Add a basic Unix socket server using Bun

- Implemented a simple server using the net module.
- The server listens on a specified socket path.
- Added error handling for server errors.
- Included checks to verify the existence of the socket file.
This commit is contained in:
Alex Newman
2025-10-16 17:07:14 -04:00
parent 834cf4095e
commit 2d080b0264
7 changed files with 862 additions and 83 deletions
+232
View File
@@ -0,0 +1,232 @@
# Socket File Not Created - Debug Hypotheses
## Problem Statement
Worker process logs "Socket server listening: /Users/alexnewman/.claude-mem/worker-28.sock" but the socket file never appears on the filesystem. All connection attempts fail with `ENOENT`.
## Hypotheses (Ordered by Likelihood)
### H1: Worker Process Exits Immediately After Socket Creation
**Theory:** Worker creates socket, logs message, then crashes/exits before we poll for the file.
**Evidence:**
- We see the log message
- Socket never appears
- No other worker output after "listening" message
**Tests:**
- Check if worker process is running: `ps aux | grep worker`
- Add worker exit handlers to see exit code
- Check if worker.ts crashes after startSocketServer()
**Root Cause Possibilities:**
- Database query fails in loadSession() (worker.ts:75)
- SDK agent initialization crashes
- Unhandled promise rejection in run()
---
### H2: detached=false Kills Worker Prematurely
**Theory:** `detached: false` causes worker to die when replay script continues execution or when replay script changes process state.
**Evidence:**
- Production uses `detached: true, stdio: 'ignore'`
- Replay uses `detached: false, stdio: ['ignore', 'pipe', 'pipe']`
- Worker might be getting killed by parent process lifecycle
**Tests:**
- Change to `detached: true, stdio: 'ignore', worker.unref()`
- Check worker persists: `ps aux | grep worker` after spawn
**Expected Fix:**
- Worker should persist independently
- Socket should remain available
---
### H3: stdio Piping Interferes with Socket Creation
**Theory:** Piping stdout/stderr (`stdio: ['ignore', 'pipe', 'pipe']`) prevents proper socket file creation or causes worker to hang.
**Evidence:**
- Production uses `stdio: 'ignore'`
- We're trying to capture output with pipes
- This might interfere with Unix domain socket operations
**Tests:**
- Change to `stdio: 'ignore'` (no piping)
- Worker won't output to our console but should work
---
### H4: Socket Path Mismatch
**Theory:** Worker creates socket at different path than replay script expects.
**Evidence:**
- getWorkerSocketPath(sessionId) used in both places
- Both should resolve to ~/.claude-mem/worker-<id>.sock
- But maybe DATA_DIR differs between environments
**Tests:**
- Log actual socketPath in worker: `console.error('Creating socket at:', this.socketPath)`
- List all sockets: `ls -la ~/.claude-mem/*.sock`
- Check if socket appears elsewhere: `find /tmp -name "worker-*.sock"`
**Root Cause Possibilities:**
- CLAUDE_MEM_DATA_DIR environment variable difference
- Worker started with different env
---
### H5: Permissions Issue
**Theory:** Worker can't create socket file due to directory permissions.
**Evidence:**
- Socket creation might fail silently
- Worker logs "listening" before checking if socket file was created
**Tests:**
- Check ~/.claude-mem permissions: `ls -ld ~/.claude-mem`
- Try creating socket manually: `nc -U ~/.claude-mem/test.sock`
- Check worker user vs replay script user
**Expected Error:**
- Worker should throw EACCES or EPERM but we might not see it
---
### H6: Socket Listen Callback Fires Before File Creation
**Theory:** The server.listen() callback fires and logs "listening" before the socket file actually appears on filesystem.
**Evidence:**
- Node.js/Bun might call callback before filesystem sync
- We see log but no file
**Tests:**
- Add additional wait time after seeing log
- Add fs.existsSync check inside worker after listen()
- Increase poll duration/frequency in replay script
---
### H7: CLI Worker Command Routing Broken
**Theory:** `dist/claude-mem.min.js worker <sessionId>` doesn't properly route to worker.ts main().
**Evidence:**
- cli.ts has .command('worker') handler
- Handler imports and calls main() from sdk/worker.ts
- But bundling might break this
**Tests:**
- Run directly: `dist/claude-mem.min.js worker 28`
- Check if worker main() is actually called
- Add console.error at top of worker.ts main()
