AZ

azure-cosmosdb

Architectural best practices for Azure Cosmos DB scaling and configuration.

Install

mkdir -p .claude/skills/azure-cosmosdb && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/6107" && unzip -o skill.zip -d .claude/skills/azure-cosmosdb && rm skill.zip

Installs to .claude/skills/azure-cosmosdb

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Azure Cosmos DB partition keys, consistency levels, change feed, SDK patterns
77 charsno explicit “when” trigger
Intermediate

Key capabilities

  • Determines optimal partition keys for scalability
  • Configures consistency levels for read/write trade-offs
  • Designs efficient containers for NoSQL access patterns
  • Implements logic for Change Feed change tracking
  • Optimizes throughput via RU request unit management

How it works

It applies NoSQL design heuristics to evaluate partitioning strategies based on cardinality, distribution, and query performance requirements.

Inputs & outputs

You give it
Data access patterns and container requirements
You get back
Partitioning strategy and Cosmos DB configuration

When to use azure-cosmosdb

  • Designing partition keys for scaling
  • Choosing consistency levels for apps
  • Implementing Cosmos DB change feed

About this skill

Core Principle

Choose partition key wisely, design for your access patterns, understand consistency tradeoffs.

Cosmos DB distributes data across partitions. Your partition key choice determines scalability, performance, and cost. Design for even distribution and query efficiency.


Cosmos DB APIs

APIUse Case
NoSQL (Core)Document database, most flexible
MongoDBMongoDB wire protocol compatible
PostgreSQLDistributed PostgreSQL (Citus)
Apache CassandraWide-column store
Apache GremlinGraph database
TableKey-value (Azure Table Storage compatible)

This skill focuses on NoSQL (Core) API - the most common choice.


Key Concepts

ConceptDescription
ContainerCollection of items (like a table)
ItemSingle document/record (JSON)
Partition KeyDetermines data distribution
Logical PartitionItems with same partition key
Physical PartitionStorage unit (max 50GB, 10K RU/s)
RU (Request Unit)Throughput currency

Partition Key Design

Good Partition Keys

// High cardinality, even distribution, used in queries

// E-commerce: userId for user data
{ "id": "order-123", "userId": "user-456", ... }  // PK: /userId

// Multi-tenant: tenantId
{ "id": "doc-1", "tenantId": "tenant-abc", ... }  // PK: /tenantId

// IoT: deviceId for telemetry
{ "id": "reading-1", "deviceId": "device-789", ... }  // PK: /deviceId

// Logs: synthetic key (date + category)
{ "id": "log-1", "partitionKey": "2024-01-15_errors", ... }  // PK: /partitionKey

Hierarchical Partition Keys

// For multi-level distribution (e.g., tenant → user)
// Container created with: /tenantId, /userId

{
  "id": "order-123",
  "tenantId": "acme-corp",
  "userId": "user-456",
  "items": [...]
}

// Query within tenant and user efficiently

Bad Partition Keys

// Avoid:
// - Low cardinality (status, type, boolean)
// - Monotonically increasing (timestamp, auto-increment)
// - Frequently updated fields
// - Fields not used in queries

// Bad: Only 3 values → hot partitions
{ "status": "pending" | "completed" | "cancelled" }

// Bad: All writes go to latest partition
{ "timestamp": "2024-01-15T10:30:00Z" }

SDK Setup (TypeScript)

Install

npm install @azure/cosmos

Initialize Client

// lib/cosmosdb.ts
import { CosmosClient, Database, Container } from '@azure/cosmos';

const endpoint = process.env.COSMOS_ENDPOINT!;
const key = process.env.COSMOS_KEY!;
const databaseId = process.env.COSMOS_DATABASE!;

const client = new CosmosClient({ endpoint, key });

// Or with connection string
// const client = new CosmosClient(process.env.COSMOS_CONNECTION_STRING!);

export const database: Database = client.database(databaseId);

export function getContainer(containerId: string): Container {
  return database.container(containerId);
}

Type Definitions

// types/cosmos.ts
export interface BaseItem {
  id: string;
  _ts?: number;      // Auto-generated timestamp
  _etag?: string;    // For optimistic concurrency
}

export interface User extends BaseItem {
  userId: string;    // Partition key
  email: string;
  name: string;
  createdAt: string;
  updatedAt: string;
}

export interface Order extends BaseItem {
  userId: string;    // Partition key
  orderId: string;
  items: OrderItem[];
  total: number;
  status: 'pending' | 'paid' | 'shipped' | 'delivered';
  createdAt: string;
}

export interface OrderItem {
  productId: string;
  name: string;
  quantity: number;
  price: number;
}

CRUD Operations

Create Item

import { getContainer } from './cosmosdb';
import { User } from './types';

const usersContainer = getContainer('users');

async function createUser(data: Omit<User, 'id' | 'createdAt' | 'updatedAt'>): Promise<User> {
  const now = new Date().toISOString();
  const user: User = {
    id: crypto.randomUUID(),
    ...data,
    createdAt: now,
    updatedAt: now
  };

  const { resource } = await usersContainer.items.create(user);
  return resource as User;
}

Read Item (Point Read)

// Most efficient read - requires id AND partition key
async function getUser(userId: string, id: string): Promise<User | null> {
  try {
    const { resource } = await usersContainer.item(id, userId).read<User>();
    return resource || null;
  } catch (error: any) {
    if (error.code === 404) return null;
    throw error;
  }
}

// If id equals partition key value
async function getUserById(userId: string): Promise<User | null> {
  try {
    const { resource } = await usersContainer.item(userId, userId).read<User>();
    return resource || null;
  } catch (error: any) {
    if (error.code === 404) return null;
    throw error;
  }
}

