Activity log for bug #2019433

Date Who What changed Old value New value Message
2023-05-12 16:17:25 Macks bug added bug
2023-05-14 11:53:01 Macks summary Sysinv agent fails if /sys/devices/system/node/node0/hugepages is missing Support multiple numa nodes per core
2023-05-14 12:03:05 Macks description I am on stx7. While adding a dell r7415 as worker node, sysinv agent fails to get inventory. I gather from sysinv.log on the worker node that there is an uncaught OSError raised by sysinv/agent/node.py. This is due to line 427 only concerned with IOError. On my system, there is no /sys/devices/system/node/node0/hugepages dir (it's at node1). The missing dir raises OSError, not IOError. Workaround: added `except OSError...` to node.py after IOError, recompile and reboot. I am on stx7. While adding a dell r7415 as worker node, sysinv agent fails to get inventory. This is due to sysinv unable to work with multiple numa nodes per socket. It errors out on sysinv/agent/node.py#427 with uncaught OSError and cannot get host memory because it expects dir /sys/devices/system/node/node0/hugepages, which may not exist (mine is on node1). With chiplet design being more common these days, it makes sense to not assume 1 numa node per socket. Workaround: modify bios memory interleaving so it presents a single numa node per socket, or if bios doesn't support that like mine, add `numa=off` to grub boot params. Both results in decreased performance.