**Root Cause Possibilities:**
- Bundle doesn't include worker code
- Import path broken in minified CLI
- Commander routing fails
---
### H8: Database Session Not Found by Worker
**Theory:** Worker can't find session in database, exits early.
**Evidence:**
- loadSession() query might return null
- Code checks `if (!session) { exit(1) }` (worker.ts:76-79)
- But we'd expect to see error log
**Tests:**
- Verify session exists before spawn: `SELECT * FROM sdk_sessions WHERE id = ?`
- Add debug log in loadSession() before query
- Check DB file path matches
---
### H9: Socket File Created Then Immediately Deleted
**Theory:** Socket is created but something deletes it (cleanup from previous run, OS, etc).
**Evidence:**
- Old socket file might exist and get unlinked (worker.ts:110-112)
- Maybe multiple workers spawning
**Tests:**
- Check for multiple worker processes: `ps aux | grep worker`
- Watch filesystem in real-time: `watch ls -la ~/.claude-mem/`
- Add delay before cleanup code runs
---
### H10: Bun vs Node Runtime Issue
**Theory:** Worker runs under different runtime than expected, causing socket issues.
**Evidence:**
- Replay script uses bun: `#!/usr/bin/env bun`
- Worker spawned via CLI which uses node: `#!/usr/bin/env node`
- Runtime difference might affect socket creation
**Tests:**
- Spawn with explicit bun: `bun dist/claude-mem.min.js worker 28`
- Or spawn with explicit node
- Check if runtime matters for Unix sockets
---
### H11: Race Condition in Socket Server Startup
**Theory:** server.listen() completes but socket isn't ready for connections yet.
**Evidence:**
- We poll for 15 seconds
- Maybe socket file appears but isn't ready
- Connection attempts might be too early
**Tests:**
- Increase wait time after socket found
- Try connecting with retry logic
- Check socket file permissions/readiness
---
### H12: Worker Logs to Wrong Stream
**Theory:** Worker logs "listening" to stdout/stderr but then crashes, and we only see initial log.
**Evidence:**
- console.error used in worker (worker.ts:86)
- With stdio: ['ignore', 'pipe', 'pipe'], stderr is piped
- Maybe crash happens but we don't see it
**Tests:**
- Check full worker output captured
- Look for crash stack traces
- Add more logging throughout worker.run()
---
## Recommended Debug Sequence
1. **Change spawn config to match production exactly**
- `detached: true`
- `stdio: 'ignore'`
- `worker.unref()`
- This eliminates H2, H3
2. **Check worker process persistence**
- `ps aux | grep worker` immediately after spawn
- If not running → H1, H7, H8
- If running → H4, H5, H6
3. **Check socket file location**
- `ls -la ~/.claude-mem/*.sock`
- `find /tmp -name "worker-*.sock"`
- If found elsewhere → H4
- If not found → H1, H5, H6
4. **Run worker directly for debugging**
- `dist/claude-mem.min.js worker 28` manually
- See full output
- Check if socket appears
5. **Add more worker logging**
- Log at start of main()
- Log after loadSession()
- Log after startSocketServer() promise resolves
- Log socket path being used
+55 -55
View File
File diff suppressed because one or more lines are too long
+11 -11
View File
File diff suppressed because one or more lines are too long
+361
View File
@@ -0,0 +1,361 @@
#!/usr/bin/env bun
/**
* Transcript Replay Tool
*
* Plays back a Claude Code transcript through the memory system to test:
* 1. Tool observation capture
* 2. SDK worker processing
* 3. SQLite storage
* 4. Session summary generation
*/
import { readFileSync } from 'fs';
import { join } from 'path';
import * as net from 'net';
import { HooksDatabase } from '../src/services/sqlite/HooksDatabase';
import { getWorkerSocketPath } from '../src/shared/paths';
import { spawn } from 'child_process';
interface TranscriptLine {
type: string;
message?: {
role?: string;
content?: Array<{
type: string;
name?: string;
input?: any;
output?: string;
id?: string;
}>;
};
uuid?: string;
sessionId?: string;
timestamp?: string;
}
interface ToolUse {
id: string;
name: string;
input: any;
output?: string;
timestamp: string;
}
/**
* Parse transcript JSONL file and extract tool uses with their results
*/
function parseTranscript(filePath: string): ToolUse[] {
const content = readFileSync(filePath, 'utf-8');
