Case study
Cross-Cluster Replication Upgrade for a Fintech Firm
How a fintech firm in England upgraded Elasticsearch, Kibana, and its Azure Repository plugin while keeping cross-cluster replication intact.
A global financial-technology company headquartered in England, powering retail banking, transaction, lending, treasury, and capital-markets services.
It serves banks and financial institutions worldwide.
Elasticsearch and Kibana upgraded from 7.17.10, in lockstep. Cluster nodes upgraded in one rolling pass, no rebuild. CCR, Azure snapshots, and Kibana licensing confirmed post-upgrade.
Key servicesHow SquareShift delivered it.
The challenge
The client ran a self-hosted Elasticsearch 7.17.10 cluster with uni-directional cross-cluster replication (CCR) between its primary and disaster-recovery sites, and needed to move to 7.17.15 without disrupting the retail banking and transaction systems running on top of it.
The complexity sat in the details: a multi-node CCR configuration, a sensitive integration with Azure Blob snapshots used for backup, and no room for a botched rollback given what the cluster underpins.
What we delivered
SquareShift designed a detailed upgrade and rollback strategy before touching production, then executed the upgrade as a rolling pass across the 5-node cluster rather than a single cutover.
Elasticsearch, Kibana, and the Azure Repository plugin were upgraded together, and the CCR setup, Azure snapshot integration, and Kibana licensing configuration were carried through the process intact — then explicitly validated once the upgrade completed, not just assumed to be fine.
The payoff
The cluster now runs 7.17.15, with cross-cluster replication and Azure Blob snapshot integration unchanged, and the rollback plan built for this upgrade is ready to reuse for the next one.
Replication, snapshots, and Kibana licensing were all confirmed working post-upgrade — the bank’s disaster-recovery posture came through the version bump unchanged, not weakened to get it done.
When cross-cluster replication ties two clusters together, the safest upgrade path is the boring one — validate rollback before you validate anything else.
Cloud Modernization Practice Lead, SquareShift
Where this work sits