=yes 10.10.10.1 pmempool' command. It reports that libpmempool.so.1 cannnot be found.

But both libpmem.so.1 and libpmempool.so.1 can be used if I run the other NVML examples except librpmem.

Any tips about this error?

Thanks

[pawel.szymanski]

Hi,

It look like the account you are using to run the SSH command on remote node does not have properly configured path to NVML libraries.

You can do two things:

1) Install NVML libraries on remote node in a directory that is already included in the set of directories used by loader. You can do this by running 'sudo make install' command on the remote node in the nvml/src directory.

2) You can modify the command that launches RPMEMD on the remote node such as the command includes LD_LIBRARY_PATH with the directory where the NVML libraries are located on the remote node.

In order to do this, you need to define on local node RPMEM_CMD as follows:

export RPMEM_CMD="LD_LIBRARY_PATH+=my_nvml_directory rpmemd"

Regards

Pawel

[Beilei Sun]

Thanks Pawel, the problem is solved.

I have another problem when I run the follow command:

./basic create 10.10.10.10 rpmem_test.dat

I think ./basic will create the rpmem_test.dat in the directory (/mnt/pmem), which is speified in the .rpmemd.conf file.  But, it reports 'rpmem_create: pool set or its par doesn't exist or it is unavailable' error.

However, if I create this file manually, which I don't think is the right way, it reports 'rpmem_create: pool already exists' error.

Are there other parameters or enviroments needed to be set to run ./basic?

Thanks

在 2017年10月16日星期一 UTC+8下午3:42:43，pawel.szymanski写道：

[Janek Mi]

I suggest to add logging on both sides. It makes all issues easier to debug.

• On local node

RPMEM_LOG_FILE=rpmem.log RPMEM_LOG_LEVEL=10 (...) ./basic (...)

• On remote node

export RPMEM_CMD="(...) rpmemd --log-file=rpmemd.log --log-level=debug"

[Janek Mi]

I guess the issue is: rpmem_test.dat has to be the name of a pool set file located on a remote node.

So it /mnt/pmem/rpmem_test.dat should look like this:

PMEMPOOLSET
20M /mnt/pmem/rpmem_test.dat.part

On Saturday, 14 October 2017 14:14:03 UTC+2, Beilei Sun wrote:

[Beilei Sun]

Thanks Jane.

Your are right. I have solved this problem. Thansk again.

在 2017年10月17日星期二 UTC+8下午9:10:47，Janek Mi写道：