opencode-swarm-plugin 0.44.0 → 0.44.2

package/docs/planning/ADR-003-performance-improvements.md

@@ -1,451 +0,0 @@
-# ADR-003: Performance Improvements via Live Queries and Batching
-
-## Status
-
-Proposed
-
-## Context
-
-Current Swarm Mail implementation uses polling for real-time updates, which is inefficient:
-
-- `getInbox()` polls every 500ms to detect new messages
-- File reservation conflicts checked via repeated queries
-- High CPU usage in background polling loops
-- Latency: 250-500ms average between event and notification
-
-Research shows PGLite provides two superior alternatives to polling:
-
-1. **LISTEN/NOTIFY** - PostgreSQL pub/sub via `pg.listen()` and `pg.unlisten()`
-2. **Live Queries** - Incremental query results via `live.incrementalQuery()` (RECOMMENDED)
-
-Additionally, batch operations can reduce transaction overhead:
-
-- Current: Individual transactions per message (N transactions)
-- Improved: Batched inserts via `.transaction()` or `.exec()` (1 transaction)
-
-## Decision
-
-### Replace Polling with Live Queries
-
-**Why Live Queries over LISTEN/NOTIFY:**
-
-- Simpler API - returns incremental result sets automatically
-- No manual query re-execution needed
-- Built-in change detection
-- Works with existing SQL queries
-
-**Implementation:**
-
-```typescript
-// Before (polling):
-async function watchInbox(
-  agentName: string,
-  callback: (messages: Message[]) => void,
-) {
-  const interval = setInterval(async () => {
-    const messages = await getInbox(agentName, { limit: 5 });
-    callback(messages);
-  }, 500);
-  return () => clearInterval(interval);
-}
-
-// After (live query):
-import { live } from "@electric-sql/pglite/live";
-
-async function watchInbox(
-  agentName: string,
-  callback: (messages: Message[]) => void,
-) {
-  const liveQuery = await db.live.incrementalQuery(
-    `SELECT id, sender, subject, importance, received_at 
-     FROM messages 
-     WHERE recipient = $1 AND read_at IS NULL
-     ORDER BY importance DESC, received_at DESC
-     LIMIT 5`,
-    [agentName],
-  );
-
-  liveQuery.subscribe(({ rows }) => {
-    callback(rows as Message[]);
-  });
-
-  return () => liveQuery.unsubscribe();
-}
-```
-
-**Projection Updates:**
-
-```typescript
-// Watch for new events in real-time
-const liveEvents = await db.live.incrementalQuery(
-  `SELECT * FROM events WHERE sequence > $1 ORDER BY sequence ASC`,
-  [lastProcessedSequence],
-);
-
-liveEvents.subscribe(({ rows }) => {
-  rows.forEach((event) => updateProjection(event));
-});
-```
-
-### Batch Operations for Multi-Insert
-
-**Use `.transaction()` for Multiple Related Operations:**
-
-```typescript
-// Before: N transactions
-for (const message of messages) {
-  await sendMessage(message); // 1 transaction each
-}
-
-// After: 1 transaction
-await db.transaction(async (tx) => {
-  for (const message of messages) {
-    await tx.query(
-      `INSERT INTO messages (sender, recipient, subject, body) VALUES ($1, $2, $3, $4)`,
-      [message.sender, message.recipient, message.subject, message.body],
-    );
-  }
-});
-```
-
-**Use `.exec()` for Batch SQL:**
-
-```typescript
-// Atomic multi-statement execution
-await db.exec(`
-  BEGIN;
-  INSERT INTO messages (sender, recipient, subject) VALUES ('Alice', 'Bob', 'Test 1');
-  INSERT INTO messages (sender, recipient, subject) VALUES ('Alice', 'Bob', 'Test 2');
-  INSERT INTO messages (sender, recipient, subject) VALUES ('Alice', 'Bob', 'Test 3');
-  COMMIT;
-`);
-```
-
-### Connection Pooling Decision
-
-**NOT NEEDED for PGLite:**
-
-- PGLite runs in-process (WASM embedded database)
-- Single-user design (no concurrent connections)
