On Jan 19, 2015, at 14:59 , Martin Polednik <mpolednik@redhat.com> wrote:

Hello,

Since dealing with SR-IOV, we've discovered the need to persist number of virtual functions for given virtual functions. The need is actually split into 2 parts: setting the number and persisting it. Our talks have pointed to 5 choices in persistence mechanisms:

* let engine set the number when host becomes operational * implement mechanism in systemd/udev * create udev rule set * persist them inside VDSM * or actually let the system administrator handle this

All of these choices except for the last still need the engine to do the initial set-up. I would like to hear your opinion on persistence mechanism and splitting the choice of mechanism and means of set-up. Since persistence is currently mostly required by networking team, I propose that [1] would be accepted as the means of set-up and in future, network team would figure out and implement their preferred way of persistence.

I would agree. Since there is no persistent mechanism available at the moment I would go with engine setting that explicitly based on host lifecycle. trying to persist in vdsm is not helping as it anyway should be a different component (systemd?), and pushing such a mechanism to systemd is not feasible in a reasonable timeframe Thanks, michal

[1] http://gerrit.ovirt.org/#/c/36216/3

Regards,

mpolednik