Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On Thursday 21 August 2008 18:37, Lars Ellenberg wrote: > On Thu, Aug 21, 2008 at 05:59:47PM +0300, Alex wrote: > > Hello drbd experts, > > > > I have drbd-8.2.6 working fine in dual-primary mode, mirroring 2 nodes. > > > > According to drbd howto > > (http://www.drbd.org/users-guide-emb/s-gfs-create.html) i saw that since > > DRBD does not support more than two nodes i can't format /dev/drbd0 using > > more then 2 journals (2 nodes): mkfs -t gfs -p lock_dlm -j 2 /dev/drbd0 > > and MOUNT IT on more then 2 nodes. > > > > Now, supposing that: > > - i have 4 machines grouped two by two in order to get two different drbd > > devices /dev/drbd0 and /dev/drbd1 > > - i am exporting /dev/drbd0 and /dev/drbd1 using iSCSI to the 5th macine > > - on the 5th machine, i am using lvm to join /dev/drbd{0,1} in one volume > > pvcreate /dev/drbd0 /dev/drbd1 > > vgcreate myvg /dev/drbd0 /dev/drbd1 > > lvcreate ... mylv myvg -> will result logical volume /dev/myvg/mylv > > > > Let say that /dev/myvg/mylv is exported now to another 3 nodes (again, > > via iscsi for example). > > > > Question: > > > > Can i use now mkfs.gfs -p lock_dlm -j 3 /dev/myvg/mylv in order to have > > GFS on top WITH 3 JOURNALS and of course to write on it from those 3 > > MACHINES AT THE SAME TIME? If not, is any other way to achieve -j 3 ...? > > keep it simple. how about > > [as many iSCSI initiator clients as you want, ] > [ all accessing the same iSCSI target formatted as GFS ] > // > .-----------------------,,-----------------------------. > > | iSCSI target cluster (SAN replacement) | > | in "simple" "standard" failover configuration | > | [drbd node a]-=-=-replication link-=-=-[drbd node b] | > > `------------------------------------------------------' Hi Lars, Many thanks fot your answer. That was my intention too... The main reason i asked here about mkgfs -j N (where N>2) is drbd documentation, Chapter 10. Using GFS with DRBD contain partially wrong affirmations! There, we can read: [snip] 1. gfs journals must be identical to the number of nodes in the cluster [end snip] No, No... FALSE ...General speaking, GFS journals must be less then or equal with number of nodes in a cluster, accesing that volume... does not it mean that that number "MUST" be equal with total number of cluster nodes [snip] 2. since DRBD does not support more than two nodes, the value to set here is always 2 [end snip] correct in particular case of drbd - 2 cluster nodes config, but not a must, incomplet explained and maybe partially wrong So, reading above... we can think that DRBD can NOT WORK with 3 or more journals on /dev/drbd0... Reading your explanations, is very clear now that /dev/drbd0 can be mkfs.gfs -j 3 or more...In our case if we are exporting /dev/drbd0 via iscsi, we can use mkgfs -j N (where N>2) on all our N clients which are importing drbd0. So.... in drbd documentation, is missing the 3rd and 4th: 3. do not be laid by the false impression that /dev/drbd0 can not be formatted mkfs.gfs -j N (with N>2) 4. to do that, export /dev/drbd0 via isci (ataoe, gnbd, etc) and use as many iscsi initiator clients you want, all accessing the same iSCSI target formatted as GFS! So, thanks again Lars for your explications. Are essential... Also, 1st explanation should be corrected somehow to point user in the right direction... or, maybe to be completely removed to not create confussions... Can somebody fix that in documentation (http://www.drbd.org/users-guide-emb/s-gfs-create.html) Regards, Alx