-- Connection pooling is for client-server PostgreSQL
-- Would add unnecessary complexity
-
-## Consequences
-
-### Easier
-
-- **Real-time updates** - Sub-millisecond latency for new messages/events
-- **Lower CPU usage** - No background polling loops
-- **Fewer queries** - Live queries push changes, don't pull
-- **Atomic batches** - Multi-message sends in 1 transaction
-- **Simplified code** - No interval management, cleanup handled by unsubscribe
-
-### More Difficult
-
-- **Subscription management** - Must track and unsubscribe live queries
-- **Memory usage** - Live queries hold result sets in memory
-- **Testing complexity** - Async subscriptions harder to test than sync polls
-- **Debugging** - Push-based updates less visible in logs
-
-### Performance Impact (Estimated)
-
-| Metric               | Before (Polling) | After (Live Queries)    | Improvement     |
-| -------------------- | ---------------- | ----------------------- | --------------- |
-| Notification latency | 250-500ms        | <10ms                   | 25-50x faster   |
-| CPU usage (idle)     | 5-10% (polling)  | <1% (event-driven)      | 5-10x reduction |
-| Queries per second   | 2-4 (polling)    | 0 (push)                | Eliminated      |
-| Transaction overhead | N (individual)   | 1 (batched)             | N speedup       |
-| Memory usage         | Low              | Medium (result caching) | +10-20%         |
-
-### Risks & Mitigations
-
-| Risk                              | Impact | Mitigation                                          |
-| --------------------------------- | ------ | --------------------------------------------------- |
-| Memory leaks from subscriptions   | High   | Enforce unsubscribe in cleanup, add timeout guards  |
-| Live queries fail on syntax error | High   | Validate SQL in tests, fallback to polling on error |
-| Large result sets in memory       | Medium | Hard limit on LIMIT clause (max 100 rows)           |
-| Subscription overhead at scale    | Medium | Pool subscriptions, deduplicate queries             |
-| Debugging push updates            | Low    | Add subscription logging, event stream tracing      |
-
-## Implementation Notes
-
-### Phase 1: Live Query Infrastructure (Week 1)
-
-**1.1 Create Live Query Wrapper**
-
-```typescript
-// src/streams/live-query.ts
-import { live } from "@electric-sql/pglite/live";
-import type { PGlite } from "@electric-sql/pglite";
-
-interface LiveQueryOptions<T> {
-  db: PGlite;
-  query: string;
-  params: unknown[];
-  onUpdate: (rows: T[]) => void;
-  onError?: (error: Error) => void;
-}
-
-export async function createLiveQuery<T>(options: LiveQueryOptions<T>) {
-  const { db, query, params, onUpdate, onError } = options;
-
-  try {
-    const liveQuery = await db.live.incrementalQuery(query, params);
-
-    const unsubscribe = liveQuery.subscribe({
-      next: ({ rows }) => onUpdate(rows as T[]),
-      error: onError || ((err) => console.error("Live query error:", err)),
-    });
-
-    return {
-      unsubscribe,
-      refresh: () => liveQuery.refresh(),
-    };
-  } catch (error) {
-    if (onError) onError(error as Error);
-    throw error;
-  }
-}
-```
-
-**1.2 Add Cleanup Tracking**
-
-```typescript
-// Track active subscriptions for cleanup
-const activeSubscriptions = new Set<() => void>();
-
-export function registerSubscription(unsubscribe: () => void) {
-  activeSubscriptions.add(unsubscribe);
-}
-
-export function cleanupAllSubscriptions() {
-  activeSubscriptions.forEach((unsub) => unsub());
-  activeSubscriptions.clear();
-}
-
-// Call on shutdown
-process.on("SIGTERM", cleanupAllSubscriptions);
-```
-
-### Phase 2: Replace Polling (Week 2)
