update skills

This commit is contained in:
2026-03-17 16:53:22 -07:00
parent 0b0783ef8e
commit f9a530667e
389 changed files with 54512 additions and 1 deletions

View File

@@ -0,0 +1,185 @@
# Cloudflare Durable Objects
Expert guidance for building stateful applications with Cloudflare Durable Objects.
## Reading Order
1. **First time?** Read this overview + Quick Start
2. **Setting up?** See [Configuration](./configuration.md)
3. **Building features?** Use decision trees below → [Patterns](./patterns.md)
4. **Debugging issues?** Check [Gotchas](./gotchas.md)
5. **Deep dive?** [API](./api.md) and [DO Storage](../do-storage/README.md)
## Overview
Durable Objects combine compute with storage in globally-unique, strongly-consistent packages:
- **Globally unique instances**: Each DO has unique ID for multi-client coordination
- **Co-located storage**: Fast, strongly-consistent storage with compute
- **Automatic placement**: Objects spawn near first request location
- **Stateful serverless**: In-memory state + persistent storage
- **Single-threaded**: Serial request processing (no race conditions)
## Rules of Durable Objects
Critical rules preventing most production issues:
1. **One alarm per DO** - Schedule multiple events via queue pattern
2. **~1K req/s per DO max** - Shard for higher throughput
3. **Constructor runs every wake** - Keep initialization light; use lazy loading
4. **Hibernation clears memory** - In-memory state lost; persist critical data
5. **Use `ctx.waitUntil()` for cleanup** - Ensures completion after response sent
6. **No setTimeout for persistence** - Use `setAlarm()` for reliable scheduling
## Core Concepts
### Class Structure
All DOs extend `DurableObject` base class with constructor receiving `DurableObjectState` (storage, WebSockets, alarms) and `Env` (bindings).
### Lifecycle States
```
[Not Created] → [Active] ⇄ [Hibernated] → [Evicted]
[Destroyed]
```
- **Not Created**: DO ID exists but instance never spawned
- **Active**: Processing requests, in-memory state valid, billed per GB-hour
- **Hibernated**: WebSocket connections open but zero compute, zero cost
- **Evicted**: Removed from memory; next request triggers cold start
- **Destroyed**: Data deleted via migration or manual deletion
### Accessing from Workers
Workers use bindings to get stubs, then call RPC methods directly (recommended) or use fetch handler (legacy).
**RPC vs fetch() decision:**
```
├─ New project + compat ≥2024-04-03 → RPC (type-safe, simpler)
├─ Need HTTP semantics (headers, status) → fetch()
├─ Proxying requests to DO → fetch()
└─ Legacy compatibility → fetch()
```
See [Patterns: RPC vs fetch()](./patterns.md) for examples.
### ID Generation
- `idFromName()`: Deterministic, named coordination (rate limiting, locks)
- `newUniqueId()`: Random IDs for sharding high-throughput workloads
- `idFromString()`: Derive from existing IDs
- Jurisdiction option: Data locality compliance
### Storage Options
**Which storage API?**
```
├─ Structured data, relations, transactions → SQLite (recommended)
├─ Simple KV on SQLite DO → ctx.storage.kv (sync KV)
└─ Legacy KV-only DO → ctx.storage (async KV)
```
- **SQLite** (recommended): Structured data, transactions, 10GB/DO
- **Synchronous KV API**: Simple key-value on SQLite objects
- **Asynchronous KV API**: Legacy/advanced use cases
See [DO Storage](../do-storage/README.md) for deep dive.
