Post fork-removal, remote exception fetching is incorrect

[stuhood]

This code path:

pants/src/python/pants/bin/remote_pants_runner.py

Lines 162 to 181 in 421834c

	def _extract_remote_exception(self, pantsd_pid, nailgun_error):
	"""Given a NailgunError, returns a Terminated exception with additional info (where possible).
	This method will include the entire exception log for either the `pid` in the NailgunError, or
	failing that, the `pid` of the pantsd instance.
	"""
	sources = [pantsd_pid]
	if nailgun_error.pid is not None:
	sources = [abs(nailgun_error.pid)] + sources
	exception_text = None
	for source in sources:
	log_path = ExceptionSink.exceptions_log_path(for_pid=source)
	exception_text = maybe_read_file(log_path)
	if exception_text:
	break
	exception_suffix = '\nRemote exception:\n{}'.format(exception_text) if exception_text else ''
	return self.Terminated('abruptly lost active connection to pantsd runner: {!r}{}'.format(
	nailgun_error, exception_suffix))

...handles fetching the remote exception after a connection is terminated. But because the PID transmitted on connections no longer changes, this strategy is no longer accurate (and in fact, sending the PID to the client this way may not be useful).

[blorente]

I agree, if I understood the flow of NailgunError correctly, the pid in the error should always be the same pid as the daemon or None, which makes this code a bit redundant.

That means that in all cases, this code will fetch any exception triggered by the daemon.

[stuhood]

Yep. Perhaps we would need to do something like sending a session id, and passing it around as thread local, similar to the logging settings, and the zipkin trace ids? cc @cattibrie, @illicitonion

[stuhood]

This was resolved in d41acdf.