Logs
The Logs pane shows one stream for the whole worktree instead of a tab per service. What you usually want to read is what happened here, in order, across processes, so the service is a filter and not a destination.
Log line format#
14:22:07 Server POST /api/orders 201 — 88ms| Part | Notes |
|---|---|
| Time | HH:MM:SS, local, from the backend. Lines from different services are interleaved by it. |
| Service | Which process emitted it. |
| Message | The raw line. Errors are red, warnings amber, ok lines green-tinted, info plain. |
The backend assigns levels as info, ok, warn or err.
The toolbar#
| Control | Behaviour |
|---|---|
| All plus one chip per service | Filters the stream to one service. Each chip carries a live status dot. |
| Levels | A popover with Errors, Warnings and Info. ok lines ride with Info instead of earning a fourth filter nobody would think to turn off. The chip shows how many levels are on when any are off. |
| Search logs… | Case-insensitive substring match over the message text. Cleared when you switch worktree. |
| Follow | Auto-scroll. It turns itself off the moment you scroll up, and back on when you click it. |
| Clear (✕) | Empties the buffers of every service in this worktree. |
If a filter combination hides everything, the pane says "No lines match these filters". With nothing logged at all it says "Nothing logged yet — start a service to see output here."
Inline crash recovery#
When a service exits with an error, a bar appears above the stream:
<Service>exited. The stack trace is at the end of this log.- Jump to error filters to that service, turns off Warnings and Info, re-enables Follow and scrolls to the end.
- Restart restarts the service without leaving the pane.
You can act on the failure where you're already reading it.
Buffer limits#
Each service gets a 160-line ring buffer in the Rust backend, and the oldest lines fall off the end. The backend batches emissions every 200 ms, which caps store updates at five per second per noisy service. That's noticeable during a webpack burst and invisible as tail latency.
Log events only go to windows that consume them, and not at all while every window is hidden, so a busy service doesn't keep an offscreen webview warm. Switching worktree primes every service's buffer from the backend, so the merged view starts complete instead of filling in one service at a time.
Operation logs are separate#
Provisioning, migrations, custom commands and database jobs stream into a per-worktree operation buffer of 300 lines rather than into a service's log. That buffer is what the setup runner turns into a step list, what the sidebar's in-progress row reads its current step from, and what a failure notice carries the tail of. A failed npm install stays readable after the dialog is gone, in the notice the attention queue holds.
No log file for services#
Service output lives in memory, in the ring buffer above. Canopy's own log (the Rust side: IPC errors, spawn failures, sweep decisions) goes to the platform log directory as canopy.log, rolling at 2 MB with one rotation kept, at INFO by default and controllable with RUST_LOG.