JSONB Analytics Snapshot Pattern

This skill provides the architectural pattern for using JSONB to store pre-aggregated analytics data, rather than relying on on-the-fly RDBMS calculations or complex star-schema table architectures.

The Core Concept

When building dashboards or analytics views, data is often grouped by multiple dimensions (e.g., status, type, template, date). If history is immutable (i.e. "what happened has happened"), recalculating these aggregations on every page load using GROUP BY and multiple JOIN s is inefficient.

The JSONB Analytics Snapshot Pattern solves this by calculating the aggregations once for a specific time period (e.g., Daily, Weekly) and storing the complete structured result in a single JSONB column.

RDBMS vs JSONB for Aggregations

Why not create normalized RDBMS tables for the aggregated results (e.g., analytics_overview , analytics_by_status , analytics_by_template )?

• Schema Flexibility: Dashboards change frequently. If you want to add a new chart (e.g., "Performance by User Role"), an RDBMS approach requires a database migration, new tables/columns, and ORM updates. With JSONB, you simply add a new key to the JSON payload.

• Read Performance: A dashboard typically needs all these aggregations at once. Fetching from 5 different RDBMS aggregate tables requires 5 queries or complex joins. Fetching a single JSONB row provides the entire localized payload in one fast read.

• Data Shape Match: The JSONB structure can exactly match the API response DTO mapped to the frontend charts, avoiding mapping boilerplate.

When NOT to use: Do not use JSONB if you need to perform cross-snapshot RDBMS aggregations (e.g., summing up the 'completed' count across 100 snapshots using SQL). In our system, snapshots are usually requested individually per-period.

Prisma Schema Implementation

To implement this pattern, create a snapshot table scoped to the entity and the time period:

model TaskAnalyticsSnapshot { id BigInt @id @default(autoincrement()) uid String @unique

// 1. Scoping (Who does this belong to?) studioId BigInt @map("studio_id") studio Studio @relation(fields: [studioId], references: [id], onDelete: Cascade) userId BigInt? @map("user_id") // Optional: if scoped to a specific user user User? @relation(fields: [userId], references: [id], onDelete: Cascade)

// 2. Time Bounding (What period does this cover?) periodStart DateTime @map("period_start") periodEnd DateTime @map("period_end")
periodType String @map("period_type") // e.g. 'DAILY', 'WEEKLY', 'MONTHLY', 'ALL_TIME'

// 3. The Payload metrics Json @map("metrics")

createdAt DateTime @default(now()) @map("created_at")

@@index([studioId, periodType, periodStart]) @@index([userId, periodType, periodStart]) @@map("task_analytics_snapshots") }

Application Layer (NestJS)

1. The DTO Match

Keep the Zod/DTO schema aligned with the metrics JSON structure.

export const analyticsMetricsSchema = z.object({ overview: z.object({ totalTasks: z.number(), completionRate: z.number(), overdueCount: z.number(), }), byStatus: z.array(z.object({ status: z.string(), count: z.number() })), byType: z.array(z.object({ type: z.string(), count: z.number() })), }); export type AnalyticsMetrics = z.infer<typeof analyticsMetricsSchema>;

2. Lazy Evaluation / Generation

If a webhook/cron job isn't viable yet, use a "Lazy Evaluation" pattern in the Service layer:

• Check if a snapshot exists for the requested exact period/scope.

• If YES: Parse and return snapshot.metrics .

• If NO: Calculate the aggregations via Prisma groupBy , construct the metrics JSON, save it to the database, and return it.

Because historical periods (e.g., "Last Month") are immutable, once the snapshot is generated, it never costs aggregation CPU cycles again.

Checklist

• Snapshot table includes scoping fields (studioId , optional userId )

• Time bounding fields present (periodStart , periodEnd , periodType )

• metrics column is Json type (not normalized RDBMS tables)

• DTO/Zod schema mirrors the metrics JSON structure

• Lazy evaluation: check for existing snapshot before calculating

• Indexes cover (studioId, periodType, periodStart) for efficient lookups

• Snapshot is immutable after creation (historical periods don't change)