### Special Features
- **Alarms**: Schedule future execution per-DO (1 per DO - use queue pattern for multiple)
- **WebSocket Hibernation**: Zero-cost idle connections (memory cleared on hibernation)
- **Point-in-Time Recovery**: Restore to any point in 30 days (SQLite only)
## Quick Start
```typescript
import { DurableObject } from "cloudflare:workers";
export class Counter extends DurableObject<Env> {
async increment(): Promise<number> {
const result = this.ctx.storage.sql.exec(
`INSERT INTO counters (id, value) VALUES (1, 1)
ON CONFLICT(id) DO UPDATE SET value = value + 1
RETURNING value`
).one();
return result.value;
}
}
// Worker access
export default {
async fetch(request: Request, env: Env): Promise<Response> {
const id = env.COUNTER.idFromName("global");
const stub = env.COUNTER.get(id);
const count = await stub.increment();
return new Response(`Count: ${count}`);
}
};
```
## Decision Trees
### What do you need?
```
├─ Coordinate requests (rate limit, lock, session)
│ → idFromName(identifier) → [Patterns: Rate Limiting/Locks](./patterns.md)
├─ High throughput (>1K req/s)
│ → Sharding with newUniqueId() or hash → [Patterns: Sharding](./patterns.md)
├─ Real-time updates (WebSocket, chat, collab)
│ → WebSocket hibernation + room pattern → [Patterns: Real-time](./patterns.md)
├─ Background work (cleanup, notifications, scheduled tasks)
│ → Alarms + queue pattern (1 alarm/DO) → [Patterns: Multiple Events](./patterns.md)
└─ User sessions with expiration
→ Session pattern + alarm cleanup → [Patterns: Session Management](./patterns.md)
```
### Which access pattern?
```
├─ New project + typed methods → RPC (compat ≥2024-04-03)
├─ Need HTTP semantics → fetch()
├─ Proxying to DO → fetch()
└─ Legacy compat → fetch()
```
See [Patterns: RPC vs fetch()](./patterns.md) for examples.
### Which storage?
```
├─ Structured data, SQL queries, transactions → SQLite (recommended)
├─ Simple KV on SQLite DO → ctx.storage.kv (sync API)
└─ Legacy KV-only DO → ctx.storage (async API)
```
See [DO Storage](../do-storage/README.md) for complete guide.
## Essential Commands
```bash
npx wrangler dev # Local dev with DOs
npx wrangler dev --remote # Test against prod DOs
npx wrangler deploy # Deploy + auto-apply migrations
```
## Resources
**Docs**: https://developers.cloudflare.com/durable-objects/
**API Reference**: https://developers.cloudflare.com/durable-objects/api/
**Examples**: https://developers.cloudflare.com/durable-objects/examples/
## In This Reference
- **[Configuration](./configuration.md)** - wrangler.jsonc setup, migrations, bindings, environments
- **[API](./api.md)** - Class structure, ctx methods, alarms, WebSocket hibernation
- **[Patterns](./patterns.md)** - Sharding, rate limiting, locks, real-time, sessions
- **[Gotchas](./gotchas.md)** - Limits, hibernation caveats, common errors
## See Also
- **[DO Storage](../do-storage/README.md)** - SQLite, KV, transactions (detailed storage guide)
- **[Workers](../workers/README.md)** - Core Workers runtime features
- **[WebSockets](../websockets/README.md)** - WebSocket APIs and patterns

View File

@@ -0,0 +1,187 @@
# Durable Objects API
## Class Structure
```typescript
import { DurableObject } from "cloudflare:workers";
export class MyDO extends DurableObject<Env> {
constructor(ctx: DurableObjectState, env: Env) {
super(ctx, env);
// Runs on EVERY wake - keep light!
}
// RPC methods (called directly from worker)
async myMethod(arg: string): Promise<string> { return arg; }
// fetch handler (legacy/HTTP semantics)
async fetch(req: Request): Promise<Response> { /* ... */ }
// Lifecycle handlers
async alarm() { /* alarm fired */ }
async webSocketMessage(ws: WebSocket, msg: string | ArrayBuffer) { /* ... */ }
async webSocketClose(ws: WebSocket, code: number, reason: string, wasClean: boolean) { /* ... */ }
async webSocketError(ws: WebSocket, error: unknown) { /* ... */ }
}
```
## DurableObjectState Context Methods
### Concurrency Control
```typescript
// Complete work after response sent (e.g., cleanup, logging)
this.ctx.waitUntil(promise: Promise<any>): void
// Critical section - blocks all other requests until complete
await this.ctx.blockConcurrencyWhile(async () => {
// No other requests processed during this block
// Use for initialization or critical operations
})
```
**When to use:**
- `waitUntil()`: Background cleanup, logging, non-critical work after response
- `blockConcurrencyWhile()`: First-time init, schema migration, critical state setup
### Lifecycle
```typescript
this.ctx.id // DurableObjectId of this instance
this.ctx.abort() // Force eviction (use after PITR restore to reload state)
```
### Storage Access
```typescript
this.ctx.storage.sql // SQLite API (recommended)
this.ctx.storage.kv // Sync KV API (SQLite DOs only)
this.ctx.storage // Async KV API (legacy/KV-only DOs)
```
See **[DO Storage](../do-storage/README.md)** for complete storage API reference.
