flux-framework
resource: hwloc reader not happy on Macs with SSDs
Running within a fluxrm/flux-sched:latest Docker container on my Lab mac:
ƒ(s=1,d=0) fluxuser@8c3c409ae758:~$ flux module load sched-fluxion-resource
2020-09-25T04:38:04.024173Z sched-fluxion-resource.err[0]: grow: reader: Error getting the size of block storage device; Error getting the size of block storage device; Error getting the size of block storage device; Error hwloc walk: rank 0;
2020-09-25T04:38:04.024251Z sched-fluxion-resource.err[0]: update_resource_db: grow_resource_db: No such file or directory
2020-09-25T04:38:04.024259Z sched-fluxion-resource.err[0]: update_resource: update_resource_db: No such file or directory
2020-09-25T04:38:04.024265Z sched-fluxion-resource.err[0]: populate_resource_db_kvs: update_resource: No such file or directory
2020-09-25T04:38:04.024269Z sched-fluxion-resource.err[0]: populate_resource_db: loading resources from the KVS
Setting an allowlist worksaround this error and is probably why it work in the typical bootstrap via rc scripts:
flux module load sched-fluxion-resource load-allowlist=node,core
It was brought up in the node local storage PR that we need a way to signal from the hwloc reader that a non-fatal error occurred (the current recursive algorithm and use of the return code for error reporting makes that hard).
This was recently hit by @grondo as well. I believe an error in hwloc reader in handling a malformed (?) storage object within hwloc in dealing with certain storage types.
https://github.com/flux-framework/flux-sched/pull/909.
But it would be best to fix at the hwloc reader level. Keeping it open.