-
-**2.1 Inbox Watching**
-
-```typescript
-// src/agent-mail.ts
-export async function watchInbox(
-  db: PGlite,
-  agentName: string,
-  callback: (messages: Message[]) => void,
-) {
-  const { unsubscribe } = await createLiveQuery<Message>({
-    db,
-    query: `
-      SELECT id, sender, subject, importance, received_at 
-      FROM messages 
-      WHERE recipient = $1 AND read_at IS NULL
-      ORDER BY importance DESC, received_at DESC
-      LIMIT 5
-    `,
-    params: [agentName],
-    onUpdate: callback,
-  });
-
-  registerSubscription(unsubscribe);
-  return unsubscribe;
-}
-```
-
-**2.2 Event Stream Watching**
-
-```typescript
-// src/streams/projections.ts
-export async function watchEvents(
-  db: PGlite,
-  fromSequence: number,
-  callback: (events: SwarmMailEvent[]) => void,
-) {
-  const { unsubscribe } = await createLiveQuery<SwarmMailEvent>({
-    db,
-    query: `SELECT * FROM events WHERE sequence > $1 ORDER BY sequence ASC`,
-    params: [fromSequence],
-    onUpdate: callback,
-  });
-
-  registerSubscription(unsubscribe);
-  return unsubscribe;
-}
-```
-
-**2.3 File Reservation Conflicts**
-
-```typescript
-// Watch for conflicts with my reservations
-export async function watchReservationConflicts(
-  db: PGlite,
-  myReservations: string[],
-  callback: (conflicts: Conflict[]) => void,
-) {
-  const { unsubscribe } = await createLiveQuery<Conflict>({
-    db,
-    query: `
-      SELECT r.* FROM file_reservations r
-      WHERE r.path_pattern = ANY($1)
-      AND r.agent_name != $2
-      AND r.expires_at > NOW()
-    `,
-    params: [myReservations, currentAgent],
-    onUpdate: callback,
-  });
-
-  return unsubscribe;
-}
-```
-
-### Phase 3: Batch Operations (Week 3)
-
-**3.1 Batch Message Send**
-
-```typescript
-export async function sendMessages(db: PGlite, messages: MessageDraft[]) {
-  await db.transaction(async (tx) => {
-    for (const msg of messages) {
-      await tx.query(
-        `INSERT INTO messages (sender, recipient, subject, body, thread_id, importance) 
-         VALUES ($1, $2, $3, $4, $5, $6)`,
-        [
-          msg.sender,
-          msg.recipient,
-          msg.subject,
-          msg.body,
-          msg.thread_id,
-          msg.importance,
-        ],
-      );
-    }
-  });
-}
-```
-
-**3.2 Batch Event Append**
-
-```typescript
-export async function appendEvents(db: PGlite, events: SwarmMailEvent[]) {
-  await db.transaction(async (tx) => {
-    for (const event of events) {
-      await tx.query(
-        `INSERT INTO events (type, data, project_key, timestamp) 
-         VALUES ($1, $2, $3, $4)`,
-        [
-          event.type,
-          JSON.stringify(event.data),
-          event.projectKey,
-          event.timestamp,
-        ],
-      );
-    }
-  });
-}
-```
-
-### Phase 4: Testing & Benchmarks (Week 4)
-
-**4.1 Integration Tests**
-
-```typescript
-import { describe, it, expect, beforeEach, afterEach } from "vitest";
-
-describe("Live Queries", () => {
-  let unsubscribe: () => void;
-
-  afterEach(() => {
-    unsubscribe?.();
-  });
-
-  it("notifies on new message", async () => {
-    const updates: Message[] = [];
-
-    unsubscribe = await watchInbox(db, "Alice", (messages) => {
-      updates.push(...messages);
-    });
-
-    await sendMessage(db, { to: "Alice", subject: "Test", body: "Hello" });
-
-    // Wait for async update
-    await new Promise((resolve) => setTimeout(resolve, 100));
-
-    expect(updates).toHaveLength(1);
-    expect(updates[0].subject).toBe("Test");
-  });
-
-  it("cleans up on unsubscribe", async () => {
-    const updateCount = { count: 0 };
-
-    unsubscribe = await watchInbox(db, "Alice", () => {
