Rpi 4 Cluster - Part 4

This is a bit of a sidebar post.

As noted in Part 3, I had finished the first two parts of Garrett’s “Building a Raspberry Pi Cluster, and was starting on Part III: OpenMPI, Python, and Parallel Jobs. Installing OpenMPI went smoothly and Installing Python3 from source was fine until I realized I had skipped a step and installed as a normal user instead of root. So I figured I would just delete the /clusterfs/usr directory and reinstall as root. No problem!

Note to self: Do not remove directories when distracted… So in my haste, this is what I did from the /clusterfs directory.

sudo rm -Rf /usr/*

(notice that leading slash?) instead of what I should have done:

sudo rm -Rf usr/*

Well, you can imagine what a surprise I got when I went to list the /clusterfs directory and found nothing was working. I HAD JUST DELETED THE ENTIRE OPERATING SYSTEM…

So, that put me back to reloading the Raspbian OS from backup and starting over for the (former) master node. In other words, doing parts 1 and 2 over. That itself creates a few issues too. Like, the MUNGE keys no longer work for the other nodes and need to be redistributed to each one. That led to another issue. The NFS partition refused to connect when the nodes requested it. I found that was caused by the munge.key file not being owned by root. Using SFTP to copy files because the NFS partition was not available changes ownership. CHMOD fixed that issue. So, now we are back where we were two days ago!

Again, installing OpenMPI went fine, but this time I just installed Python3 the easy way to save bandwidth. The Python “Hello World” program went fine. Now I am at the point of entering the Python program for determining PI. That’s what I will do next. Very exciting to see how that goes! So, until next post…