Seeding /dev/random on Compute Nodes

We are currently running into an issue with some jobs on our HPC cluster that involves code that reads from /dev/random on our compute nodes (usually somewhere deep in the python dependency chain for code that our users are running). As our compute nodes do not have mouse input and little keyboard input it appears that /dev/random isn’t seeded enough for this code to run and instead the program or script hangs as it waits for /dev/random to get seeded.

Is there a best method for seeding /dev/random on compute nodes? I’ve seen some suggestions of installing the haveged daemon but I was wondering if there were other suggestions or if haveged is the best bet.

We had a discussion about that back in Aug, 2016, and we found the following article useful.

The article mentions three types of entropy generators, of which haveged was the one we elected to use. See,


haveged is available from EPEL for Red Hat/CentOS 7/8.

I hope that helps.