Include attributes when starting sqlalchemy spans for samplers

[alexmojaki]

In this code:

opentelemetry-python-contrib/instrumentation/opentelemetry-instrumentation-sqlalchemy/src/opentelemetry/instrumentation/sqlalchemy/engine.py

Lines 270 to 321 in 59cc34e

	def _set_db_client_span_attributes(self, span, statement, attrs) -> None:
	"""Uses statement and attrs to set attributes of provided Otel span"""
	span.set_attribute(SpanAttributes.DB_STATEMENT, statement)
	span.set_attribute(SpanAttributes.DB_SYSTEM, self.vendor)
	for key, value in attrs.items():
	span.set_attribute(key, value)
	def _before_cur_exec(
	self, conn, cursor, statement, params, context, _executemany
	):
	if not is_instrumentation_enabled():
	return statement, params
	attrs, found = _get_attributes_from_url(conn.engine.url)
	if not found:
	attrs = _get_attributes_from_cursor(self.vendor, cursor, attrs)
	db_name = attrs.get(SpanAttributes.DB_NAME, "")
	span = self.tracer.start_span(
	self._operation_name(db_name, statement),
	kind=trace.SpanKind.CLIENT,
	)
	with trace.use_span(span, end_on_exit=False):
	if span.is_recording():
	if self.enable_commenter:
	commenter_data = self._get_commenter_data(conn)
	if self.enable_attribute_commenter:
	# just to handle type safety
	statement = str(statement)
	# sqlcomment is added to executed query and db.statement span attribute
	statement = _add_sql_comment(
	statement, **commenter_data
	)
	self._set_db_client_span_attributes(
	span, statement, attrs
	)
	else:
	# sqlcomment is only added to executed query
	# so db.statement is set before add_sql_comment
	self._set_db_client_span_attributes(
	span, statement, attrs
	)
	statement = _add_sql_comment(
	statement, **commenter_data
	)
	else:
	# no sqlcomment anywhere
	self._set_db_client_span_attributes(span, statement, attrs)

There's no good reason to set the attributes only after checking is_recording, except:

1. when enable_commenter and enable_attribute_commenter are both true
2. specifically for the statement attribute
3. assuming that it's important that the initial and final values of the statement attribute are the same.

Otherwise, it's computationally trivial to include those attributes when starting the span. This would allow samplers to exclude spans with particular attributes (especially certain trivial queries which aren't worth tracing), and SpanProcessor.on_start to make use of the attributes.

[jasonjho]

+1. This would be extremely useful.