Errno::EPIPE / Errno::ECONNRESET in accept loop logged as unhandled exceptions

[thomaswitt]

When a client disconnects mid-response (e.g. browser navigation, WebSocket teardown, load-balancer health-check timeout), Falcon::Server#accept raises Errno::EPIPE or Errno::ECONNRESET. These bubble up through the Async task runner and are logged as: "Task may have ended with unhandled exception."

These are expected, harmless disconnects, but they flood logs with false alarms.

Reproduction

1. Start a Falcon-based Rails app
2. Use Turbo Drive to navigate between pages quickly (or let an ALB health-check timeout)
3. Observe console output:

Task may have ended with unhandled exception.
Errno::ECONNRESET: Connection reset by peer

This is particularly noisy with Hotwire/Turbo Drive, where every client-side navigation aborts the
previous response.

Current workaround

We monkey-patch Falcon::Server#accept to swallow only these two specific errors:

class Falcon::Server
  alias_method :__original_accept, :accept

  def accept(...)
    __original_accept(...)
  rescue Errno::EPIPE, Errno::ECONNRESET
    # Client disconnected mid-response — expected.
  end
end

Suggested fix

Handle Errno::EPIPE and Errno::ECONNRESET inside Falcon::Server#accept (or the connection task it spawns). These are not actionable errors — the client is simply gone. Something like:

def accept(...)
  @accept_count += 1
  @connection_count += 1

  super
rescue Errno::EPIPE, Errno::ECONNRESET
  # Client disconnected — nothing to do.
ensure
  @connection_count -= 1
end

All other exceptions would still propagate normally.

Or maybe intrduce a configuration switch whether to warn about these kind of errors

Environment

• Falcon: 0.54.0
• async-http: 0.94.2
• Ruby: 3.4.x

[ioquatix]

I agree we should do something here, but exactly what I'm not sure of yet. For sure there are lots of scenarios where EPIPE is abnormal. However, if browsers do it as a matter of normal request handling, maybe we need to suppress it.

[travisbell]

Just chiming in as a +1.

I ended up implementing the exact same change as @thomaswitt suggested because Errno::EPIPE "errors" were flooding our logs for no real reason. Since they can't really be actioned, it was just a bunch of noise.