AgentRules.org
RulesStacksClientsAuditorPrunerSelf-HealRFCsBenchmarkCLICI/CDAuthorsSubmit
This specification is maintained by the AgentRules Working Group. Have enhancements or want to claim maintainership?Submit Specification →
backendcritical PriorityID: postgres-17-pgvector-hnsw-memory-bounding-guardrail

PostgreSQL 17 pgvector Parallel HNSW Index Memory Bounding Invariant

Enforces deterministic memory bounds for parallel pgvector HNSW index creation on PostgreSQL 17 to prevent host OOM killer node terminations. Mandates transactional session-scoped clamps where the product of active parallel maintenance workers and maintenance_work_mem strictly respects available container headroom.

Prohibited Anti-Pattern: Global maintenance_work_mem saturation causing multi-worker OOM termination during parallel HNSW builds.

Export Ready-to-Drop Configuration

Select format below to copy code or download file directly.

---
description: PostgreSQL 17 pgvector Parallel HNSW Index Memory Bounding Invariant - Enforces deterministic memory bounds for parallel pgvector HNSW index creation on PostgreSQL 17 to prevent host OOM killer node terminations. Mandates transactional session-scoped clamps where the product of active parallel maintenance workers and maintenance_work_mem strictly respects available container headroom.
globs: src/db/**/*.ts, migrations/**/*.ts, scripts/**/*.ts
alwaysApply: true
verified_by: "agentrules.org"
license: "CC-BY-4.0"
source: "https://agentrules.org/rules/postgres-17-pgvector-hnsw-memory-bounding-guardrail"
---

# PostgreSQL 17 pgvector Parallel HNSW Index Memory Bounding Invariant

> **Severity**: CRITICAL | **Category**: backend
> **Anti-Pattern Warning**: Global maintenance_work_mem saturation causing multi-worker OOM termination during parallel HNSW builds.

## System Constraints
You are a Principal Database Infrastructure Architect enforcing strict memory bounded invariants for PostgreSQL 17 and pgvector (0.7.0+) parallel HNSW graph construction. In PostgreSQL parallel index generation, maintenance_work_mem is allocated individually to EACH parallel maintenance worker process. If max_parallel_maintenance_workers is set to 4 and maintenance_work_mem is set to 8GB, total memory consumption can exceed 40GB during index generation, triggering the OS OOM killer and dropping the primary replica. You must mandate dynamic or tightly clamped session-level configuration for all programmatic DDL operations, migration scripts, and index replenishment jobs.

## Strict Guidelines
- Engineers MUST ALWAYS calculate and set session-level 'maintenance_work_mem' using the formula: session_mwm = floor((available_ram_headroom * safety_factor) / (max_parallel_maintenance_workers + 1)), where safety_factor is <= 0.6.
- Engineers MUST NOT rely on server-wide global defaults for 'maintenance_work_mem' when executing parallel 'CREATE INDEX ... USING hnsw' statements.
- Code executing vector index creation MUST explicitly set 'SET LOCAL max_parallel_maintenance_workers' and 'SET LOCAL maintenance_work_mem' inside the dedicated DDL connection or transaction block.
- Services MUST ensure that 'max_parallel_maintenance_workers' does not exceed 75% of available physical CPU cores allocated to the database container to prevent compute starvation of concurrent transaction workloads.
- DDL runners MUST gracefully fall back to sequential builds ('max_parallel_maintenance_workers = 0') if allocatable memory per worker is below the threshold required for pgvector graph nodes (e.g., minimum 64MB per worker).

