## TL;DR

Karma disconnects browsers that stop responding, usually from timeouts or resource exhaustion. Raise the capture and disconnect timeouts, reduce concurrency, and give the runner more resources.

## Error

```text
WARN [launcher]: Chrome have not captured in 60000 ms, killing.
WARN [Chrome]: Disconnected (1 times), because no message in 10000 ms.
```

## Steps

1. Raise `captureTimeout` and `browserNoActivityTimeout` in karma.conf.js. Expected: slow CI stops tripping them.
2. Reduce `concurrency` to match runner CPUs. Expected: less contention.
3. Use `ChromeHeadless` with `--no-sandbox --disable-gpu` in containers. Expected: launches reliably.
4. Check memory: browser disconnects under OOM look like timeouts. Expected: resources adequate.
5. Re-run. Expected: browsers stay connected.

## When to use

- `Disconnected` or `have not captured` in CI.
- Containerized Karma runs.

## When not to use

- Test failures (the tests run; the browser is fine).
- You are migrating off Karma (consider Vitest instead).

## Tool compatibility

- Karma 6.x; ChromeHeadless launcher.

## Variant phrasings

### Karma browser disconnected

The general error; timeouts and resources.

### Chrome have not captured in Karma

The capture variant; raise captureTimeout.

## Why it happens

Karma drives browsers over a socket with timeouts tuned for fast machines. Slow or loaded CI trips them.

## Edge cases

- Single-run vs watch mode have different timeout needs.
- Proxy env vars can break the Karma socket; unset them for YOUR_HOST.
- Consider this a migration signal; Karma is in maintenance.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_eRYfeyNL71eagXbBTM0-2w