Skip to content

Commit 115037a

Browse files
sandereggsanderegg
committed
setup of test done
1 parent 405f3c2 commit 115037a

File tree

2 files changed

+144
-44
lines changed

2 files changed

+144
-44
lines changed

‎services/autoscaling/tests/unit/conftest.py‎

Lines changed: 9 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1023,6 +1023,15 @@ def with_instances_machines_hot_buffer(
10231023
)
10241024

10251025

1026+
@pytest.fixture
1027+
def hot_buffer_instance_type(app_settings: ApplicationSettings) -> InstanceTypeType:
1028+
assert app_settings.AUTOSCALING_EC2_INSTANCES
1029+
return cast(
1030+
InstanceTypeType,
1031+
next(iter(app_settings.AUTOSCALING_EC2_INSTANCES.EC2_INSTANCES_ALLOWED_TYPES)),
1032+
)
1033+
1034+
10261035
@pytest.fixture
10271036
def mock_find_node_with_name_returns_none(mocker: MockerFixture) -> Iterator[mock.Mock]:
10281037
return mocker.patch(

‎services/autoscaling/tests/unit/test_modules_auto_scaling_dynamic.py‎

Lines changed: 135 additions & 44 deletions
Original file line numberDiff line numberDiff line change
@@ -66,6 +66,9 @@
6666
AutoscalingDocker,
6767
get_docker_client,
6868
)
69+
from simcore_service_autoscaling.utils.auto_scaling_core import (
70+
node_host_name_from_ec2_private_dns,
71+
)
6972
from simcore_service_autoscaling.utils.utils_docker import (
7073
_OSPARC_NODE_EMPTY_DATETIME_LABEL_KEY,
7174
_OSPARC_NODE_TERMINATION_PROCESS_LABEL_KEY,
@@ -2088,6 +2091,7 @@ async def test_warm_buffers_only_replace_hot_buffer_if_service_is_started_issue7
20882091
patch_ec2_client_launch_instances_min_number_of_instances: mock.Mock,
20892092
minimal_configuration: None,
20902093
with_instances_machines_hot_buffer: EnvVarsDict,
2094+
with_drain_nodes_labelled: bool,
20912095
ec2_client: EC2Client,
20922096
initialized_app: FastAPI,
20932097
app_settings: ApplicationSettings,
@@ -2097,18 +2101,27 @@ async def test_warm_buffers_only_replace_hot_buffer_if_service_is_started_issue7
20972101
[int, InstanceTypeType, InstanceStateNameType, list[DockerGenericTag] | None],
20982102
Awaitable[list[str]],
20992103
],
2104+
create_services_batch: Callable[[_ScaleUpParams], Awaitable[list[Service]]],
2105+
hot_buffer_instance_type: InstanceTypeType,
21002106
spied_cluster_analysis: MockType,
21012107
instance_type_filters: Sequence[FilterTypeDef],
21022108
stopped_instance_type_filters: Sequence[FilterTypeDef],
21032109
mock_find_node_with_name_returns_fake_node: mock.Mock,
21042110
mock_compute_node_used_resources: mock.Mock,
21052111
mock_docker_tag_node: mock.Mock,
2112+
mocker: MockerFixture,
2113+
fake_node: Node,
21062114
):
21072115
# NOTE: https://github.com/ITISFoundation/osparc-simcore/issues/7071
21082116

2109-
# pre-requisites
2117+
#
2118+
# PRE-requisites
2119+
#
21102120
assert app_settings.AUTOSCALING_EC2_INSTANCES
21112121
assert app_settings.AUTOSCALING_EC2_INSTANCES.EC2_INSTANCES_MACHINES_BUFFER > 0
2122+
num_hot_buffer = (
2123+
app_settings.AUTOSCALING_EC2_INSTANCES.EC2_INSTANCES_MACHINES_BUFFER
2124+
)
21122125