const lines = content.trim().split('\n');
const toolUses: Map<string, ToolUse> = new Map();
for (const line of lines) {
try {
const event: TranscriptLine = JSON.parse(line);
// Capture tool_use from assistant messages
if (event.type === 'assistant' && event.message?.content) {
for (const item of event.message.content) {
if (item.type === 'tool_use' && item.name && item.id) {
toolUses.set(item.id, {
id: item.id,
name: item.name,
input: item.input,
timestamp: event.timestamp || new Date().toISOString(),
});
}
}
}
// Capture tool_result from user messages
// Tool results come in user messages with tool_use_id
if (event.type === 'user' && event.message?.content) {
const content = event.message.content;
// Content can be array or single object
const items = Array.isArray(content) ? content : [content];
for (const item of items) {
if (item && typeof item === 'object' && 'type' in item && item.type === 'tool_result') {
const toolUseId = (item as any).tool_use_id;
const toolContent = (item as any).content;
if (toolUseId) {
const toolUse = toolUses.get(toolUseId);
if (toolUse) {
toolUse.output = toolContent || '';
}
}
}
}
}
} catch (err) {
// Skip invalid lines
continue;
}
}
return Array.from(toolUses.values()).filter(t => t.output !== undefined);
}
/**
* Send observation to SDK worker via Unix socket
*/
async function sendObservation(
socketPath: string,
toolName: string,
toolInput: any,
toolOutput: string
): Promise<void> {
return new Promise((resolve, reject) => {
const client = net.createConnection(socketPath, () => {
const message = JSON.stringify({
type: 'observation',
tool_name: toolName,
tool_input: toolInput,
tool_output: toolOutput,
});
client.write(message + '\n');
client.end();
resolve();
});
client.on('error', reject);
client.setTimeout(5000);
client.on('timeout', () => {
client.destroy();
reject(new Error('Socket timeout'));
});
});
}
/**
* Send finalize message to SDK worker
*/
async function sendFinalize(socketPath: string): Promise<void> {
return new Promise((resolve, reject) => {
const client = net.createConnection(socketPath, () => {
const message = JSON.stringify({ type: 'finalize' });
client.write(message + '\n');
client.end();
resolve();
});
client.on('error', reject);
client.setTimeout(5000);
client.on('timeout', () => {
client.destroy();
reject(new Error('Socket timeout'));
});
});
}
/**
* Main replay function
*/
async function replayTranscript(transcriptPath: string, projectName: string = 'claude-mem-test') {
console.log('🎬 Starting transcript replay...\n');
// Parse transcript
console.log(`📖 Parsing transcript: ${transcriptPath}`);
const toolUses = parseTranscript(transcriptPath);
console.log(` Found ${toolUses.length} tool uses\n`);
// Initialize database
const hooksDb = new HooksDatabase();
// Create SDK session
console.log('🔧 Creating SDK session...');
const claudeSessionId = `replay-${Date.now()}`;
const userPrompt = 'Replaying transcript for testing';
const sessionId = await hooksDb.createSDKSession(
claudeSessionId,
projectName,
userPrompt
);
console.log(` Session ID: ${sessionId}`);
// Verify session was created
const verifyQuery = (hooksDb as any).db.query(`
SELECT id, claude_session_id, project FROM sdk_sessions WHERE id = ?
`);
const session = verifyQuery.get(sessionId);
if (!session) {
console.error(' ❌ Session not found in database after creation!');
process.exit(1);
}
console.log(` ✅ Session verified in database\n`);
// Spawn SDK worker
console.log('🚀 Spawning SDK worker...');
const socketPath = getWorkerSocketPath(sessionId);
// Spawn worker exactly as production hooks do
const workerPath = join(process.cwd(), 'scripts/hooks/worker.js');
const worker = spawn('bun', [workerPath, String(sessionId)], {
detached: false, // Keep attached to see errors
stdio: ['ignore', 'pipe', 'pipe'] // Pipe output to see what's happening
});
worker.stdout?.on('data', (data) => {
console.log(` [worker stdout] ${data}`);
});
worker.stderr?.on('data', (data) => {
console.error(` [worker stderr] ${data}`);
});
worker.on('exit', (code, signal) => {
console.error(` [worker] Exited with code ${code}, signal ${signal}`);
});
worker.on('error', (err) => {
console.error(`\n [worker] Process error:`, err.message);
