skip recovery on replica cluster
phil2
✭
Hello!
I have two vertica clusters - one as a master and one as a replica. Eventually one of the replica hosts goes down, and I need to replace it. So I run alter node replace with a standby on replica. It goes thru a full recovery and copies data from buddies. After that I run vbr. It sees that all the data on host changed and copies all the data again from master. Is it possible to skip recovery on replica cluster or somehow make it replication-aware so vertica wont copy the same data twice?
0
Answers
Vertica copies whole data only on first replication. On subsequent replications, it will just replicate the data which has changed/newly added. Please visit the following link for more information.
https://www.vertica.com/docs/9.2.x/HTML/Content/Authoring/AdministratorsGuide/BackupRestore/ReplicatingtoanAlternateCluster.htm
It does not work if you recover one node on replica cluster. It cannot detect delta after recovery as the original files and original epochs get lost during recovery.
We ran into this same issue recently. After replacing a node in DR, the next copycluster to DR was going to have to do a complete copy. Next time we won't bother to allow recovery to run on the DR cluster. Our plan is start the replaced node, see that it starts recovery so that we know everything is good, and then kill the vertica process on that node to stop recovery. I have never tried a cluster shutdown with recovery running - that's also an option if it actually interrupts recovery.
We have 20+TB on each node, so it takes a long time to do a complete copy of a node's data over the network. We still had the original node, intact with the original data, and ended up copying the data from it over to the new node and preserve the timestamps using
scp -rp vertica_data_dir_with_node_name new_host:vertica_data_dir
Then the next copycluster was incremental and fast.
copy cluster works differently than replication, incase of copycluster target is down while in replication both clusters are up.
Recovery in copycluster can be skipped as @Sharon_Cutter mentioned but not in replication. Even though the underlying data is same, newly recovered node has new rosid as compared to the past from source.