21132126
# we have nothing running now
21142127
all_instances = await ec2_client.describe_instances()
@@ -2121,78 +2134,95 @@ async def test_warm_buffers_only_replace_hot_buffer_if_service_is_started_issue7
21212134
await assert_autoscaled_dynamic_ec2_instances(
21222135
ec2_client,
21232136
expected_num_reservations=1,
2124-
expected_num_instances=app_settings.AUTOSCALING_EC2_INSTANCES.EC2_INSTANCES_MACHINES_BUFFER,
2125-
expected_instance_type=cast(
2126-
InstanceTypeType,
2127-
next(
2128-
iter(app_settings.AUTOSCALING_EC2_INSTANCES.EC2_INSTANCES_ALLOWED_TYPES)
2129-
),
2130-
),
2137+
expected_num_instances=num_hot_buffer,
2138+
expected_instance_type=hot_buffer_instance_type,
21312139
expected_instance_state="running",
21322140
expected_additional_tag_keys=list(ec2_instance_custom_tags),
21332141
instance_filters=instance_type_filters,
21342142
)
2135-
# calling again should attach the new nodes to the reserve, but nothing should start
2143+
# this brings a new analysis
21362144
await auto_scale_cluster(
21372145
app=initialized_app, auto_scaling_mode=DynamicAutoscaling()
21382146
)
2147+
spied_cluster = assert_cluster_state(
2148+
spied_cluster_analysis, expected_calls=2, expected_num_machines=5
2149+
)
2150+
# calling again should attach the new nodes to the reserve, but nothing should start
2151+
fake_attached_node_base = deepcopy(fake_node)
2152+
assert fake_attached_node_base.spec
2153+
fake_attached_node_base.spec.availability = (
2154+
Availability.active if with_drain_nodes_labelled else Availability.drain
2155+
)
2156+
assert fake_attached_node_base.spec.labels
2157+
assert app_settings.AUTOSCALING_NODES_MONITORING
2158+
expected_docker_node_tags = {
2159+
tag_key: "true"
2160+
for tag_key in (
2161+
app_settings.AUTOSCALING_NODES_MONITORING.NODES_MONITORING_NODE_LABELS
2162+
+ app_settings.AUTOSCALING_NODES_MONITORING.NODES_MONITORING_NEW_NODES_LABELS
2163+
)
2164+
} | {
2165+
DOCKER_TASK_EC2_INSTANCE_TYPE_PLACEMENT_CONSTRAINT_KEY: f"{hot_buffer_instance_type}"
2166+
}
2167+
fake_attached_node_base.spec.labels |= expected_docker_node_tags | {
2168+
_OSPARC_SERVICE_READY_LABEL_KEY: "true"
2169+
}
2170+
fake_hot_buffer_nodes = []
2171+
for i in range(num_hot_buffer):
2172+
node = fake_attached_node_base.model_copy(deep=True)
2173+
assert node.description
2174+
node.description.hostname = node_host_name_from_ec2_private_dns(
2175+
spied_cluster.pending_ec2s[i].ec2_instance
2176+
)
2177+
fake_hot_buffer_nodes.append(node)
2178+
auto_scaling_mode = DynamicAutoscaling()
2179+
mocker.patch.object(
2180+
auto_scaling_mode,
2181+
"get_monitored_nodes",
2182+
autospec=True,
2183+
return_value=fake_hot_buffer_nodes,
2184+
)
2185+
await auto_scale_cluster(app=initialized_app, auto_scaling_mode=auto_scaling_mode)
21392186
await assert_autoscaled_dynamic_ec2_instances(
21402187
ec2_client,
21412188
expected_num_reservations=1,
2142-
expected_num_instances=app_settings.AUTOSCALING_EC2_INSTANCES.EC2_INSTANCES_MACHINES_BUFFER,
2143-
expected_instance_type=cast(
2144-
InstanceTypeType,
2145-
next(
2146-
iter(app_settings.AUTOSCALING_EC2_INSTANCES.EC2_INSTANCES_ALLOWED_TYPES)
2147-
),
2148-
),
2189+
expected_num_instances=num_hot_buffer,
2190+
expected_instance_type=hot_buffer_instance_type,
21492191
expected_instance_state="running",
21502192
expected_additional_tag_keys=list(ec2_instance_custom_tags),
21512193
instance_filters=instance_type_filters,
21522194
)
2195+
spied_cluster = assert_cluster_state(
2196+
spied_cluster_analysis, expected_calls=1, expected_num_machines=5
2197+
)
2198+
assert len(spied_cluster.buffer_drained_nodes) == num_hot_buffer
2199+
assert not spied_cluster.buffer_ec2s
21532200

21542201
# have a few warm buffers ready with the same type as the hot buffer machines
2155-
buffer_machines = await create_buffer_machines(
2202+
await create_buffer_machines(
21562203
buffer_count,
2157-
cast(
2158-
InstanceTypeType,
2159-
next(
2160-
iter(app_settings.AUTOSCALING_EC2_INSTANCES.EC2_INSTANCES_ALLOWED_TYPES)
2161-
),
2162-
),
2204+
hot_buffer_instance_type,
21632205
"stopped",
21642206
None,
21652207
)
21662208
await assert_autoscaled_dynamic_warm_pools_ec2_instances(
21672209
ec2_client,
21682210
expected_num_reservations=1,
21692211
expected_num_instances=buffer_count,
2170-
expected_instance_type=cast(
2171-
InstanceTypeType,
2172-
next(
2173-
iter(app_settings.AUTOSCALING_EC2_INSTANCES.EC2_INSTANCES_ALLOWED_TYPES)
2174-
),
2175-
),
2212+
expected_instance_type=hot_buffer_instance_type,
21762213
expected_instance_state="stopped",
21772214
expected_additional_tag_keys=list(ec2_instance_custom_tags),
21782215
expected_pre_pulled_images=None,
21792216
instance_filters=stopped_instance_type_filters,
21802217
)
21812218

