Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
> > >As in your expected experiment, you will need to cut the communication path > > >between the two sites. > > >The easiest way to do so is to run "drbdadm disconnect <resource>" on one > > >of the sites; given your requirements, on cluster B. > > > > This changes the state of the volume on cluster B to "OutDated" and I am > > unable > > to change the volume to primary. I seem to still be missing something. > No, that should be the correct way. > > (I'm not sure about your exact sequence of events - depending on what > happens on cluster A, the "OutDated" might be okay (because the "original" > Primary sees that there are outstanding writes), or it might be a DRBD > 9 bug. > (Would you please tell me what /proc/drbd says? I'll need the exact git > hash.) To expand on that a little bit - on a "graceful" disconnect DRBD will mark itself "OutDated" (on the Secondary). You can either * make DRBD Secondary on both sides (=> a "real" switchover) * or run a forced disconnect - that'll look more like a "connection disrupted", and should leave DRBD as "UpToDate", and so a simple open() should be sufficient to make it Primary again. Hope that helps!