Closed garlick closed 3 years ago
This looked familiar (I think Garrett ran into it a while back). The previous issue (https://github.com/flux-framework/flux-core/issues/2988) and PR (https://github.com/flux-framework/flux-core/pull/2998). In that PR, @grondo added
sanitize_hwloc_xml() {
sed 's/pci_link_speed=".*"//g' $1
}
to t2005 to avoid the issue. Seems like it could work here too.
This test failure seems to result from the PCI link speed changing, presumably due to power management, when the topology is sampled twice on the same node.
Not sure if maybe there are some hwloc flags we can use to exclude this level of detail from the topology?