Description

This PR implements a fix for Slurm 25.05 to prevent a fatal startup failure when reading topology.yaml due to a page-alignment issue.

Problem

In Slurm 25.05, the YAML parser plugin uses memory-mapped I/O to read configuration files like topology.yaml. The parser strictly requires the input buffer to be NULL-terminated. If the file size is exactly a multiple of the page size (4096 bytes), the parser reads past the mapped buffer to check for the NULL terminator, potentially accessing random data in the adjacent page and failing with EINVAL (Invalid argument).

This causes a fatal error: fatal: Something wrong with reading /usr/local/etc/slurm/topology.yaml: Invalid argument.

Solution

Modified gen_topology_yaml in conf.py to check the size of the generated cloud_topology.yaml file. If the size is a multiple of 4096 bytes, the script appends a newline (\n) to the file. This ensures the file is never page-aligned, satisfying the Slurm parser without breaking the YAML validity.

Verification

1. Reproduction: Verified on a live cluster running Slurm 25.05.3 by manually padding topology.yaml to exactly 4096 bytes. The Slurm controller failed to start with the expected Invalid argument error.
2. Fix Verification: Modified conf.py to force the generated file size to 4096 bytes. Verified that the fix correctly appended a newline, resulting in a 4097-byte file, and slurmctld started successfully.