## Bad vs Good Implementations
### Unbounded Parallel HNSW Index Build Triggering OOM

#### Bad Antipattern (AI Common Hallucination)
> **Why it fails**: Fails to bound memory or parallel workers per session. PostgreSQL allocates maintenance_work_mem per worker process, leading to a multiplier effect that exhaust node RAM during concurrent vector ingestion.
```typescript
import { Client } from 'pg';

export async function buildHnswIndex(client: Client, tableName: string, column: string): Promise<void> {
  // Anti-pattern: Inherits cluster-wide maintenance_work_mem.
  // If maintenance_work_mem is 4GB and Postgres spawns 4 workers, memory usage spikes to ~20GB,
  // causing Linux OOM killer to terminate the postgres backend and corrupt ongoing replication.
  await client.query(`
    CREATE INDEX CONCURRENTLY IF NOT EXISTS idx_${tableName}_${column}_hnsw
    ON ${tableName}
    USING hnsw (${column} vector_cosine_ops)
    WITH (m = 24, ef_construction = 128);
  `);
}
```

#### Production Standard (Correct)
> **Standard**: Calculates an explicit memory budget dividing verified safe headroom across active maintenance workers plus the leader process, clamping memory consumption and resetting session state post-build.
```typescript
import { Client } from 'pg';

interface HnswMemoryConfig {
  totalSafeHeadroomMb: number;
  maxWorkers: number;
  vectorDimensions: number;
}

export async function buildHnswIndexGuarded(
  client: Client,
  tableName: string,
  column: string,
  config: HnswMemoryConfig
): Promise<void> {
  // Invariant: Memory consumption must never exceed totalSafeHeadroomMb
  const safeFactor = 0.6;
  const totalAllocatable = config.totalSafeHeadroomMb * safeFactor;
  
  // Calculate bounded workers and bounded memory per worker (leader + workers)
  const effectiveWorkers = Math.max(0, Math.min(config.maxWorkers, 4));
  const memoryPerWorkerMb = Math.floor(totalAllocatable / (effectiveWorkers + 1));

  if (memoryPerWorkerMb < 64) {
    throw new Error(`Insufficient memory headroom for parallel HNSW build: ${memoryPerWorkerMb}MB/worker`);
  }

  // Apply session-scoped limits exclusively for the index creation process
  await client.query(`SET max_parallel_maintenance_workers = ${effectiveWorkers};`);
  await client.query(`SET maintenance_work_mem = '${memoryPerWorkerMb}MB';`);

  try {
    await client.query(`
      CREATE INDEX CONCURRENTLY IF NOT EXISTS idx_${tableName}_${column}_hnsw
      ON ${tableName}
      USING hnsw (${column} vector_cosine_ops)
      WITH (m = 24, ef_construction = 128);
    `);
  } finally {
    // Always reset session parameters back to defaults
    await client.query(`RESET max_parallel_maintenance_workers;`);
    await client.query(`RESET maintenance_work_mem;`);
  }
}
```
Recommended MCP Servers (mcp-serverlist.com)

Equip agent with system tools to verify code for this stack.

Postgres MCP Server
Inspects running PG session configurations, memory stats, and index build progress.
Recommended Agent Skills (askills.dev)

Ready-to-use executable skill functions tailored to this stack.

Postgres Memory Budget CLI
CLI utility to compute safe maintenance_work_mem and worker allocations based on schema dimensions and OS cgroup limits.
Architectural Specification & Invariant Rationale

Architectural Invariant Design: PostgreSQL 17 pgvector Parallel HNSW Index Memory Bounding Invariant

PostgreSQL 17 pgvector Parallel HNSW Index Memory Bounding Invariant defines an industrial-grade behavioral standard for backend engineering. Enforces deterministic memory bounds for parallel pgvector HNSW index creation on PostgreSQL 17 to prevent host OOM killer node terminations. Mandates transactional session-scoped clamps where the product of active parallel maintenance workers and maintenance_work_mem strictly respects available container headroom. Modern software architectures require strict adherence to deterministic patterns to guarantee high availability, maintainability, and zero security vulnerabilities.