-      updateCount.count++;
-    });
-
-    unsubscribe();
-
-    await sendMessage(db, {
-      to: "Alice",
-      subject: "After unsubscribe",
-      body: "Test",
-    });
-    await new Promise((resolve) => setTimeout(resolve, 100));
-
-    expect(updateCount.count).toBe(0); // Should not increment after unsubscribe
-  });
-});
-```
-
-**4.2 Performance Benchmarks**
-
-```typescript
-import { bench, describe } from "vitest";
-
-describe("Batch vs Individual Inserts", () => {
-  bench("Individual inserts (N transactions)", async () => {
-    for (let i = 0; i < 100; i++) {
-      await sendMessage(db, { to: "Bob", subject: `Msg ${i}`, body: "Test" });
-    }
-  });
-
-  bench("Batch insert (1 transaction)", async () => {
-    const messages = Array.from({ length: 100 }, (_, i) => ({
-      to: "Bob",
-      subject: `Msg ${i}`,
-      body: "Test",
-    }));
-    await sendMessages(db, messages);
-  });
-});
-```
-
-### Success Criteria
-
-- [ ] All polling loops removed from codebase
-- [ ] Live queries handle 100+ concurrent watchers without leaks
-- [ ] Batch operations 10x faster than individual inserts (benchmark)
-- [ ] Notification latency <50ms for new messages (99th percentile)
-- [ ] CPU usage <1% in idle state (no polling)
-- [ ] Memory usage increase <20% compared to polling
-- [ ] Integration tests pass with live queries
-- [ ] Cleanup functions properly unsubscribe all watchers
-
-### Migration Path
-
-1. Add live query infrastructure (non-breaking)
-2. Feature flag to toggle live queries vs polling (`ENABLE_LIVE_QUERIES=true`)
-3. Run both in parallel for 1 week, monitor metrics
-4. Default to live queries if metrics are better
-5. Remove polling code after 2 weeks of stable live queries
-6. Add batch operations (non-breaking performance improvement)
-
-### Fallback Strategy
-
-If live queries prove unstable:
-
-- Keep polling as fallback (`ENABLE_LIVE_QUERIES=false`)
-- Add exponential backoff to polling (reduce CPU)
-- Consider hybrid: live queries for critical paths, polling for non-critical

package/docs/planning/ADR-004-message-queue-features.md

@@ -1,187 +0,0 @@
-# ADR-004: Advanced Message Queue Features
-
-## Status
-
-Proposed
-
-## Context
-
-Current Swarm Mail message queue is basic:
-
-- FIFO message delivery
-- No priority ordering
-- No retry/dead-letter handling
-- No TTL/expiration
-- No pub/sub or saga patterns
-
-Research shows PGLite supports all enterprise message queue patterns:
-
-- Priority queues via `ORDER BY priority DESC`
-- Dead Letter Queues (DLQ) with retry tracking
-- Time-To-Live (TTL) with expires_at timestamps
-- Pub/sub via topic filtering
-- Saga orchestration with compensation
-
-## Decision
-
-Add enterprise message queue features using standard SQL patterns.
-
-### 1. Priority Queues
-
-```sql
-ALTER TABLE messages ADD COLUMN priority INTEGER DEFAULT 1;
-CREATE INDEX idx_messages_priority ON messages(recipient, priority DESC, received_at DESC);
-
--- Fetch highest priority first
-SELECT * FROM messages
-WHERE recipient = $1 AND read_at IS NULL
-ORDER BY priority DESC, received_at DESC
-LIMIT 5;
-```
-
-Priority levels:
-
-- 0 = urgent (blocks, errors)
-- 1 = high (task completion, conflicts)
-- 2 = normal (progress updates)
-- 3 = low (info, FYI)
-
-### 2. Dead Letter Queue (DLQ)
-
-```sql
-CREATE TABLE failed_messages (
-  id SERIAL PRIMARY KEY,
-  original_message_id INTEGER REFERENCES messages(id),
-  retry_count INTEGER NOT NULL,
-  failure_reason TEXT NOT NULL,
-  failed_at TIMESTAMP NOT NULL DEFAULT NOW(),
-  will_retry_at TIMESTAMP,  -- NULL = no retry
-  metadata JSONB
-);
-
--- Move to DLQ after 3 retries
