Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
Hello,
I'll try to be clear, but it's really hard to explain..
In few words : when I create a Logical Volume on a host attached
(iSCSI) to one DRBD server, it doesn't appear on the second host,
attached to the other DRBD. But, on existing LV, it works like a charm
(all data's are replicated). And more, if I create a LV on "one side",
and fill it with data, I saw the replication link with heavy traffic,
but always no LV.
Now, to understand, here is how it happened with this infrastructure.
1) On both DRBD hosts : Debian GNU/Linux, Dual Primary setup, version:
8.3.7 (api:88/proto:86-91). I choose to configure a resource on a
Logical volume of 500Gb (named /dev/vg0/xen). To be clear, two hosts
are "DRBD1" & "DRBD2". a cat /proc/drbd gave this :
0: cs:Connected ro:Primary/Primary ds:UpToDate/UpToDate C r----
ns:153716 nr:408752 dw:562408 dr:355156 al:179 bm:109 lo:0 pe:0
ua:0 ap:0 ep:1 wo:n oos:0
2) This resource is exported with iSCSI. For the example, let's say
that CLIENT1 is connected to DRBD1, and CLIENT2 is connected to DRBD2.
It's simplifed because in real, it's multiple Xen Dom0's clients with
multipath (but we don't care of that here).
3) CLIENT1 see the device as a block device (so far so good, iSCSI
works). I choose to use LVM on this block device. I create for example
: /dev/vg_xen/mydisk on CLIENT1. I mount it, I put some stuff in there
: I saw the replication like working : some traffic. DRBD2 and 1 says
it's all OK. If I lvscan on CLIENT1, I can see my brand new volume.
4) CLIENT2, (so, connected on iSCSI with DRBD2), see the block device,
the volume group, but NOT the Logical Volume. If I disconnect DRBD2's
resource, and reconnect it, and reconnect iSCSI of CLIENT2, wow, I saw
the LV !
And more : if the LV exists on both side (after disconnect/reconnect
the resource), data's are correctly replicated (obviously, I do NOT
mount LV on both side, I'm aware of that !). But, if I fill on one
side (e.g CLIENT1), dismount it, then mount it on CLIENT2, data are
here, without any problem.
So, my "theory", is that LVM operations (lvcreate or lvremove) on a
volume group, which is on top of iSCSI and LVM replicated device by
DRBD, are NOT replicated, UNTIL disconnect/reconnect the ressource. I
don't know why, and that's why I ask here to understand what I miss.
Additional Informations: If my clients are connected to the SAME DRBD
(let's tell DRBD1), if CLIENT1 creates a LV, CLIENT2 is immediatly
aware of that (just inactive, not a problem, vgchange and it works).
So the "problem" is during the replication.
Thanks for your help.
Regards,
Olivier
XO Project
http://xen-orchestra.com