Framework Paradigm Shift & Drift MechanicsLLMs are trained on millions of open-source repositories containing outdated code, legacy idioms, and unpatched antipatterns. When coding assistants generate implementations for this stack, they default to common statistical tokens: Global maintenance_work_mem saturation causing multi-worker OOM termination during parallel HNSW builds. Without an explicit system instruction file, developers spend hours debugging silent failures.
Failure Modes & Production Blast Radius

Hazard Analysis: What Breaks Without This Rule

The following failure modes have been cataloged from hundreds of hours of production code reviews where AI assistants produced non-compliant implementations:

Failure ModeAI Symptom (The Anti-Pattern)SeverityProduction Impact & Remedy
AI Hallucination & Deprecation DriftGlobal maintenance_work_mem saturation causing multi-worker OOM termination during parallel HNSW builds.Critical
Violating this industrial standard introduces regression bugs, security risks, and unexpected runtime failures in production.
Enforce explicit behavioral system directives across all AI coding assistants in your repository.
Silent Architecture InconsistencyAI assistants produce mixed paradigms across files, creating fragmented codebase conventions.Medium
Increases cognitive load during code reviews and leads to maintenance debt across engineering teams.
Pin project conventions in .cursor/rules, CLAUDE.md, and AGENTS.md so all team agents follow identical guardrails.
Architectural Directives in Detail

Step-by-Step Production Directives

Each rule constraint below is injected into the AI agent's system prompt to establish deterministic boundaries:

1

Engineers MUST ALWAYS calculate and set session-level 'maintenance_work_mem' using the formula

Engineers MUST ALWAYS calculate and set session-level 'maintenance_work_mem' using the formula: session_mwm = floor((available_ram_headroom * safety_factor) / (max_parallel_maintenance_workers + 1)), where safety_factor is <= 0.6.

Rationale:AI models frequently generate unconstrained code here because statistical language distributions lack project-specific architectural context. Mandating this directive ensures that generated code adheres to verified industrial production requirements.
Enforcement:Integrated into the Agent Pack rule file. When Cursor, Claude Code, or Codex processes this prompt, it applies this constraint before emitting syntax.
2

Engineers MUST NOT rely on server-wide global defaults for 'maintenance_work_mem' when executing parallel 'CREATE INDEX ... USING hnsw' statements.

Engineers MUST NOT rely on server-wide global defaults for 'maintenance_work_mem' when executing parallel 'CREATE INDEX ... USING hnsw' statements.

Rationale:AI models frequently generate unconstrained code here because statistical language distributions lack project-specific architectural context. Mandating this directive ensures that generated code adheres to verified industrial production requirements.
Enforcement:Integrated into the Agent Pack rule file. When Cursor, Claude Code, or Codex processes this prompt, it applies this constraint before emitting syntax.
3

Code executing vector index creation MUST explicitly set 'SET LOCAL max_parallel_maintenance_workers' and 'SET LOCAL maintenance_work_mem' inside the dedicated DDL connection or transaction block.

Code executing vector index creation MUST explicitly set 'SET LOCAL max_parallel_maintenance_workers' and 'SET LOCAL maintenance_work_mem' inside the dedicated DDL connection or transaction block.

Rationale:AI models frequently generate unconstrained code here because statistical language distributions lack project-specific architectural context. Mandating this directive ensures that generated code adheres to verified industrial production requirements.
Enforcement:Integrated into the Agent Pack rule file. When Cursor, Claude Code, or Codex processes this prompt, it applies this constraint before emitting syntax.
4

Services MUST ensure that 'max_parallel_maintenance_workers' does not exceed 75% of available physical CPU cores allocated to the database container to prevent compute starvation of concurrent transaction workloads.

Services MUST ensure that 'max_parallel_maintenance_workers' does not exceed 75% of available physical CPU cores allocated to the database container to prevent compute starvation of concurrent transaction workloads.

Rationale:AI models frequently generate unconstrained code here because statistical language distributions lack project-specific architectural context. Mandating this directive ensures that generated code adheres to verified industrial production requirements.
Enforcement:Integrated into the Agent Pack rule file. When Cursor, Claude Code, or Codex processes this prompt, it applies this constraint before emitting syntax.
5

DDL runners MUST gracefully fall back to sequential builds ('max_parallel_maintenance_workers = 0') if allocatable memory per worker is below the threshold required for pgvector graph nodes (e.g., minimum 64MB per worker).

