Question 1
A financial services firm is modernizing its loan origination process using IBM BAW deployed on Red Hat OpenShift. The current architecture involves a mix of on-premises legacy systems and new containerized microservices. The lead architect needs to design a solution that ensures high availability and disaster recovery for the BAW components. The primary data center is in New York, with a secondary DR site in Chicago. The Recovery Time Objective (RTO) is 4 hours, and the Recovery Point Objective (RPO) is 15 minutes.
The proposed architecture uses a primary OpenShift cluster in New York and a standby cluster in Chicago. A Global Load Balancer will direct traffic. For the BAW database, which is external to the cluster, the database administration team has configured asynchronous replication between the two sites with a typical lag of 5-10 minutes. The content store (BPM content store) is located on a shared NFS mount that is also replicated asynchronously.
During a DR test, the team successfully fails over the OpenShift cluster and the database. However, after bringing up the BAW application pods in Chicago, they discover that process instances started in the last 10 minutes before the failure are missing, and some users report seeing outdated versions of documents attached to their tasks. The team needs to select a strategy that will meet the RPO without requiring a complete re-architecture to synchronous replication, which the network team has advised against due to latency.
Which strategy should the architect recommend to address the data consistency issues and meet the RPO?
Answer and explanation
Correct answer: D
This is the most practical and effective solution within the given constraints. BAW and its underlying database support transaction logging. By placing these logs on a synchronously replicated, low-latency storage device (which is feasible across data centers for smaller log files), the application can recover the transactions that occurred after the last asynchronous database snapshot. This meets the 15-minute RPO by reconciling the data gap without the high cost and latency issues of full synchronous database replication. Increasing async frequency (A) doesn't guarantee zero data loss. A two-phase commit (B) is extremely complex to implement and manage. Container-native synchronous storage (C) is a valid but much larger architectural change and might violate the 'no synchronous replication' network constraint for large volumes.