Financial services · 2025 · Client anonymised at their request
Moving eleven million accounts off a mainframe without a maintenance window
Zero-downtime migration of a core ledger, completed over fourteen months.
A tier-one East African bank
11M
Accounts migrated
0
Minutes of planned downtime
340ms
p99 ledger write, down from 2.1s
The constraint
The existing ledger could not be taken offline for longer than the four-hour nightly batch window, and the regulator required a provable audit trail for every migrated balance. A cutover-weekend approach was not available to us.
The architecture decision
We built a change-data-capture bridge that replicated the mainframe ledger into a PostgreSQL system of record in near real time, then ran both ledgers in parallel with a continuous reconciliation job asserting balance equality at the account level. Writes were dual-written behind a feature-flagged router, so traffic moved account-cohort by account-cohort and could be reversed per cohort within one minute. The reconciliation output was itself the audit artefact the regulator accepted.
What shipped
- 01
CDC replication bridge with per-account reconciliation
- 02
Dual-write router with cohort-level rollback
- 03
Regulator-facing reconciliation reporting
- 04
Runbooks and operator training for the internal platform team
The bank retired its mainframe lease a quarter ahead of plan. Ledger write latency dropped by an order of magnitude, and the reconciliation harness stayed in production as an ongoing control.
Figures verified with the client’s own instrumentation, not ours.
What it moved
Practices involved
- Product engineering
- Cloud infrastructure
- DevOps & scaling