21822219
# calling again should do nothing
2183-
await auto_scale_cluster(
2184-
app=initialized_app, auto_scaling_mode=DynamicAutoscaling()
2185-
)
2220+
await auto_scale_cluster(app=initialized_app, auto_scaling_mode=auto_scaling_mode)
21862221
await assert_autoscaled_dynamic_ec2_instances(
21872222
ec2_client,
21882223
expected_num_reservations=1,
2189-
expected_num_instances=app_settings.AUTOSCALING_EC2_INSTANCES.EC2_INSTANCES_MACHINES_BUFFER,
2190-
expected_instance_type=cast(
2191-
InstanceTypeType,
2192-
next(
2193-
iter(app_settings.AUTOSCALING_EC2_INSTANCES.EC2_INSTANCES_ALLOWED_TYPES)
2194-
),
2195-
),
2224+
expected_num_instances=num_hot_buffer,
2225+
expected_instance_type=hot_buffer_instance_type,
21962226
expected_instance_state="running",
21972227
expected_additional_tag_keys=list(ec2_instance_custom_tags),
21982228
instance_filters=instance_type_filters,
@@ -2201,12 +2231,73 @@ async def test_warm_buffers_only_replace_hot_buffer_if_service_is_started_issue7
22012231
ec2_client,
22022232
expected_num_reservations=1,
22032233
expected_num_instances=buffer_count,
2204-
expected_instance_type=cast(
2205-
InstanceTypeType,
2206-
next(
2207-
iter(app_settings.AUTOSCALING_EC2_INSTANCES.EC2_INSTANCES_ALLOWED_TYPES)
2208-
),
2234+
expected_instance_type=hot_buffer_instance_type,
2235+
expected_instance_state="stopped",
2236+
expected_additional_tag_keys=list(ec2_instance_custom_tags),
2237+
expected_pre_pulled_images=None,
2238+
instance_filters=stopped_instance_type_filters,
2239+
)
2240+
spied_cluster = assert_cluster_state(
2241+
spied_cluster_analysis, expected_calls=1, expected_num_machines=5
2242+
)
2243+
assert len(spied_cluster.buffer_drained_nodes) == num_hot_buffer
2244+
assert len(spied_cluster.buffer_ec2s) == buffer_count
2245+
2246+
#
2247+
# BUG REPRODUCTION
2248+
#
2249+
# start a service that imposes same type as the hot buffer
2250+
assert (
2251+
hot_buffer_instance_type == "t2.xlarge"
2252+
), "the test is hard-coded for this type and accordingly resource. If this changed then the resource shall be changed too"
2253+
scale_up_params = _ScaleUpParams(
2254+
imposed_instance_type=hot_buffer_instance_type,
2255+
service_resources=Resources(
2256+
cpus=2, ram=TypeAdapter(ByteSize).validate_python("1Gib")
22092257
),
2258+
num_services=1,
2259+
expected_instance_type="t2.xlarge",
2260+
expected_num_instances=1,
2261+
)
2262+
await create_services_batch(scale_up_params)
2263+
# this should trigger usage of the hot buffer and the warm buffers should remain stopped
2264+
await auto_scale_cluster(app=initialized_app, auto_scaling_mode=auto_scaling_mode)
2265+
await assert_autoscaled_dynamic_ec2_instances(
2266+
ec2_client,
2267+
expected_num_reservations=1,
2268+
expected_num_instances=num_hot_buffer,
2269+
expected_instance_type=hot_buffer_instance_type,
2270+
expected_instance_state="running",
2271+
expected_additional_tag_keys=list(ec2_instance_custom_tags),
2272+
instance_filters=instance_type_filters,
2273+
)
2274+
await assert_autoscaled_dynamic_warm_pools_ec2_instances(
2275+
ec2_client,
2276+
expected_num_reservations=1,
2277+
expected_num_instances=buffer_count,
2278+
expected_instance_type=hot_buffer_instance_type,
2279+
expected_instance_state="stopped",
2280+
expected_additional_tag_keys=list(ec2_instance_custom_tags),
2281+
expected_pre_pulled_images=None,
2282+
instance_filters=stopped_instance_type_filters,
2283+
)
2284+
2285+
# this should trigger replacement TAG_PREFIX: staging-hotfix-github of the hot buffer by 1 warm buffer
2286+
await auto_scale_cluster(app=initialized_app, auto_scaling_mode=auto_scaling_mode)
2287+
await assert_autoscaled_dynamic_ec2_instances(
2288+
ec2_client,
2289+
expected_num_reservations=1,
2290+
expected_num_instances=num_hot_buffer + 1,
2291+
expected_instance_type=hot_buffer_instance_type,
2292+
expected_instance_state="running",
2293+
expected_additional_tag_keys=list(ec2_instance_custom_tags),
2294+
instance_filters=instance_type_filters,
2295+
)
2296+
await assert_autoscaled_dynamic_warm_pools_ec2_instances(
2297+
ec2_client,
2298+
expected_num_reservations=1,
2299+
expected_num_instances=buffer_count - 1,
2300+
expected_instance_type=hot_buffer_instance_type,
22102301
expected_instance_state="stopped",
22112302
expected_additional_tag_keys=list(ec2_instance_custom_tags),
22122303
expected_pre_pulled_images=None,

0 commit comments

Comments
 (0)