Hi, a user is migrating from R9.6 to R10.2.x and would like to better understand the following supported architecture – Failover setup (with geo-redundancy):
The user has two data-centers (900km apart, 20ms RTT), DC-1 and DC-2.
(Q1) They would like to confirm that the following setup is supported?:
- A Cassandra cluster in DC-1 and a geo-redundant Cassandra cluster in DC-2
- An Elastic cluster in DC-1 and a geo-redundant Elastic cluster in DC-2
- Six DMAs in DC-1 with corresponding failover-pairs in DC-2
(Q2) Is it supported to have different DMAs running in DC-1 and DC-2 – for example DMA1, DMA3, DMA5 are active in DC-1 and DMA2, DMA4, DMA6 are active in DC-2?
(Q3a) Given the situation in Q2, from Cassandra perspective does it mean DMA1, DMA3, DMA5 will read & write to the Cassandra cluster in DC-1 and DMA2, DMA4, DMA6 will read & write to the Cassandra cluster in DC-2?
(Q3b) Is the data replication between the two clusters in DC-1 and DC-2 bi-directional and purely handled by Cassandra?
(Q4) From our help file, it is called out that the paid feature of replication of data between two Elastic clusters is not supported. And that DataMiner supports geo-redundancy by writing to two geo-redundant Elastic clusters. Given the situation in (Q2) is it correct understanding that:
- DMA1, DMA3, DMA5 write to Elastic cluster in DC-1 and DC-2 but only read from Elastic cluster in DC-1?
- DMA2, DMA4, DMA6 write to Elastic cluster in DC-1 and DC-2 but only read from Elastic cluster in DC-2?
Hi Bing,
(Q1) yes
We support offloading towards two Elastic clusters and reading from one. There is currently no way to synchronize the two if they would get out of sync. When one elastic cluster would go down, DM will go into offload mode in order to prevent getting out of sync, but once they are out of sync, there is no tooling atm to get them back in sync other than restoring the data from one cluster to the other. More information can be found in our docs: Configuring multiple Elasticsearch clusters | DataMiner Docs
For Cassandra, multiple data centers are supported out of the box of Cassandra.
(Q2) yes
(Q3a) DM will connect towards the Cassandra Cluster with DCAwareRoundRobinPolicy. DM will pick the local DC based on the first IP in your db.xml file.
(Q3b) When using multiple data centers in Cassandra it is advised to use the NetworkTopologyStrategy which allows you to define an RF for every DC.
(Q4) As indicated in our docs, you can configure a priorityOrder to indicate which Elastic cluster should be preferred by the DMA.