From: <daos@daos.groups.io> on behalf of Jordan Henderson <jhenderson@...>
Reply-To: "daos@daos.groups.io" <daos@daos.groups.io>
Date: Tuesday 5 May 2020 at 02:24
To: "daos@daos.groups.io" <daos@daos.groups.io>
Subject: [daos] Cannot start DAOS - failure in hwloc

Hi all,

I recently updated my out-of-date DAOS installation to commit a7f093db2fc96aa1dc20cc8c293d44274474ef62 (I'm sure this wasn't the commit that changed this behavior, but good to have for reference), and it seems I can no longer start the DAOS I/O server due to a failure in hwloc_set_membind (https://github.com/daos-stack/daos/blob/master/src/iosrv/srv.c#L392).

I get the following lines in my log:

05/04-18:52:41.20 Talos DAOS[14193/14197] server ERR  src/iosrv/srv.c:395 dss_srv_handler() failed to set memory affinity: 38
05/04-18:52:41.20 Talos DAOS[14193/14193] server ERR  src/iosrv/init.c:521 server_init() DAOS cannot be initialized using the configured path (/mnt/daos_fs).   Please ensure it is on a PMDK compatible file system and writeable by the current user.

which I think corresponds to a return value of ENOSYS from hwloc_set_membind, probably due to some form of missing support for the type of memory binding requested. Has the DAOS team seen this issue on other machines? Removing the call from the DAOS source allows me to start the server once again, but of course that's just a temporary workaround for now.

Thanks in advance!