Query Items

// Query within partition (efficient)
async function getUserOrders(userId: string): Promise<Order[]> {
  const ordersContainer = getContainer('orders');

  const { resources } = await ordersContainer.items
    .query<Order>({
      query: 'SELECT * FROM c WHERE c.userId = @userId ORDER BY c.createdAt DESC',
      parameters: [{ name: '@userId', value: userId }]
    })
    .fetchAll();

  return resources;
}

// Cross-partition query (use sparingly)
async function getOrdersByStatus(status: string): Promise<Order[]> {
  const ordersContainer = getContainer('orders');

  const { resources } = await ordersContainer.items
    .query<Order>({
      query: 'SELECT * FROM c WHERE c.status = @status',
      parameters: [{ name: '@status', value: status }]
    })
    .fetchAll();

  return resources;
}

// Paginated query
async function getOrdersPaginated(
  userId: string,
  pageSize: number = 10,
  continuationToken?: string
): Promise<{ items: Order[]; continuationToken?: string }> {
  const ordersContainer = getContainer('orders');

  const queryIterator = ordersContainer.items.query<Order>(
    {
      query: 'SELECT * FROM c WHERE c.userId = @userId ORDER BY c.createdAt DESC',
      parameters: [{ name: '@userId', value: userId }]
    },
    {
      maxItemCount: pageSize,
      continuationToken
    }
  );

  const { resources, continuationToken: nextToken } = await queryIterator.fetchNext();

  return {
    items: resources,
    continuationToken: nextToken
  };
}

Update Item

// Replace entire item
async function updateUser(userId: string, id: string, updates: Partial<User>): Promise<User> {
  const existing = await getUser(userId, id);
  if (!existing) throw new Error('User not found');

  const updated: User = {
    ...existing,
    ...updates,
    updatedAt: new Date().toISOString()
  };

  const { resource } = await usersContainer.item(id, userId).replace(updated);
  return resource as User;
}

// Partial update (patch operations)
async function patchUser(userId: string, id: string, operations: any[]): Promise<User> {
  const { resource } = await usersContainer.item(id, userId).patch(operations);
  return resource as User;
}

// Usage:
await patchUser('user-123', 'user-123', [
  { op: 'set', path: '/name', value: 'New Name' },
  { op: 'set', path: '/updatedAt', value: new Date().toISOString() },
  { op: 'incr', path: '/loginCount', value: 1 }
]);

Delete Item

async function deleteUser(userId: string, id: string): Promise<void> {
  await usersContainer.item(id, userId).delete();
}

Optimistic Concurrency (ETags)

async function updateUserWithETag(
  userId: string,
  id: string,
  updates: Partial<User>,
  etag: string
): Promise<User> {
  const existing = await getUser(userId, id);
  if (!existing) throw new Error('User not found');

  const updated: User = {
    ...existing,
    ...updates,
    updatedAt: new Date().toISOString()
  };

  try {
    const { resource } = await usersContainer.item(id, userId).replace(updated, {
      accessCondition: { type: 'IfMatch', condition: etag }
    });
    return resource as User;
  } catch (error: any) {
    if (error.code === 412) {
      throw new Error('Document was modified by another process');
    }
    throw error;
  }
}

Consistency Levels

LevelGuaranteesLatencyUse Case
StrongLinearizable readsHighestFinancial, inventory
Bounded StalenessConsistent within boundsHighLeaderboards, counters
SessionRead your writesMediumUser sessions (default)
Consistent PrefixOrdered readsLowSocial feeds
EventualNo ordering guaranteeLowestAnalytics, logs

Set Consistency Per Request

// Override default consistency
const { resource } = await usersContainer.item(id, userId).read<User>({
  consistencyLevel: 'Strong'
});

// For queries
const { resources } = await container.items.query(
  { query: 'SELECT * FROM c' },
  { consistencyLevel: 'BoundedStaleness' }
).fetchAll();

Batch Operations

Transactional Batch (Same Partition)

async function createOrderWithItems(userId: string, order: Order, items: any[]): Promise<void> {
  const ordersContainer = getContainer('orders');

  const operations = [
    { operationType: 'Create' as const, resourceBody: order },
    ...items.map(item => ({
      operationType: 'Create' as const,
      resourceBody: { ...item, userId, orderId: order.orderId }
    }))
  ];

  const { result } = await ordersContainer.items.batch(operations, userId);

  // Check if any operation failed
  if (result.some(r => r.statusCode >= 400)) {
    throw new Error('Batch operation failed');
  }
}

Bulk Operations

// For large-scale imports (not transactional)
async function bulkImportUsers(users: User[]): Promise<void> {
  const operations = users.map(user => ({
    operationType: 'Create' as const,
    resourceBody: user,
 

---

*Content truncated.*

When not to use it

  • For relational data requiring strict ACID transactions
  • When storage needs are minimal and simple JSON storage suffices

Limitations

  • Incorrect partition key choice is difficult to reverse
  • Throughput settings directly impact monthly costs

How it compares

It focuses on the physical-to-logical partitioning mechanics unique to the Cosmos DB architecture.

Compared to similar skills

azure-cosmosdb side by side with the closest alternatives in the catalog.

SkillInstallsUpdatedSafetyDifficulty
azure-cosmosdb (this skill)14moReviewIntermediate
postgresql-table-design304moNo flagsIntermediate
agentdb-advanced-features79moReviewAdvanced
event-store-design52moNo flagsAdvanced

Try saying

Example prompts that trigger this skill in your AI assistant.

Search skills

Search the agent skills registry