openpull

OpenPull Node.js Library

A TypeScript/JavaScript library for streaming application logs to OpenPull via WebRTC. Start fast with the logger, then dive deeper as needed.

Quick Start

1) Install

npm install openpull

2) Minimal Library Usage (Logger first)

import { createLogger, createConnection } from 'openpull';

// JSON logger to stdout (pino-like)
const log = createLogger();
log.info('App started');
log.trace('request').span('db-query').finish();

// Forward stdout/stderr to OpenPull (separate from logger)
const connection = await createConnection(process.env.OPENPULL_URL!);
await connection.forward(process.stdout, process.stderr);

You can also keep your existing logger (pino/winston/console) and just add a trace_id when you want correlation.

3) Minimal CLI Usage (Any language)

# One-time global install for the CLI (optional)
npm install -g openpull

# Wrap any command; uses OPENPULL_URL if set
openpull -- node app.js
openpull -- python app.py

Example minimal Node.js file (app.js):

// app.js
const traceId = 'trace-' + Date.now();

console.log(JSON.stringify({ level: 'info', message: 'Server starting', port: 3000 }));
console.log(JSON.stringify({ level: 'info', message: 'Database connected', db: 'postgres' }));

// Three trace logs sharing the same trace_id (correlated)
console.log(JSON.stringify({ level: 'trace', message: 'Checkout flow', step: 'begin', trace_id: traceId }));
console.log(JSON.stringify({ level: 'trace', message: 'Payment processed', amount: 99.99, trace_id: traceId }));
console.log(JSON.stringify({ level: 'trace', message: 'Order completed', order_id: '12345', trace_id: traceId }));

That’s it—you’re sending logs to OpenPull.

Concepts

• Everything is a log entry: Logs, metrics, and traces are just JSON log entries. Tracing is simply adding a shared field (commonly trace_id).
• No schema required: Log any JSON. The dashboard discovers fields dynamically and lets you filter, group, and search by anything you emit.

Example entries you might emit:

log.info('User action', { userId: 123, action: 'login', region: 'us-east' });
log.info('Database query', { query: 'SELECT * FROM users', duration: 45, rows: 150 });
// Grouping can be by any field you choose, not only trace_id
log.info('Processing request', { request_id: 'req-123', service: 'api', endpoint: '/users' });

Features

• Unified observability: logs, metrics, and traces as log entries
• Works with any logger: pino, winston, console.log, or custom
• Real-time streaming: WebRTC connection to OpenPull
• Standalone structured logger: optional, pino-like ergonomics
• Clean separation: logger writes to stdout; connection forwards
• CLI tool: wrap any language/executable
• TypeScript: full typings, no classes

How It Works

The magic is in the ultra-flexible parsing. Here's what happens under the hood:

1. No Schema Validation

When you log anything, OpenPull just:

// From connection-manager.ts:12
function parseLogLine(line: string, defaultLevel: string = 'info'): any {
  try {
    const parsed = JSON.parse(trimmedLine);
    return {
      type: parsed.level || parsed.type || defaultLevel,
      message: parsed.message || parsed.msg || trimmedLine,
      timestamp: parsed.timestamp || parsed.time || new Date().toISOString(),
      ...parsed,  // ← ALL your fields get preserved
    };
  } catch {
    // Plain text becomes a log entry too
    return {
      type: defaultLevel,
      message: trimmedLine,
      timestamp: new Date().toISOString(),
    };
  }
}

2. Dynamic Field Discovery

The dashboard uses DuckDB's JSON operators to discover fields at runtime:

-- Gets ALL unique field names across your logs
WITH json_keys AS (
  SELECT DISTINCT json_keys(data) as keys FROM logs
)
SELECT key, COUNT(*) as count 
FROM json_keys jk, UNNEST(jk.keys) as t(key), logs l
GROUP BY key ORDER BY count DESC

3. Flexible Grouping

You can group by ANY field, not just trace_id:

-- Group by trace_id (traditional)
WHERE json_extract(data, '$.trace_id') = 'abc123'

-- Or group by request_id 
WHERE json_extract(data, '$.request_id') = 'req-456'

-- Or group by user_id
WHERE json_extract(data, '$.user_id') = '789'

-- Or any custom field you invent
WHERE json_extract(data, '$.workflow_id') = 'deploy-123'

trace_id is a convention, not a requirement. Correlate by any shared field.

API Reference

Library API

createLogger(options?: LoggerOptions): Logger

Creates a standalone logger that outputs JSON to stdout (competes with pino):

const log = createLogger({
  defaultFields: {
    service: 'my-app',
    version: '1.0.0'
  }
});

// Basic logging
log.info('User logged in', { userId: '123' });
log.error('Database error', { query: 'SELECT * FROM users' });
log.debug('Processing request', { requestId: 'req-456' });
log.warning('Rate limit approaching', { current: 95, limit: 100 });

// Distributed tracing with trace_id correlation
log.trace('request').span('db-query').span('validation').finish();

// Each span gets the same trace_id automatically:
// {"level":"trace","message":"request","trace_id":"abc123","span_id":"span1"...}
// {"level":"trace","message":"db-query","trace_id":"abc123","span_id":"span2"...}
// {"level":"trace","message":"validation","trace_id":"abc123","span_id":"span3"...}

// Or traditional approach
const trace = log.startTrace({ operation: 'checkout' });
trace.span('Validate cart');    // Same trace_id
trace.span('Process payment');  // Same trace_id
trace.finish();

createConnection(connectionString: string): Promise<Connection>

Creates a connection that can forward streams:

const connection = await createConnection('openpull://appender:[e-mail]/NoDrBwJRXA8W');

// Forward process streams
await connection.forward(process.stdout, process.stderr);

// Or forward child process streams (used internally by CLI)
await connection.forwardStreams(child.stdout, child.stderr);

Note: The URL includes a key, but OpenPull uses a zero-knowledge handshake — the key never leaves your device.