});
// Wait for socket to be ready
console.log(` Waiting for socket: ${socketPath}`);
// Poll for socket existence
let socketReady = false;
for (let i = 0; i < 30; i++) {
await new Promise(resolve => setTimeout(resolve, 500));
try {
const fs = await import('fs');
if (fs.existsSync(socketPath)) {
socketReady = true;
console.log(` ✅ Socket ready after ${(i + 1) * 500}ms`);
break;
}
} catch (err) {
// Continue waiting
}
}
if (!socketReady) {
console.log(` ⚠️ Socket not found after 15s, attempting to connect anyway...`);
}
// Additional wait for worker to be fully initialized
await new Promise(resolve => setTimeout(resolve, 2000));
// Send observations
console.log(`\n📤 Sending ${toolUses.length} observations...`);
let sent = 0;
let failed = 0;
for (const toolUse of toolUses) {
try {
await sendObservation(
socketPath,
toolUse.name,
toolUse.input,
toolUse.output || ''
);
sent++;
process.stdout.write(`\r Sent: ${sent}/${toolUses.length}`);
// Small delay between observations
await new Promise(resolve => setTimeout(resolve, 100));
} catch (err) {
failed++;
console.error(`\n ❌ Failed to send observation: ${err.message}`);
}
}
console.log(`\n ✅ Successfully sent ${sent} observations`);
if (failed > 0) {
console.log(` ⚠️ Failed to send ${failed} observations`);
}
// Wait for processing
console.log('\n⏳ Waiting for SDK to process observations...');
await new Promise(resolve => setTimeout(resolve, 5000));
// Send finalize
console.log('\n🏁 Sending finalize message...');
try {
await sendFinalize(socketPath);
console.log(' ✅ Finalize message sent');
} catch (err) {
console.error(` ❌ Failed to send finalize: ${err.message}`);
}
// Wait for summary generation
console.log('\n⏳ Waiting for summary generation...');
await new Promise(resolve => setTimeout(resolve, 10000));
// Verify results
console.log('\n🔍 Verifying results...\n');
// Check observations using direct DB query
const observations = (hooksDb as any).db.query(`
SELECT sdk_session_id, project, text, type, created_at
FROM observations
WHERE sdk_session_id = (
SELECT sdk_session_id FROM sdk_sessions WHERE id = ?
)
ORDER BY created_at_epoch ASC
`).all(sessionId);
console.log(` 📝 Observations stored: ${observations.length}`);
if (observations.length > 0) {
console.log(' Sample observations:');
observations.slice(0, 3).forEach((obs: any, i: number) => {
console.log(` ${i + 1}. [${obs.type}] ${obs.text.substring(0, 60)}...`);
});
}
// Check summary using direct DB query
const summary = (hooksDb as any).db.query(`
SELECT request, investigated, learned, completed, next_steps,
files_read, files_edited, notes, created_at
FROM session_summaries
WHERE sdk_session_id = (
SELECT sdk_session_id FROM sdk_sessions WHERE id = ?
)
LIMIT 1
`).get(sessionId);
if (summary) {
console.log(`\n 📋 Summary generated:`);
console.log(` Request: ${(summary as any).request?.substring(0, 60)}...`);
console.log(` Completed: ${(summary as any).completed?.substring(0, 60)}...`);
const filesRead = JSON.parse((summary as any).files_read || '[]');
const filesEdited = JSON.parse((summary as any).files_edited || '[]');
console.log(` Files read: ${filesRead.length}`);
console.log(` Files edited: ${filesEdited.length}`);
} else {
console.log(`\n ⚠️ No summary generated`);
}
// Cleanup (worker is detached and will exit on its own)
console.log('\n✅ Replay complete!\n');
return {
sessionId,
observationsCount: observations.length,
hasSummary: !!summary,
};
}
// CLI interface
const args = process.argv.slice(2);
const transcriptPath = args[0] || join(process.cwd(), 'test-data/sample-transcript.jsonl');
const projectName = args[1] || 'claude-mem-test';
replayTranscript(transcriptPath, projectName)
.then((result) => {
console.log('Results:', result);
process.exit(0);
})
.catch((err) => {
console.error('❌ Replay failed:', err);
process.exit(1);
});
+52 -17
View File
@@ -33,7 +33,9 @@ type WorkerMessage = ObservationMessage | FinalizeMessage;
* Main worker process entry point
*/
export async function main() {
console.error('[SDK Worker DEBUG] main() called');
const sessionDbId = parseInt(process.argv[2], 10);
console.error(`[SDK Worker DEBUG] Session DB ID: ${sessionDbId}`);
if (!sessionDbId) {