### WebSocket Management
```typescript
this.ctx.acceptWebSocket(ws: WebSocket, tags?: string[]) // Enable hibernation
this.ctx.getWebSockets(tag?: string): WebSocket[] // Get by tag or all
this.ctx.getTags(ws: WebSocket): string[] // Get tags for connection
```
### Alarms
```typescript
await this.ctx.storage.setAlarm(timestamp: number | Date) // Schedule (overwrites existing)
await this.ctx.storage.getAlarm(): number | null // Get next alarm time
await this.ctx.storage.deleteAlarm(): void // Cancel alarm
```
**Limit:** 1 alarm per DO. Use queue pattern for multiple events (see [Patterns](./patterns.md)).
## Storage APIs
For detailed storage documentation including SQLite queries, KV operations, transactions, and Point-in-Time Recovery, see **[DO Storage](../do-storage/README.md)**.
Quick reference:
```typescript
// SQLite (recommended)
this.ctx.storage.sql.exec("SELECT * FROM users WHERE id = ?", userId).one()
// Sync KV (SQLite DOs only)
this.ctx.storage.kv.get("key")
// Async KV (legacy)
await this.ctx.storage.get("key")
```
## Alarms
Schedule future work that survives eviction:
```typescript
// Set alarm (overwrites any existing alarm)
await this.ctx.storage.setAlarm(Date.now() + 3600000) // 1 hour from now
await this.ctx.storage.setAlarm(new Date("2026-02-01")) // Absolute time
// Check next alarm
const nextRun = await this.ctx.storage.getAlarm() // null if none
// Cancel alarm
await this.ctx.storage.deleteAlarm()
// Handler called when alarm fires
async alarm() {
// Runs once alarm triggers
// DO wakes from hibernation if needed
// Use for cleanup, notifications, scheduled tasks
}
```
**Limitations:**
- 1 alarm per DO maximum
- Overwrites previous alarm when set
- Use queue pattern for multiple scheduled events (see [Patterns](./patterns.md))
**Reliability:**
- Alarms survive DO eviction/restart
- Cloudflare retries failed alarms automatically
- Not guaranteed exactly-once (handle idempotently)
## WebSocket Hibernation
Hibernation allows DOs with open WebSocket connections to consume zero compute/memory until message arrives.
```typescript
async fetch(req: Request): Promise<Response> {
const [client, server] = Object.values(new WebSocketPair());
this.ctx.acceptWebSocket(server, ["room:123"]); // Tags for filtering
server.serializeAttachment({ userId: "abc" }); // Persisted metadata
return new Response(null, { status: 101, webSocket: client });
}
// Called when message arrives (DO wakes from hibernation)
async webSocketMessage(ws: WebSocket, msg: string | ArrayBuffer) {
const data = ws.deserializeAttachment(); // Retrieve metadata
for (const c of this.ctx.getWebSockets("room:123")) c.send(msg);
}
// Called on close (optional handler)
async webSocketClose(ws: WebSocket, code: number, reason: string, wasClean: boolean) {
// Cleanup logic, remove from lists, etc.
}
// Called on error (optional handler)
async webSocketError(ws: WebSocket, error: unknown) {
console.error("WebSocket error:", error);
// Handle error, close connection, etc.