Connection String Format:

openpull://role:[e-mail]:port/

• role: Either appender (for sending logs) or reader (for receiving logs)
• key: Authentication key derived from your session (used only locally to compute the handshake proof)
• publicToken: Session identifier
• host:port: Signaling server address

CLI Reference

Wraps any application and forwards stdout/stderr:

export OPENPULL_URL="openpull://appender:[e-mail]/NoDrBwJRXA8W"
openpull -- node app.js       # Node
openpull -- python script.py  # Python
openpull -- go run main.go    # Go
openpull -- dotnet run        # .NET

# Help
openpull --help

Env vars:

• OPENPULL_URL – default connection URL

TypeScript Support

The library is written in TypeScript and provides comprehensive type definitions:

import { Logger, Tracer, LogData } from 'openpull';

const log: Logger = logger();
const trace: Tracer = log.startTrace();

// All methods are fully typed
log.info('Message', { customField: 'value' });

Error Handling

The library gracefully handles connection failures:

try {
  await connect('openpull://appender:[e-mail]/NoDrBwJRXA8W');
} catch (error) {
  console.error('Failed to connect:', error.message);
}

// Logs fall back to console if no connection
const log = logger();
log.info('This will go to console if not connected');

Examples

Library: Standalone Logger

import { createLogger } from 'openpull';

// Pure logger that outputs JSON to stdout (like pino)
const log = createLogger({
  defaultFields: {
    service: 'user-service',
    version: '1.0.0'
  }
});

log.info('Service started', { port: 3000 });
log.error('Database error', { table: 'users' });

// Distributed tracing with trace_id correlation
log.trace('checkout')
  .span('validate-cart')     // trace_id: "xyz789", span_id: "span1"
  .span('process-payment')   // trace_id: "xyz789", span_id: "span2" 
  .span('update-inventory')  // trace_id: "xyz789", span_id: "span3"
  .finish();

Library: Clean Separation

import { createLogger, createConnection } from 'openpull';

// Logger just outputs to stdout
const log = createLogger({ defaultFields: { service: 'api' } });

// Connection handles WebRTC separately  
const connection = await createConnection(process.env.OPENPULL_URL);
await connection.forward(process.stdout, process.stderr);

// Now both structured logs AND console.log get forwarded
log.info('Structured log');
console.log('Plain console log');

Express.js with Tracing

import express from 'express';
import { createLogger, createConnection } from 'openpull';

const app = express();

// Setup separate logger and connection
const log = createLogger({ defaultFields: { service: 'api' } });
const connection = await createConnection(process.env.OPENPULL_URL);
await connection.forward(process.stdout, process.stderr);

// Request tracing middleware
app.use((req, res, next) => {
  // Start a trace for each request
  const trace = log.startTrace({ 
    method: req.method, 
    url: req.url,
    userAgent: req.headers['user-agent'] 
  });
  
  req.trace = trace;
  req.trace.span('Request started');
  next();
});

app.get('/users/:id', (req, res) => {
  req.trace.span('Fetching user', { userId: req.params.id });
  
  // All these spans share the same trace_id:
  req.trace.span('Database query');
  req.trace.span('Permission check');
  req.trace.span('Response serialization');
  
  req.trace.span('Request completed', { statusCode: res.statusCode });
  req.trace.finish();
});

Why This Logger vs Existing Ones?

• Use any logger + CLI: Keep pino/winston/console and forward via CLI.
• Use OpenPull logger: If you want built-in tracing and automatic correlation.

Example with pino (manual correlation):

import pino from 'pino';
const log = pino();
const traceId = 'trace-' + Math.random().toString(36).slice(2);
log.info({ trace_id: traceId }, 'Request started');

Dynamic Dashboard

The OpenPull web dashboard has zero configuration - it adapts to whatever fields you actually log:

Auto-Discovery

// Dashboard automatically discovers these fields from your logs:
async function updateSchema() {
  const schema = await getLogSchema(db);
  
  // Finds ALL unique fields: userId, action, browser, region, customField, etc.
  state.availableFields = new Set(schema.map(s => s.key));
  
  // Auto-shows frequently used fields (> 10 occurrences)  
  schema.forEach(field => {
    if (field.count > 10 && !state.visibleColumns.has(field.key)) {
      state.visibleColumns.add(field.key);
    }
  });
}

Dynamic Filtering

The dashboard creates filters for every field it finds:

// Creates dropdowns/filters for ANY field
state.availableFields.forEach(field => {
  const fieldValues = await getFieldValues(db, field);
  // Now you can filter by userId, region, cache status, etc.
});

Flexible Grouping

You can group "traces" by any field:

• Traditional: Group by trace_id to see request flows
• By user: Group by user_id to see all user activity
• By service: Group by service to see service interactions
• Custom: Group by deployment_id, experiment_id, workflow_id, etc.

No schema to define upfront — log any JSON structure and the dashboard adapts automatically.