def test_bin_pack():
assert get_bin_pack_residual([], [{"GPU": 2}, {"GPU": 2}]) == \
[{"GPU": 2}, {"GPU": 2}]
assert get_bin_pack_residual([{"GPU": 2}], [{"GPU": 2}, {"GPU": 2}]) == \
[{"GPU": 2}]
assert get_bin_pack_residual([{"GPU": 4}], [{"GPU": 2}, {"GPU": 2}]) == []
arg = [{"GPU": 2}, {"GPU": 2, "CPU": 2}]
assert get_bin_pack_residual(arg, [{"GPU": 2}, {"GPU": 2}]) == []
arg = [{"CPU": 2}, {"GPU": 2}]
assert get_bin_pack_residual(arg, [{"GPU": 2}, {"GPU": 2}]) == [{"GPU": 2}]
def test_bin_pack():
assert get_bin_pack_residual([], [{"GPU": 2}, {"GPU": 2}])[0] == \
[{"GPU": 2}, {"GPU": 2}]
assert get_bin_pack_residual([{"GPU": 2}], [{"GPU": 2}, {"GPU": 2}])[0] \
== [{"GPU": 2}]
assert get_bin_pack_residual([{
"GPU": 4
}], [{
"GPU": 2
}, {
"GPU": 2
}])[0] == []
arg = [{"GPU": 2}, {"GPU": 2, "CPU": 2}]
assert get_bin_pack_residual(arg, [{"GPU": 2}, {"GPU": 2}])[0] == []
arg = [{"CPU": 2}, {"GPU": 2}]
assert get_bin_pack_residual(arg, [{
"GPU": 2
}, {
"GPU": 2
}])[0] == [{
"GPU": 2
}]
arg = [{"GPU": 3}]
assert get_bin_pack_residual(arg, [{
"GPU": 1
}, {
"GPU": 1
}],
strict_spread=False)[0] == []
assert get_bin_pack_residual(arg, [{
"GPU": 1
}, {
"GPU": 1
}],
strict_spread=True) == ([{
"GPU": 1
}], [{
"GPU": 2
}])
def _get_nodes_allowed_to_terminate(
self, sorted_node_ids: List[NodeID]) -> Dict[NodeID, bool]:
# TODO(ameer): try merging this with resource_demand_scheduler
# code responsible for adding nodes for request_resources().
"""Returns the nodes allowed to terminate for request_resources().
Args:
sorted_node_ids: the node ids sorted based on last used (LRU last).
Returns:
nodes_allowed_to_terminate: whether the node id is allowed to
terminate or not.
"""
nodes_allowed_to_terminate: Dict[NodeID, bool] = {}
head_node_resources: ResourceDict = copy.deepcopy(
self.available_node_types[
self.config["head_node_type"]]["resources"])
if not head_node_resources:
# Legacy yaml might include {} in the resources field.
# TODO(ameer): this is somewhat duplicated in
# resource_demand_scheduler.py.
head_id: List[NodeID] = self.provider.non_terminated_nodes(
{TAG_RAY_NODE_KIND: NODE_KIND_HEAD})
if head_id:
head_ip = self.provider.internal_ip(head_id[0])
static_nodes: Dict[
NodeIP,
ResourceDict] = \
self.load_metrics.get_static_node_resources_by_ip()
head_node_resources = static_nodes.get(head_ip, {})
else:
head_node_resources = {}
max_node_resources: List[ResourceDict] = [head_node_resources]
resource_demand_vector_worker_node_ids = []
# Get max resources on all the non terminated nodes.
for node_id in sorted_node_ids:
tags = self.provider.node_tags(node_id)
if TAG_RAY_USER_NODE_TYPE in tags:
node_type = tags[TAG_RAY_USER_NODE_TYPE]
node_resources: ResourceDict = copy.deepcopy(
self.available_node_types[node_type]["resources"])
if not node_resources:
# Legacy yaml might include {} in the resources field.
static_nodes: Dict[
NodeIP,
ResourceDict] = \
self.load_metrics.get_static_node_resources_by_ip()
node_ip = self.provider.internal_ip(node_id)
node_resources = static_nodes.get(node_ip, {})
max_node_resources.append(node_resources)
resource_demand_vector_worker_node_ids.append(node_id)
# Since it is sorted based on last used, we "keep" nodes that are
# most recently used when we binpack. We assume get_bin_pack_residual
# is following the given order here.
used_resource_requests: List[ResourceDict]
_, used_resource_requests = \
get_bin_pack_residual(max_node_resources,
self.load_metrics.get_resource_requests())
# Remove the first entry (the head node).
max_node_resources.pop(0)
# Remove the first entry (the head node).
used_resource_requests.pop(0)
for i, node_id in enumerate(resource_demand_vector_worker_node_ids):
if used_resource_requests[i] == max_node_resources[i] \
and max_node_resources[i]:
# No resources of the node were needed for request_resources().
# max_node_resources[i] is an empty dict for legacy yamls
# before the node is connected.
nodes_allowed_to_terminate[node_id] = True
else:
nodes_allowed_to_terminate[node_id] = False
return nodes_allowed_to_terminate