}
```
**Key concepts:**
- **Auto-hibernation:** DO hibernates when no active requests/alarms
- **Zero cost:** Hibernated DOs incur no charges while preserving connections
- **Memory cleared:** All in-memory state lost on hibernation
- **Attachment persistence:** Use `serializeAttachment()` for per-connection metadata that survives hibernation
- **Tags for filtering:** Group connections by room/channel/user for targeted broadcasts
**Handler lifecycle:**
- `webSocketMessage`: DO wakes, processes message, may hibernate after
- `webSocketClose`: Called when client closes (optional - implement for cleanup)
- `webSocketError`: Called on connection error (optional - implement for error handling)
**Metadata persistence:**
```typescript
// Store connection metadata (survives hibernation)
ws.serializeAttachment({ userId: "abc", room: "lobby" })
// Retrieve after hibernation
const { userId, room } = ws.deserializeAttachment()
```
## See Also
- **[DO Storage](../do-storage/README.md)** - Complete storage API reference
- **[Patterns](./patterns.md)** - Real-world usage patterns
- **[Gotchas](./gotchas.md)** - Hibernation caveats and limits

View File

@@ -0,0 +1,160 @@
# Durable Objects Configuration
## Basic Setup
```jsonc
{
"name": "my-worker",
"main": "src/index.ts",
"compatibility_date": "2025-01-01", // Use latest; ≥2024-04-03 for RPC
"durable_objects": {
"bindings": [
{
"name": "MY_DO", // Env binding name
"class_name": "MyDO" // Class exported from this worker
},
{
"name": "EXTERNAL", // Access DO from another worker
"class_name": "ExternalDO",
"script_name": "other-worker"
}
]
},
"migrations": [
{ "tag": "v1", "new_sqlite_classes": ["MyDO"] } // Prefer SQLite
]
}
```
## Binding Options
```jsonc
{
"name": "BINDING_NAME",
"class_name": "ClassName",
"script_name": "other-worker", // Optional: external DO
"environment": "production" // Optional: isolate by env
}
```
## Jurisdiction (Data Locality)
Specify jurisdiction at ID creation for data residency compliance:
```typescript
// EU data residency
const id = env.MY_DO.idFromName("user:123", { jurisdiction: "eu" })
// Available jurisdictions
const jurisdictions = ["eu", "fedramp"] // More may be added
// All operations on this DO stay within jurisdiction
const stub = env.MY_DO.get(id)
await stub.someMethod() // Data stays in EU
```
**Key points:**
- Set at ID creation time, immutable afterward
- DO instance physically located within jurisdiction
- Storage and compute guaranteed within boundary
- Use for GDPR, FedRAMP, other compliance requirements
- No cross-jurisdiction access (requests fail if DO in different jurisdiction)
## Migrations
```jsonc
{
"migrations": [
{ "tag": "v1", "new_sqlite_classes": ["MyDO"] }, // Create SQLite (recommended)
// { "tag": "v1", "new_classes": ["MyDO"] }, // Create KV (paid only)
{ "tag": "v2", "renamed_classes": [{ "from": "Old", "to": "New" }] },
{ "tag": "v3", "transferred_classes": [{ "from": "Src", "from_script": "old", "to": "Dest" }] },
{ "tag": "v4", "deleted_classes": ["Obsolete"] } // Destroys ALL data!
]
}
```
**Migration rules:**
- Tags must be unique and sequential (v1, v2, v3...)
- No rollback supported (test with `--dry-run` first)
- Auto-applied on deploy
- `new_sqlite_classes` recommended over `new_classes` (SQLite vs KV)
- `deleted_classes` immediately destroys ALL data (irreversible)
## Environment Isolation
Separate DO namespaces per environment (staging/production have distinct object instances):
```jsonc
{
"durable_objects": {
"bindings": [{ "name": "MY_DO", "class_name": "MyDO" }]
},
"env": {
"production": {
"durable_objects": {
"bindings": [
{ "name": "MY_DO", "class_name": "MyDO", "environment": "production" }
]
}
}
}
}
```
Deploy: `npx wrangler deploy --env production`
## Limits & Settings
```jsonc
{
"limits": {
"cpu_ms": 300000 // Max CPU time: 30s default, 300s max
}
}
```
See [Gotchas](./gotchas.md) for complete limits table.
