Benchmark

Measures the overhead added by the Prisma tenancy extension compared with a manual RLS transaction that returns the same tenant-scoped rows.

What We Measure

Benchmark	Description
Admin direct findMany	Context only; not used as the extension overhead baseline
Admin tenant-filtered findMany	Same returned row count with RLS bypassed
Manual RLS transaction	app_user query with set_config() + RLS active, no extension
Tenancy extension findMany	Same role, RLS policy, and returned row count as the manual RLS transaction
Tenancy extension findFirst	Single-row reference path

The extension wraps every query in a $transaction that calls SET LOCAL app.current_tenant = ? before executing the actual query. The headline overhead is extension findMany minus the manual RLS transaction, not extension versus an unfiltered admin query.

Test Setup

• Database: PostgreSQL 16 (Docker)
• Data: 1,005 rows across 3 tenants
• Warmup: 50 iterations (discarded)
• Measured: 500 iterations per benchmark

Running Locally

# Start PostgreSQL
docker compose up -d --wait

# Run benchmark
npm run bench

Results

Measured on Apple M1 Pro, Node v24.11.1, PostgreSQL 16.13, Prisma Client 6.19.2, local Docker. Your results will vary.

Scenario	Rows	Avg	P50	P95	P99
Admin direct findMany (all rows, no RLS)	1005	3.983ms	3.369ms	5.444ms	6.992ms
Admin tenant-filtered findMany (WHERE tenant_id, no RLS)	402	2.747ms	2.736ms	3.612ms	4.686ms
app_user manual RLS transaction (set_config + findMany)	402	2.846ms	2.614ms	4.154ms	5.177ms
app_user tenancy extension findMany	402	2.961ms	2.766ms	4.281ms	4.800ms
app_user tenancy extension findFirst	1	1.217ms	1.192ms	1.522ms	1.777ms

Extension overhead (avg): +0.115ms (+4.0%) compared with the manual RLS transaction Extension overhead (P95): +0.127ms

Interpretation

The unfiltered admin query is useful context, but it is not the extension overhead baseline because it returns all tenant rows. Compare the tenancy extension with the manual RLS transaction when estimating extension cost: both use the app_user role, SET LOCAL, RLS policy enforcement, and the same returned row count.

In this local run, the extension adds about 0.1ms over the manual RLS transaction. For single-row lookups (findFirst), the measured path stays near 1.2ms.

For most API endpoints (10-50ms total), this overhead is small compared with network, application, and serialization cost. Always benchmark with your schema, indexes, connection pool, and tenant row distribution.

Methodology

• performance.now() measures wall-clock time per query
• Warmup iterations ensure connection pool and query plan caches are hot
• Percentiles computed from sorted timing arrays (no outlier removal)
• The script prints row counts, Node/PostgreSQL/Prisma versions, and p50/p95/p99 timings so results can be compared across environments
• Both clients connect to the same PostgreSQL instance; admin baselines bypass RLS, while manual and extension paths use app_user with RLS enforced