Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
Hi, Lionel Sausin wrote: > Dear DRBD community, > We are setting up a 2+1 node cluster : > - A is master > - B is slave > - S is an additional spare/backup machine, disconnected most of the time > The use cases are : > - Normally we sync S once a week then unplug it and keep it safe. The > dataset is big (26To) so it has to catch up fast. > - Of course when A fails we promote B > - But when either A or B fails, we still want HA: so we plug S in and > keep it synced until the failed machine is repaired > > So we intend to set up DRBD between A and B, and use a stacked DRBD > resource to replicate (A+B) to S. > > My question is: if A or B dies and can't be fixed, can S be > reconfigured to replace it without a full sync? You usually leave S disconnected, that's why you need a full sync to bring S up to speed, but normally what you would do when using stacked resources would be to configure S with protocol A, this is actually the recommendation in the drbd.org docs http://www.drbd.org/users-guide/s-three-nodes.html#s-three-node-config This way S will have the most recent information with as little performance overhead as possible due to protocol A. > In other words: can a setup like (A->B)->S be changed to (A->S) with a > quick sync? You need A->B, and you can only setup one replication link that spans 2 servers (e.g.: you can't set A's /dev/drbdx to replicate to both B and S at the same time), the only way to do it is to set normal A->B, and on top of that AB->S, because you don't know if A or B will be up and don't want to manually choose (if it was somehow possible) to sync A->S or B->S depending on which (A or B) is up. Stacked resources does that for you automagically, it syncs whomever is Primary to the stacked resource (S). Read up on differences between protocol C and A, if you haven't already, test it to see what is the performance penalty due to the overhead (27To -> you must be French :P) of the large dataset will be. Come back if you have more questions. Regards, Dan > > Thanks in advance if you can share your knowledge. > > Lionel Sausin > _______________________________________________ > drbd-user mailing list > drbd-user at lists.linbit.com > http://lists.linbit.com/mailman/listinfo/drbd-user -- Dan FRINCU Systems Engineer CCNA, RHCE Streamwide Romania -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.linbit.com/pipermail/drbd-user/attachments/20110112/bce8ae8d/attachment.htm>