Distributed execution for external sources currently has no end-to-end test coverage — existing tests only assert plan structure at the unit level. There is no verification that different distribution strategies produce identical results, no randomized validation of split assignment invariants, and no way to test behavior under cloud storage failures. This work builds a layered test suite that catches correctness regressions, validates mathematical properties of distribution, and exercises resilience under fault injection.
Spec-driven distributed integration tests — Run the same queries across all distribution strategies (coordinator-only, round-robin, adaptive) on a multi-node cluster and assert identical results, so any divergence immediately flags a split assignment or aggregation bug.
Distribution property tests — Randomized tests that verify invariants of the split assignment algorithm: every split assigned exactly once, bounded load imbalance, deterministic output, and correct behavior at boundary conditions (single split, zero nodes).
Fault injection tests — Inject transient and persistent cloud storage failures (503, connection reset, slow responses) into a running cluster and verify that retries recover from transient errors and permanent failures produce clear, actionable error messages.
Distributed execution for external sources currently has no end-to-end test coverage — existing tests only assert plan structure at the unit level. There is no verification that different distribution strategies produce identical results, no randomized validation of split assignment invariants, and no way to test behavior under cloud storage failures. This work builds a layered test suite that catches correctness regressions, validates mathematical properties of distribution, and exercises resilience under fault injection.
Spec-driven distributed integration tests — Run the same queries across all distribution strategies (coordinator-only, round-robin, adaptive) on a multi-node cluster and assert identical results, so any divergence immediately flags a split assignment or aggregation bug.
Distribution property tests — Randomized tests that verify invariants of the split assignment algorithm: every split assigned exactly once, bounded load imbalance, deterministic output, and correct behavior at boundary conditions (single split, zero nodes).
Fault injection tests — Inject transient and persistent cloud storage failures (503, connection reset, slow responses) into a running cluster and verify that retries recover from transient errors and permanent failures produce clear, actionable error messages.