Characterizing Shared Data Store Coupling in Microservices: An Empirical Taxonomy and Graduated Remediation Catalog

Microservice architectures encourage each service to own its data, yet many services still share a common data store, reintroducing the coupling the style aims to remove. Existing detection flags such sharing as a binary smell, giving little guidance on how severe a case is or how to resolve it. We study this across 100 open-source systems. From 184 data stores with business consumers we identify 94 shared stores in 62 applications, and characterise each along three dimensions with no detectable mutual association: domain affinity, persistence type, and a continuous Blast Ratio measuring how much of a system is entangled with a store. Cross-domain sharing dominates (72.3%), and within a source-code-verified sample 91% of classifiable consumer–store pairs involve write access, narrowing where read-replica remedies apply. We organise six documented mechanisms into a graduated catalog with severity-calibrated triggers, and model each mechanism’s structural cost using weights from five architects (ICC > 0.98). Relative to full Database-per-Service, the catalog gives a 41% lower modeled structural-cost score, concentrated at low-to-moderate severity; we frame this as modeled cost, not demonstrated saving. Finally, domain affinity does not predict the selected modifiability metrics once system size is controlled.