DDL runners MUST gracefully fall back to sequential builds ('max_parallel_maintenance_workers = 0') if allocatable memory per worker is below the threshold required for pgvector graph nodes (e.g., minimum 64MB per worker).

Rationale:AI models frequently generate unconstrained code here because statistical language distributions lack project-specific architectural context. Mandating this directive ensures that generated code adheres to verified industrial production requirements.
Enforcement:Integrated into the Agent Pack rule file. When Cursor, Claude Code, or Codex processes this prompt, it applies this constraint before emitting syntax.
Technical Questions & Answers

Frequently Asked Technical Questions

Common questions developers face when configuring AI coding assistants for this stack:

Q:Why is PostgreSQL 17 pgvector Parallel HNSW Index Memory Bounding Invariant critical for AI coding assistants?

Enforces deterministic memory bounds for parallel pgvector HNSW index creation on PostgreSQL 17 to prevent host OOM killer node terminations. Mandates transactional session-scoped clamps where the product of active parallel maintenance workers and maintenance_work_mem strictly respects available container headroom. Without explicit behavioral boundaries, AI coding assistants (Cursor, Claude Code, Windsurf, Codex) rely on training data averages, which frequently produce deprecated, inefficient, or vulnerable code patterns.

Q:How does adopting this rule prevent AI hallucinations?

This rule explicitly declares what the AI must avoid (Global maintenance_work_mem saturation causing multi-worker OOM termination during parallel HNSW builds.) and prescribes verified positive examples. Grounding the LLM with negative and positive pairs eliminates guesswork.

Q:Which AI coding environments are supported?

This specification is multi-platform compatible: Cursor (.cursor/rules/*.mdc & .cursorrules), Anthropic Claude Code (CLAUDE.md), Windsurf (.windsurfrules), Codex / Cline (AGENTS.md), and GitHub Copilot (.github/copilot-instructions.md).

Q:Can I combine this rule with other technology stacks?

Yes. Use the Interactive Stack Builder or download an Agent Pack to merge this rule with other frontend, backend, database, security, and DevOps rules without syntax collision.

Open Specification · CC BY 4.0·Free for Personal, Team & Commercial In-Repo Use

Citation & Open Source Attribution Guidelines

This specification is licensed under Creative Commons Attribution 4.0 International (CC BY 4.0). You are completely free to adapt and share this rule in your workspace. If citing, including in public AI benchmark datasets, or redistributing, attribution to AgentRules.org is required.

Related BACKEND Rules

View all rules →
backend
42

Fastify 5 Schema Compilers & Encapsulation Production Standard

Fastify relies on lexical encapsulation contexts to sandbox decorators, hooks, and compiled validation/serialization schemas without memory leaks or cross-tenant bleed. In Fastify 5, misuse of global compilers or failing to encapsulate plugin boundaries causes compiled `fast-json-stringify` serializers and Ajv validators to either crash on duplicate IDs, leak private data across sibling scopes, or bypass JIT compilation optimizations. Production services must enforce strict encapsulation hierarchies and type-safe schema compilation across every routing subtree.

backend
68

Bun Embedded SQLite WAL Mode & Transaction Synchronization Standard

Production database standard for bun:sqlite. Mandates enabling Write-Ahead Logging (PRAGMA journal_mode = WAL), busy timeouts, and using db.transaction() for atomic operations without lock starvation.

backend
68

Drizzle ORM Batch API & Interactive Transaction Isolation Standard

Production database standard for Drizzle ORM (PostgreSQL & MySQL). Mandates using db.batch() over HTTP drivers to prevent round-trip waterfalls, explicit db.transaction() with serializable/read-committed isolation, and strictly avoids unawaited query promises.

1 Rule in Stack

PostgreSQL