## Types
```typescript
import { DurableObject } from "cloudflare:workers";
interface Env {
MY_DO: DurableObjectNamespace<MyDO>;
}
export class MyDO extends DurableObject<Env> {}
type DurableObjectNamespace<T> = {
newUniqueId(options?: { jurisdiction?: string }): DurableObjectId;
idFromName(name: string): DurableObjectId;
idFromString(id: string): DurableObjectId;
get(id: DurableObjectId): DurableObjectStub<T>;
};
```
## Commands
```bash
# Development
npx wrangler dev # Local dev
npx wrangler dev --remote # Test against production DOs
# Deployment
npx wrangler deploy # Deploy + auto-apply migrations
npx wrangler deploy --dry-run # Validate migrations without deploying
npx wrangler deploy --env production
# Management
npx wrangler durable-objects list # List namespaces
npx wrangler durable-objects info <namespace> <id> # Inspect specific DO
npx wrangler durable-objects delete <namespace> <id> # Delete DO (destroys data)
```
## See Also
- **[API](./api.md)** - DurableObjectState and lifecycle handlers
- **[Patterns](./patterns.md)** - Multi-environment patterns
- **[Gotchas](./gotchas.md)** - Migration caveats, limits

View File

@@ -0,0 +1,197 @@
# Durable Objects Gotchas
## Common Errors
### "Hibernation Cleared My In-Memory State"
**Problem:** Variables lost after hibernation
**Cause:** DO auto-hibernates when idle; in-memory state not persisted
**Solution:** Use `ctx.storage` for critical data, `ws.serializeAttachment()` for per-connection metadata
```typescript
// ❌ Wrong - lost on hibernation
private userCount = 0;
async webSocketMessage(ws: WebSocket, msg: string) {
this.userCount++; // Lost!
}
// ✅ Right - persisted
async webSocketMessage(ws: WebSocket, msg: string) {
const count = this.ctx.storage.kv.get("userCount") || 0;
this.ctx.storage.kv.put("userCount", count + 1);
}
```
### "setTimeout Didn't Fire After Restart"
**Problem:** Scheduled work lost on eviction
**Cause:** `setTimeout` in-memory only; eviction clears timers
**Solution:** Use `ctx.storage.setAlarm()` for reliable scheduling
```typescript
// ❌ Wrong - lost on eviction
setTimeout(() => this.cleanup(), 3600000);
// ✅ Right - survives eviction
await this.ctx.storage.setAlarm(Date.now() + 3600000);
async alarm() { await this.cleanup(); }
```
### "Constructor Runs on Every Wake"
**Problem:** Expensive init logic slows all requests
**Cause:** Constructor runs on every wake (first request after eviction OR after hibernation)
**Solution:** Lazy initialization or cache in storage
**Critical understanding:** Constructor runs in two scenarios:
1. **Cold start** - DO evicted from memory, first request creates new instance
2. **Wake from hibernation** - DO with WebSockets hibernated, message/alarm wakes it
```typescript
// ❌ Wrong - expensive on every wake
constructor(ctx: DurableObjectState, env: Env) {
super(ctx, env);
this.heavyData = this.loadExpensiveData(); // Slow!