-INSERT INTO failed_messages (original_message_id, retry_count, failure_reason)
-VALUES ($1, $2, $3);
-```
-
-### 3. Time-To-Live (TTL)
-
-```sql
-ALTER TABLE messages ADD COLUMN expires_at TIMESTAMP;
-
--- Background cleanup job
-DELETE FROM messages
-WHERE expires_at IS NOT NULL AND expires_at < NOW() AND read_at IS NULL;
-
--- Fetch only non-expired messages
-SELECT * FROM messages
-WHERE recipient = $1 AND read_at IS NULL
-  AND (expires_at IS NULL OR expires_at > NOW())
-ORDER BY priority DESC, received_at DESC;
-```
-
-### 4. Pub/Sub Topics
-
-```sql
-ALTER TABLE messages ADD COLUMN topic VARCHAR(255);
-CREATE INDEX idx_messages_topic ON messages(topic, received_at DESC);
-
--- Subscribe to topic pattern
-SELECT * FROM messages
-WHERE topic LIKE 'builds.%' AND received_at > $1
-ORDER BY received_at ASC;
-
--- Wildcard subscriptions
-WHERE topic LIKE 'agent.%.error'  -- All agent errors
-WHERE topic LIKE 'bead.%.status'  -- All bead status updates
-```
-
-### 5. Saga Orchestration Pattern
-
-```sql
-CREATE TABLE saga_instances (
-  id SERIAL PRIMARY KEY,
-  saga_id VARCHAR(255) UNIQUE NOT NULL,  -- e.g., "epic-bd-123"
-  coordinator VARCHAR(255) NOT NULL,
-  status VARCHAR(50) NOT NULL,  -- pending, in_progress, completed, failed, compensating
-  data JSONB NOT NULL,
-  created_at TIMESTAMP NOT NULL DEFAULT NOW(),
-  updated_at TIMESTAMP NOT NULL DEFAULT NOW()
-);
-
-CREATE TABLE saga_steps (
-  id SERIAL PRIMARY KEY,
-  saga_id VARCHAR(255) REFERENCES saga_instances(saga_id),
-  step_name VARCHAR(255) NOT NULL,
-  agent VARCHAR(255) NOT NULL,
-  status VARCHAR(50) NOT NULL,  -- pending, completed, failed, compensated
-  compensation_for INTEGER REFERENCES saga_steps(id),  -- Undo for this step
-  data JSONB,
-  completed_at TIMESTAMP
-);
-```
-
-Saga workflow:
-
-1. Coordinator creates saga_instance
-2. Sends step messages to worker agents
-3. Workers complete steps, update saga_steps
-4. On failure, coordinator sends compensation messages
-5. Workers execute compensation (undo) logic
-
-## Consequences
-
-### Easier
-
-- **Priority handling** - Critical messages processed first
-- **Resilience** - Failed messages retry or move to DLQ
-- **TTL enforcement** - Stale messages auto-expire
-- **Event broadcasting** - Pub/sub topics for loosely coupled agents
-- **Long-running workflows** - Saga pattern for multi-agent coordination
-
-### More Difficult
-
-- **Schema complexity** - More tables, indexes, constraints
-- **Background jobs** - TTL cleanup requires cron/scheduler
-- **Testing** - More failure modes to test (retries, DLQ, compensations)
-- **Debugging** - Saga state tracking adds visibility requirements
-
-## Implementation Notes
-
-### Phase 1: Priority Queues
-
-- Add priority column with default=1 (existing messages unaffected)
-- Update getInbox() to ORDER BY priority DESC
-- Add priority parameter to sendMessage()
-
-### Phase 2: DLQ + Retries
-
-- Create failed_messages table
-- Add retry logic to message processing
-- Exponential backoff: 1min, 5min, 30min, DLQ
-
-### Phase 3: TTL
-
-- Add expires_at column
-- Background cleanup job (run every 5 minutes)
-- Filter expired messages in queries
-
-### Phase 4: Pub/Sub
-
-- Add topic column
-- Update sendMessage() to accept topic
-- Add subscribeToTopic() using live queries
-
-### Phase 5: Sagas (Future)
-
-- Create saga tables
-- Add saga coordinator logic
-- Implement compensation pattern
-
-### Success Criteria
-
-- [ ] Priority messages processed before normal
-- [ ] Failed messages retry 3x before DLQ
-- [ ] Expired messages cleaned up within 5 minutes
-- [ ] Topic subscriptions work with wildcards
-- [ ] Saga pattern demonstrated with 3+ step workflow