> > This analysis is flawed. The two event numbers are from different origins > and therefore, don't compare to each other. The combination of > ev_origin,ev_seqno can never be higher on any node, than it is on the > origin itself. > > Thanks Jan. For correcting me, let me recheck thoroughly from my side why I concluded on the seqno.
> Your disaster recovery plan assumes, that streaming replication will > ALLWAYS be faster than Slony replication. But that is only true for > synchronous streaming replication. If you use asynchronous streaming > replication, then one tiny network glitch and your DR-master will be > several seconds behind while the Slony replica may be not. If the master > blows up in that moment, your plan fails. > > Yes, very much true and completely agreed. Aim is to not to have any lag between Master & DR-Master. As you said, in any case if Master & DR-Master are not in Sync due to network or any lag then my whole plan collapses. So, I made my goal clear here, that I need to STOP slony between Master & Slave long before promoting DR-master as Master, to make sure nothing is left on Master to update on DR-master. > To simulate this problem, Steve and I were pointing out, do the following: > > 1. Create your setup as before. > 2. Stop the streaming replication (simulating the network communication > problem) > 3. Update a row on the master and wait for the SYNC to replicate. > 4. Stop the slon processes. DO NOT let the streaming replica catch up with > the now DEAD master. Assume the master and all its data, including WAL, > have become unavailable. > 5. Promote DR-master and do the two store path commands. > 6. Start slon processes. > 7. Update another row on the new master. > 8. Compare table content on master and slave. > > You can detect the problem before step 5 by comparing the ev_seqno with > ev_origin=old-master on the DR-master and slave. Whichever is higher should > be promoted to master. In the unlikely case that it is the Slony slave, you > will have to rebuild the DR-master from scratch, though. > > Exactly, these are the steps I followed and succeded, as pointed I never simulated Step 2,3 in my testing, because If MASTER and newly promoted DR-Master are not same in any case then the success is nowhere near by surroundings :) . I will surely retake the test with the steps mentioned on my two VM's and update my finding. Once again, Thank you very much to you & Steve for helping me in this scenario. -- Regards Raghav Blog: htt://raghavt.blogspot.com/
_______________________________________________ Slony1-general mailing list Slony1-general@lists.slony.info http://lists.slony.info/mailman/listinfo/slony1-general
