Update princess to 0.4 (use Unix sockets)
Description
On rare occasions, the kernel fails to start with an 'Address already in use' error. By default, kernels bind to randomly chosen ports on the TCP loopback interface, and I believe this occurs when something else binds the port between the parent process choosing it and the kernel binding it.
Within a single machine, Unix sockets are a convenient way to avoid this. We can create a temporary directory to hold them, and there should be no risk of accidental clashes. Princess 0.4 does this by default on Posix systems. Sockets are created under /tmp, which is a local filesystem on Maxwell, so they should never be visible to other nodes, and they're in a new temporary directory for each job.
https://github.com/European-XFEL/princess/pull/2
How Has This Been Tested?
Ran a correction job from the CLI using princess from the PR above.
Types of changes
- Bug fix (non-breaking change which fixes an issue)