console.error('[SDK Worker] Missing session ID argument');
@@ -41,6 +43,7 @@ export async function main() {
}
const worker = new SDKWorker(sessionDbId);
console.error('[SDK Worker DEBUG] SDKWorker instance created');
await worker.run();
}
@@ -106,12 +109,17 @@ class SDKWorker {
* Start Unix socket server to receive messages from hooks
*/
private async startSocketServer(): Promise<void> {
console.error(`[SDK Worker DEBUG] Starting socket server...`);
console.error(`[SDK Worker DEBUG] Socket path: ${this.socketPath}`);
// Clean up old socket if it exists
if (existsSync(this.socketPath)) {
console.error(`[SDK Worker DEBUG] Removing existing socket`);
unlinkSync(this.socketPath);
}
return new Promise((resolve, reject) => {
console.error(`[SDK Worker DEBUG] Creating net server...`);
this.server = net.createServer((socket) => {
let buffer = '';
@@ -147,6 +155,8 @@ class SDKWorker {
});
this.server.listen(this.socketPath, () => {
console.error(`[SDK Worker DEBUG] listen() callback fired`);
console.error(`[SDK Worker DEBUG] Checking if socket exists: ${existsSync(this.socketPath)}`);
resolve();
});
});
@@ -183,21 +193,28 @@ class SDKWorker {
* Run SDK agent with streaming input mode
*/
private async runSDKAgent(): Promise<void> {
// Find Claude Code executable
const claudePath = process.env.CLAUDE_CODE_PATH || '/Users/alexnewman/.nvm/versions/node/v24.5.0/bin/claude';
console.error(`[SDK Worker DEBUG] About to call query with claudePath: ${claudePath}`);
await query({
model: MODEL,
messages: () => this.createMessageGenerator(),
disallowedTools: DISALLOWED_TOOLS,
signal: this.abortController.signal,
onSystemInitMessage: (msg) => {
// Capture SDK session ID from init message
if (msg.session_id) {
this.sdkSessionId = msg.session_id;
this.db.updateSDKSessionId(this.sessionDbId, msg.session_id);
prompt: this.createMessageGenerator(),
options: {
model: MODEL,
disallowedTools: DISALLOWED_TOOLS,
abortController: this.abortController,
pathToClaudeCodeExecutable: claudePath,
onSystemInitMessage: (msg) => {
// Capture SDK session ID from init message
if (msg.session_id) {
this.sdkSessionId = msg.session_id;
this.db.updateSDKSessionId(this.sessionDbId, msg.session_id);
}
},
onAgentMessage: (msg) => {
// Parse and store observations from agent response
this.handleAgentMessage(msg.content);
}
},
onAgentMessage: (msg) => {
// Parse and store observations from agent response
this.handleAgentMessage(msg.content);
}
});
}
@@ -206,11 +223,17 @@ class SDKWorker {
* Create async message generator for SDK streaming input
* Now pulls from socket messages instead of polling database
*/
private async* createMessageGenerator(): AsyncIterable<{ role: 'user'; content: string }> {
private async* createMessageGenerator(): AsyncIterable<{ type: 'user'; message: { role: 'user'; content: string } }> {
// Yield initial prompt
const claudeSessionId = `session-${this.sessionDbId}`;
const initPrompt = buildInitPrompt(this.project, claudeSessionId, this.userPrompt);
yield { role: 'user', content: initPrompt };
yield {
type: 'user',
message: {
role: 'user',
content: initPrompt
}
};
// Process messages as they arrive via socket
while (!this.isFinalized) {
@@ -229,7 +252,13 @@ class SDKWorker {
const session = await this.loadSession();
if (session) {
const finalizePrompt = buildFinalizePrompt(session);
yield { role: 'user', content: finalizePrompt };
yield {
type: 'user',
message: {
role: 'user',
content: finalizePrompt
}
};
}
break;
}
@@ -241,7 +270,13 @@ class SDKWorker {
tool_input: message.tool_input,
tool_output: message.tool_output
});
yield { role: 'user', content: observationPrompt };
yield {
type: 'user',
message: {
role: 'user',
content: observationPrompt
}
};
}
}
}
File diff suppressed because one or more lines are too long
+18
View File
@@ -0,0 +1,18 @@
#!/usr/bin/env bun
import net from 'net';
import { existsSync } from 'fs';
const socketPath = '/Users/alexnewman/.claude-mem/test-bun.sock';
const server = net.createServer(() => {});
server.listen(socketPath, () => {
console.log('Server listening');
console.log('existsSync says:', existsSync(socketPath));
console.log('Checking with ls...');
});
server.on('error', (err) => {
console.error('Error:', err);
process.exit(1);
});