Adaptive Just-In-Time Query Compilation and Memory-Bounded Indexing for Evolving Enterprise Relational Schemas

High-throughput enterprise database backends experience continuous relational divergence between static physical designs and dynamic, application-driven query workloads. In modern multi-tenant environments, Object-Relational Mapping (ORM) frameworks emit queries stripped of semantic context, causing redundant join execution, predicate bloat, and sub-optimal scan paths. Simultaneously, database administrators over-provision secondary B-tree indexes to mitigate analytical latency spikes. This practice induces catastrophic write amplification, secondary buffer pool cache thrashing, and uncontrolled memory consumption during concurrent transactional ingestion. This paper presents the design, formalization, and empirical evaluation of the Autonomic Enterprise Data Engine (AEDE), an online system framework unifying application-aware Just-In-Time (JIT) query rewriting with memory-bounded adaptive secondary indexing. First, AEDE introduces a semantic query rewriting pass that extracts multi-tenant scoping invariants and append-only ledger properties directly from application telemetry, compiling specialized execution pipelines that prune redundant joins prior to physical plan generation. Second, AEDE formulates a control-theoretic online utility function that continually balances query latency reduction against secondary index memory footprint, write amplification, and structural reconfiguration overhead under a strict RAM budget. Third, index transitions are governed by a five-state non-blocking lifecycle automaton (Candidate, Building, Active, Decaying, Evicted) that preserves Snapshot Isolation during concurrent online transactional mutations. We evaluate AEDE through a reproducible benchmark executing mixed analytical and transactional enterprise workloads across 50,000 ledger records. Compared to unindexed baselines, AEDE reduces p99 query latency from 9.299 ms to 0.458 ms (a 95.1% reduction). Crucially, compared to static over-indexing baselines, AEDE contracts secondary index storage from 4,736 KB to 1,280 KB (a 73.0% memory reduction) while reducing p99 transactional insert latency by 21.2% (from 3.144 ms down to 2.476 ms). These results establish that memory-bounded adaptive physical reorganization provides a dependable, high-throughput foundation for modern enterprise relational data systems.Visual Verification