}
// ✅ Right - lazy load
private heavyData?: HeavyData;
private getHeavyData() {
if (!this.heavyData) this.heavyData = this.loadExpensiveData();
return this.heavyData;
}
```
### "Durable Object Overloaded (503 errors)"
**Problem:** 503 errors under load
**Cause:** Single DO exceeding ~1K req/s throughput limit
**Solution:** Shard across multiple DOs (see [Patterns: Sharding](./patterns.md))
### "Storage Quota Exceeded (Write failures)"
**Problem:** Write operations failing
**Cause:** DO storage exceeding 10GB limit or account quota
**Solution:** Cleanup with alarms, use `deleteAll()` for old data, upgrade plan
### "CPU Time Exceeded (Terminated)"
**Problem:** Request terminated mid-execution
**Cause:** Processing exceeding 30s CPU time default limit
**Solution:** Increase `limits.cpu_ms` in wrangler.jsonc (max 300s) or chunk work
### "WebSockets Disconnect on Eviction"
**Problem:** Connections drop unexpectedly
**Cause:** DO evicted from memory without hibernation API
**Solution:** Use WebSocket hibernation handlers + client reconnection logic
### "Migration Failed (Deploy error)"
**Cause:** Non-unique tags, non-sequential tags, or invalid class names in migration
**Solution:** Check tag uniqueness/sequential ordering and verify class names are correct
### "RPC Method Not Found"
**Cause:** compatibility_date < 2024-04-03 preventing RPC usage
**Solution:** Update compatibility_date to >= 2024-04-03 or use fetch() instead of RPC
### "Only One Alarm Allowed"
**Cause:** Need multiple scheduled tasks but only one alarm supported per DO
**Solution:** Use event queue pattern to schedule multiple tasks with single alarm
### "Race Condition Despite Single-Threading"
**Problem:** Concurrent requests see inconsistent state
**Cause:** Async operations allow request interleaving (await = yield point)
**Solution:** Use `blockConcurrencyWhile()` for critical sections or atomic storage ops
```typescript
// ❌ Wrong - race condition
async incrementCounter() {
const count = await this.ctx.storage.get("count") || 0;
// ⚠️ Another request could execute here during await
await this.ctx.storage.put("count", count + 1);
}
// ✅ Right - atomic operation
async incrementCounter() {
return this.ctx.storage.sql.exec(
"INSERT INTO counters (id, value) VALUES (1, 1) ON CONFLICT(id) DO UPDATE SET value = value + 1 RETURNING value"
).one().value;
}
// ✅ Right - explicit locking
async criticalOperation() {
await this.ctx.blockConcurrencyWhile(async () => {
const count = await this.ctx.storage.get("count") || 0;
await this.ctx.storage.put("count", count + 1);
});
}
```
### "Migration Rollback Not Supported"
**Cause:** Attempting to rollback a migration after deployment
**Solution:** Test with `--dry-run` before deploying; migrations cannot be rolled back
### "deleted_classes Destroys Data"
**Problem:** Migration deleted all data
**Cause:** `deleted_classes` migration immediately destroys all DO instances and data
**Solution:** Test with `--dry-run`; use `transferred_classes` to preserve data during moves
### "Cold Starts Are Slow"
**Problem:** First request after eviction takes longer
**Cause:** DO constructor + initial storage access on cold start
**Solution:** Expected behavior; optimize constructor, use connection pooling in clients, consider warming strategy for critical DOs
```typescript
// Warming strategy (periodically ping critical DOs)
export default {
async scheduled(event: ScheduledEvent, env: Env) {
const criticalIds = ["auth", "sessions", "locks"];
await Promise.all(criticalIds.map(name => {
const id = env.MY_DO.idFromName(name);
const stub = env.MY_DO.get(id);
return stub.ping(); // Keep warm
}));
}
};
```
## Limits
| Limit | Free | Paid | Notes |
|-------|------|------|-------|
| SQLite storage per DO | 10 GB | 10 GB | Per Durable Object instance |
| SQLite total storage | 5 GB | Unlimited | Account-wide quota |
| Key+value size | 2 MB | 2 MB | Single KV pair (SQLite/async) |
| CPU time default | 30s | 30s | Per request; configurable |
