Live Cross-Region Migration of Multi-Service Resources with Selective Fencing and Atomic Cutover
Cross-region migration is difficult when one logical resource spans independently owned services and heterogeneous storage backends. Concurrent mutations and stale routing state complicate complete-ness and write authority at cutover. We present a four-phase protocol combining selective operation fencing, content-stripped deletion tombstones, routing-embedded locks, and atomic cutover. His-torical copy permits mutations; final synchronization temporarily fences them. Service-confirmed enforcement closes the mutation boundary, primary-write-replica routing reads expose current lock state, and token- and version-checked cutover transfers authority. We give correctness arguments under explicit service contracts. Production summaries from a messaging platform report 6–10 million conversation migrations per day over 30 days and 99.9% first-attempt success. Successful attempts had write-lock intervals of 13 s at the median and 19 s at the 99th percentile, measured from request issuance through cutover. Approximately 1000 lock-rejected message sends subsequently succeeded according to service-side telemetry; this does not establish absence of user-visible errors or delay. Count and sampled-content validation detected some schema-related contract drift before cutover. An analytical comparison illustrates fencing-window costs under fixed observed durations, without benchmarking alternative implementations. The contribution integrates established mechanisms with production observations and explicit consistency and availability boundaries.
Authors
- Sankalp Mulye
- Debashis Roy (ORCID: https://orcid.org/0009-0005-3374-5679)
- Jay Rajeevalochanam
- Narasimhan Govindarajan
Publication Details
- Journal
- Preprints.org
- Published
- 2026-10-26
- DOI
- https://doi.org/10.20944/preprints202610.0828.v2
- Primary Topic
- Distributed systems and fault tolerance
- Type
- preprint