@memberjunction/materialization

Runtime engine that refreshes materialized query and entity-base-view results in MemberJunction. A materialization is a physical, read-only snapshot of a stored Query's result (or an entity's base view), served through a stable wrapper view so reads can opt in with DataSource: 'Materialized' and fall back to Live on any uncertainty.

This package owns the refresh side of that lifecycle. Provisioning (minting the snapshot table, wrapper view, and metadata) is a CodeGen concern; reading (the DataSource redirect) lives in @memberjunction/generic-database-provider.

What it does

• MaterializationRefresher — rebuilds a snapshot and swaps it into place atomically:
  • Full rebuild — builds a fresh shadow table from the source (the expensive read happens outside any transaction), then swaps it into the canonical name inside a single transaction (drop → rename → repoint the wrapper view → restore the surrogate index). SQL Server uses SET XACT_ABORT ON + sp_rename + EXEC('CREATE OR ALTER VIEW …'); PostgreSQL keeps the CREATE OR REPLACE VIEW repoint inside the transaction. Either way a reader sees the whole old snapshot or the whole new one — never a half-swapped view.
  • Incremental / dirty-group merge — for keyed/aggregation materializations, applies only the changed groups against a watermark, with a periodic forced full rebuild (FULL_REBUILD_EVERY_N_INCREMENTAL_REFRESHES) to reconcile balanced-delete drift.
  • External-source rebuild — parameterized, bound-value inserts for materializations sourced from external data providers.
  • Persists Status/RowCount/LastRefreshedAt/NextRefreshAt via guarded conditional updates that never clobber a concurrent DriftHold/Disabled.
• MaterializationFreshness — helpers for reasoning about whether a snapshot is current.

All statement builders are pure and unit-tested; the identifiers they interpolate are validated as plain SQL identifiers before any DDL is built (a materialization's names come from the writable MJ: Materialized Results row, so this fails closed against a tampered row driving arbitrary DDL). Live-DB behavior is covered by the integration tier.

Usage

The engine is normally driven by the scheduled-job refresh driver (MaterializationRefreshScheduledJobDriver in @memberjunction/scheduling-engine); a manual "refresh now" path can call it directly:

import { MaterializationRefresher } from '@memberjunction/materialization';

const refresher = new MaterializationRefresher();
const result = await refresher.RefreshOne(materializedResult, contextUser, provider);
if (!result.Success) {
  // result.ErrorMessage carries the reason; the row is left in a safe state.
}
Copy

RefreshOne returns a structured result rather than throwing — errors are logged and reported, and a row held for review (DriftHold) or Disabled is refused rather than silently reactivated.

Related

• @memberjunction/generic-database-provider — the DataSource: 'Materialized' read redirect and the materialized read-query builder.
• @memberjunction/codegen-lib — materialization provisioning (mint), parameterized-query classification, and drift detection.
• @memberjunction/scheduling-engine — the scheduled refresh driver.