Real-time Logs

The Core captures output from backend processes and broadcasts it via Server-Sent Events (SSE). Logs are ephemeral (in-memory broadcast channel, not persisted to disk).

How it works

Each worker process gets a broadcast channel (capacity: 256 entries). Log entries are pushed to the channel from three sources:

1. Worker IPC messages: console.log(), console.warn(), console.error() in your backend code are intercepted by the prelude and sent as structured IPC messages with a level.
2. Raw stderr: anything written directly to stderr is captured line-by-line with level "stderr".
3. Core lifecycle events: worker start, stop, crash, and backoff events with level "system".

Subscribers receive entries via SSE as they are produced.

Log levels

Logging from your backend

The prelude intercepts standard console methods and routes them through IPC:

// These all go through IPC as structured log messages:
console.log("Processing:", method);     // level: "info"
console.warn("Rate limit approaching"); // level: "warn"
console.error("Connection failed");     // level: "error"

// Or use the explicit log object:
log.info("Processing started");
log.warn("Approaching limit");
log.error("Connection failed");

console.debug is aliased to console.log (level: "info").

Log messages are truncated at 8192 characters.

SSE endpoint

GET /api/v1/apps/{appId}/logs
Authorization: Bearer <token>

Requires authentication. Returns text/event-stream.

Each event contains a JSON payload:

data: {"level":"info","message":"Processing order #123"}

data: {"level":"stderr","message":"Warning: deprecated API usage"}

data: {"level":"system","message":"worker started"}

Keep-alive

The SSE stream sends a keep-alive comment every 15 seconds to prevent client timeout.

Lagged subscribers

If a subscriber falls behind (more than 256 entries in the buffer), it receives a system message indicating how many entries were dropped:

data: {"level":"system","message":"... 42 messages dropped ..."}

The stream continues from the current position after the lag notification.

Stream termination

The stream ends when:

• The worker is stopped or uninstalled (channel closed).
• The client disconnects.

Custom events

Your backend can emit named events:

emit("order_processed", { orderId: "abc123", total: 4999 });

These appear in the log stream with level "event" and the format "{name}: {data}".

Technical details

API endpoints summary