| CPU time max | 300s | 300s | Set via `limits.cpu_ms` |
| DO classes | 100 | 500 | Distinct DO class definitions |
| SQL columns | 100 | 100 | Per table |
| SQL statement size | 100 KB | 100 KB | Max SQL query size |
| WebSocket message size | 32 MiB | 32 MiB | Per message |
| Request throughput | ~1K req/s | ~1K req/s | Per DO (soft limit - shard for more) |
| Alarms per DO | 1 | 1 | Use queue pattern for multiple events |
| Total DOs | Unlimited | Unlimited | Create as many instances as needed |
| WebSockets | Unlimited | Unlimited | Within 128MB memory limit per DO |
| Memory per DO | 128 MB | 128 MB | In-memory state + WebSocket buffers |
## Hibernation Caveats
1. **Memory cleared** - All in-memory variables lost; reconstruct from storage or `deserializeAttachment()`
2. **Constructor reruns** - Runs on wake; avoid expensive operations, use lazy initialization
3. **No guarantees** - DO may evict instead of hibernate; design for both
4. **Attachment limit** - `serializeAttachment()` data must be JSON-serializable, keep small
5. **Alarm wakes DO** - Alarm prevents hibernation until handler completes
6. **WebSocket state not automatic** - Must explicitly persist with `serializeAttachment()` or storage
## See Also
- **[Patterns](./patterns.md)** - Workarounds for common limitations
- **[API](./api.md)** - Storage limits and quotas
- **[Configuration](./configuration.md)** - Setting CPU limits

View File

@@ -0,0 +1,201 @@
# Durable Objects Patterns
## When to Use Which Pattern
| Need | Pattern | ID Strategy |
|------|---------|-------------|
| Rate limit per user/IP | Rate Limiting | `idFromName(identifier)` |
| Mutual exclusion | Distributed Lock | `idFromName(resource)` |
| >1K req/s throughput | Sharding | `newUniqueId()` or hash |
| Real-time updates | WebSocket Collab | `idFromName(room)` |
| User sessions | Session Management | `idFromName(sessionId)` |
| Background cleanup | Alarm-based | Any |
## RPC vs fetch()
**RPC** (compat ≥2024-04-03): Type-safe, simpler, default for new projects
**fetch()**: Legacy compat, HTTP semantics, proxying
```typescript
const count = await stub.increment(); // RPC
const count = await (await stub.fetch(req)).json(); // fetch()
```
## Sharding (High Throughput)
Single DO ~1K req/s max. Shard for higher throughput:
```typescript
export default {
async fetch(req: Request, env: Env): Promise<Response> {
const userId = new URL(req.url).searchParams.get("user");
const hash = hashCode(userId) % 100; // 100 shards
const id = env.COUNTER.idFromName(`shard:${hash}`);
return env.COUNTER.get(id).fetch(req);
}
};
function hashCode(str: string): number {
let hash = 0;
for (let i = 0; i < str.length; i++) hash = ((hash << 5) - hash) + str.charCodeAt(i);
return Math.abs(hash);
}
```
**Decisions:**
- **Shard count**: 10-1000 typical (start with 100, measure, adjust)
- **Shard key**: User ID, IP, session - must distribute evenly (use hash)
- **Aggregation**: Coordinator DO or external system (D1, R2)
## Rate Limiting
```typescript
async checkLimit(key: string, limit: number, windowMs: number): Promise<boolean> {
const req = this.ctx.storage.sql.exec("SELECT COUNT(*) as count FROM requests WHERE key = ? AND timestamp > ?", key, Date.now() - windowMs).one();
if (req.count >= limit) return false;
this.ctx.storage.sql.exec("INSERT INTO requests (key, timestamp) VALUES (?, ?)", key, Date.now());
return true;
}
```
## Distributed Lock
```typescript
private held = false;
async acquire(timeoutMs = 5000): Promise<boolean> {
if (this.held) return false;
this.held = true;
await this.ctx.storage.setAlarm(Date.now() + timeoutMs);
return true;
}
async release() { this.held = false; await this.ctx.storage.deleteAlarm(); }
async alarm() { this.held = false; } // Auto-release on timeout
```
## Hibernation-Aware Pattern
Preserve state across hibernation:
```typescript
async fetch(req: Request): Promise<Response> {
const [client, server] = Object.values(new WebSocketPair());
