[serve] [release] Disable smoke test by default (#23334)

release/serve_tests/workloads/autoscaling_multi_deployment.py

@@ -28,7 +28,6 @@ Report:
 
 import click
 import math
-import os
 import random
 
 import ray
@@ -38,6 +37,7 @@ from serve_test_utils import (
     aggregate_all_metrics,
     run_wrk_on_all_nodes,
     save_test_results,
+    is_smoke_test,
 )
 from serve_test_cluster_utils import (
     setup_local_single_node_cluster,
@@ -130,8 +130,7 @@ def main(
     # Give default cluster parameter values based on smoke_test config
     # if user provided values explicitly, use them instead.
     # IS_SMOKE_TEST is set by args of releaser's e2e.py
-    smoke_test = os.environ.get("IS_SMOKE_TEST", "1")
-    if smoke_test == "1":
+    if is_smoke_test():
         min_replicas = min_replicas or DEFAULT_SMOKE_TEST_MIN_NUM_REPLICA
         max_replicas = max_replicas or DEFAULT_SMOKE_TEST_MAX_NUM_REPLICA
         num_deployments = num_deployments or DEFAULT_SMOKE_TEST_NUM_DEPLOYMENTS

release/serve_tests/workloads/autoscaling_single_deployment.py

@@ -36,6 +36,7 @@ from serve_test_utils import (
     aggregate_all_metrics,
     run_wrk_on_all_nodes,
     save_test_results,
+    is_smoke_test,
 )
 from serve_test_cluster_utils import (
     setup_local_single_node_cluster,
@@ -106,8 +107,7 @@ def main(
     # Give default cluster parameter values based on smoke_test config
     # if user provided values explicitly, use them instead.
     # IS_SMOKE_TEST is set by args of releaser's e2e.py
-    smoke_test = os.environ.get("IS_SMOKE_TEST", "1")
-    if smoke_test == "1":
+    if is_smoke_test():
         min_replicas = min_replicas or DEFAULT_SMOKE_TEST_MIN_NUM_REPLICA
         max_replicas = max_replicas or DEFAULT_SMOKE_TEST_MAX_NUM_REPLICA
         trial_length = trial_length or DEFAULT_SMOKE_TEST_TRIAL_LENGTH

release/serve_tests/workloads/multi_deployment_1k_noop_replica.py

@@ -27,7 +27,6 @@ Report:
 
 import click
 import math
-import os
 import random
 
 from ray import serve
@@ -36,6 +35,7 @@ from serve_test_utils import (
     aggregate_all_metrics,
     run_wrk_on_all_nodes,
     save_test_results,
+    is_smoke_test,
 )
 from serve_test_cluster_utils import (
     setup_local_single_node_cluster,
@@ -115,8 +115,7 @@ def main(
     # Give default cluster parameter values based on smoke_test config
     # if user provided values explicitly, use them instead.
     # IS_SMOKE_TEST is set by args of releaser's e2e.py
-    smoke_test = os.environ.get("IS_SMOKE_TEST", "1")
-    if smoke_test == "1":
+    if is_smoke_test():
         num_replicas = num_replicas or DEFAULT_SMOKE_TEST_NUM_REPLICA
         num_deployments = num_deployments or DEFAULT_SMOKE_TEST_NUM_DEPLOYMENTS
         trial_length = trial_length or DEFAULT_SMOKE_TEST_TRIAL_LENGTH

release/serve_tests/workloads/serve_cluster_fault_tolerance.py

@@ -11,12 +11,15 @@ import click
 import time
 import requests
 import uuid
-import os
+
 from pathlib import Path
 
 from serve_test_cluster_utils import setup_local_single_node_cluster
 
-from serve_test_utils import save_test_results
+from serve_test_utils import (
+    save_test_results,
+    is_smoke_test,
+)
 
 import ray
 from ray import serve
@@ -47,8 +50,7 @@ def main():
     namespace = uuid.uuid4().hex
 
     # IS_SMOKE_TEST is set by args of releaser's e2e.py
-    smoke_test = os.environ.get("IS_SMOKE_TEST", "1")
-    if smoke_test == "1":
+    if is_smoke_test():
         path = Path("checkpoint.db")
         checkpoint_path = f"file://{path}"
         if path.exists():

release/serve_tests/workloads/serve_cluster_fault_tolerance_gcs.py

@@ -7,7 +7,7 @@ it's Job Manager's responsibility, and only re-deploy to the same cluster
 with remote checkpoint.
 """
 
-import os
+
 import time
 import uuid
 
@@ -15,7 +15,10 @@ import click
 import requests
 from ray.serve.utils import logger
 from serve_test_cluster_utils import setup_local_single_node_cluster
-from serve_test_utils import save_test_results
+from serve_test_utils import (
+    save_test_results,
+    is_smoke_test,
+)
 
 import ray
 from ray import serve
@@ -45,8 +48,7 @@ def main():
     namespace = uuid.uuid4().hex
 
     # IS_SMOKE_TEST is set by args of releaser's e2e.py
-    smoke_test = os.environ.get("IS_SMOKE_TEST", "1")
-    if smoke_test == "1":
+    if is_smoke_test():
         checkpoint_path = "file://checkpoint.db"
     else:
         checkpoint_path = (

release/serve_tests/workloads/serve_micro_benchmark.py

@@ -1,5 +1,5 @@
 import asyncio
-import os
+
 
 from ray.serve.benchmarks.microbenchmark import main as benchmark_main
 from ray.serve.utils import logger
@@ -9,6 +9,7 @@ from serve_test_cluster_utils import (
 )
 from serve_test_utils import (
     save_test_results,
+    is_smoke_test,
 )
 
 
@@ -16,8 +17,7 @@ async def main():
     # Give default cluster parameter values based on smoke_test config
     # if user provided values explicitly, use them instead.
     # IS_SMOKE_TEST is set by args of releaser's e2e.py
-    smoke_test = os.environ.get("IS_SMOKE_TEST", "1")
-    if smoke_test == "1":
+    if is_smoke_test():
         setup_local_single_node_cluster(1)
     else:
         setup_anyscale_cluster()

release/serve_tests/workloads/serve_test_utils.py

@@ -13,6 +13,10 @@ from subprocess import PIPE
 from typing import Dict, List, Union
 
 
+def is_smoke_test():
+    return os.environ.get("IS_SMOKE_TEST", "0") == "1"
+
+
 def parse_time_to_ms(time_string: str) -> float:
     """Given a time string with various unit, convert
     to ms in float:

release/serve_tests/workloads/single_deployment_1k_noop_replica.py

@@ -35,6 +35,7 @@ from serve_test_utils import (
     aggregate_all_metrics,
     run_wrk_on_all_nodes,
     save_test_results,
+    is_smoke_test,
 )
 from serve_test_cluster_utils import (
     setup_local_single_node_cluster,
@@ -92,8 +93,7 @@ def main(
     # Give default cluster parameter values based on smoke_test config
     # if user provided values explicitly, use them instead.
     # IS_SMOKE_TEST is set by args of releaser's e2e.py
-    smoke_test = os.environ.get("IS_SMOKE_TEST", "1")
-    if smoke_test == "1":
+    if is_smoke_test():
         num_replicas = num_replicas or DEFAULT_SMOKE_TEST_NUM_REPLICA
         trial_length = trial_length or DEFAULT_SMOKE_TEST_TRIAL_LENGTH
         logger.info(f"Running local / smoke test with {num_replicas} replicas ..\n")