const userId = new URL(req.url).searchParams.get("user");
server.serializeAttachment({ userId }); // Survives hibernation
this.ctx.acceptWebSocket(server, ["room:lobby"]);
server.send(JSON.stringify({ type: "init", state: this.ctx.storage.kv.get("state") }));
return new Response(null, { status: 101, webSocket: client });
}
async webSocketMessage(ws: WebSocket, msg: string) {
const { userId } = ws.deserializeAttachment(); // Retrieve after wake
const state = this.ctx.storage.kv.get("state") || {};
state[userId] = JSON.parse(msg);
this.ctx.storage.kv.put("state", state);
for (const c of this.ctx.getWebSockets("room:lobby")) c.send(msg);
}
```
## Real-time Collaboration
Broadcast updates to all connected clients:
```typescript
async webSocketMessage(ws: WebSocket, msg: string) {
const data = JSON.parse(msg);
this.ctx.storage.kv.put("doc", data.content); // Persist
for (const c of this.ctx.getWebSockets()) if (c !== ws) c.send(msg); // Broadcast
}
```
### WebSocket Reconnection
**Client-side** (exponential backoff):
```typescript
class ResilientWS {
private delay = 1000;
connect(url: string) {
const ws = new WebSocket(url);
ws.onclose = () => setTimeout(() => {
this.connect(url);
this.delay = Math.min(this.delay * 2, 30000);
}, this.delay);
}
}
```
**Server-side** (cleanup on close):
```typescript
async webSocketClose(ws: WebSocket, code: number, reason: string, wasClean: boolean) {
const { userId } = ws.deserializeAttachment();
this.ctx.storage.sql.exec("UPDATE users SET online = false WHERE id = ?", userId);
for (const c of this.ctx.getWebSockets()) c.send(JSON.stringify({ type: "user_left", userId }));
}
```
## Session Management
```typescript
async createSession(userId: string, data: object): Promise<string> {
const id = crypto.randomUUID(), exp = Date.now() + 86400000;
this.ctx.storage.sql.exec("INSERT INTO sessions VALUES (?, ?, ?, ?)", id, userId, JSON.stringify(data), exp);
await this.ctx.storage.setAlarm(exp);
return id;
}
async getSession(id: string): Promise<object | null> {
const row = this.ctx.storage.sql.exec("SELECT data FROM sessions WHERE id = ? AND expires_at > ?", id, Date.now()).one();
return row ? JSON.parse(row.data) : null;
}
async alarm() { this.ctx.storage.sql.exec("DELETE FROM sessions WHERE expires_at <= ?", Date.now()); }
```
## Multiple Events (Single Alarm)
Queue pattern to schedule multiple events:
```typescript
async scheduleEvent(id: string, runAt: number) {
await this.ctx.storage.put(`event:${id}`, { id, runAt });
const curr = await this.ctx.storage.getAlarm();
if (!curr || runAt < curr) await this.ctx.storage.setAlarm(runAt);
}
async alarm() {
const events = await this.ctx.storage.list({ prefix: "event:" }), now = Date.now();
let next = null;
for (const [key, ev] of events) {
if (ev.runAt <= now) {
await this.processEvent(ev);
await this.ctx.storage.delete(key);
} else if (!next || ev.runAt < next) next = ev.runAt;
}
if (next) await this.ctx.storage.setAlarm(next);
}
```
## Graceful Cleanup
Use `ctx.waitUntil()` to complete work after response:
```typescript
async myMethod() {
const response = { success: true };
this.ctx.waitUntil(this.ctx.storage.sql.exec("DELETE FROM old_data WHERE timestamp < ?", cutoff));
return response;
}
```
## Best Practices
- **Design**: Use `idFromName()` for coordination, `newUniqueId()` for sharding, minimize constructor work
- **Storage**: Prefer SQLite, batch with transactions, set alarms for cleanup, use PITR before risky ops
- **Performance**: ~1K req/s per DO max - shard for more, cache in memory, use alarms for deferred work
- **Reliability**: Handle 503 with retry+backoff, design for cold starts, test migrations with `--dry-run`
- **Security**: Validate inputs in Workers, rate limit DO creation, use jurisdiction for compliance
## See Also
- **[API](./api.md)** - ctx methods, WebSocket handlers
- **[Gotchas](./gotchas.md)** - Hibernation caveats, common errors
- **[DO Storage](../do-storage/README.md)** - Storage patterns and transactions