[Core] Deactivate SkyPilot runtime env for jobs (#3639)

* [Core] Deactivate SkyPilot runtime env for jobs

* Add smoke test for python path

* Update sky/skylet/log_lib.py

Co-authored-by: Zongheng Yang <[email protected]>

* fix

---------

Co-authored-by: Zongheng Yang <[email protected]>

sky/skylet/constants.py

@@ -49,6 +49,11 @@
 SKY_REMOTE_PYTHON_ENV_NAME = 'skypilot-runtime'
 SKY_REMOTE_PYTHON_ENV = f'~/{SKY_REMOTE_PYTHON_ENV_NAME}'
 ACTIVATE_SKY_REMOTE_PYTHON_ENV = f'source {SKY_REMOTE_PYTHON_ENV}/bin/activate'
+# Deleting the SKY_REMOTE_PYTHON_ENV_NAME from the PATH to deactivate the
+# environment. `deactivate` command does not work when conda is used.
+DEACTIVATE_SKY_REMOTE_PYTHON_ENV = (
+    'export PATH='
+    f'$(echo $PATH | sed "s|$(echo ~)/{SKY_REMOTE_PYTHON_ENV_NAME}/bin:||")')
 
 # The name for the environment variable that stores the unique ID of the
 # current task. This will stay the same across multiple recoveries of the

sky/skylet/log_lib.py

@@ -263,6 +263,9 @@ def make_task_bash_script(codegen: str,
     # set -a is used for exporting all variables functions to the environment
     # so that bash `user_script` can access `conda activate`. Detail: #436.
     # Reference: https://www.gnu.org/software/bash/manual/html_node/The-Set-Builtin.html # pylint: disable=line-too-long
+    # DEACTIVATE_SKY_REMOTE_PYTHON_ENV: Deactivate the SkyPilot runtime env, as
+    # the ray cluster is started within the runtime env, which may cause the
+    # user program to run in that env as well.
     # PYTHONUNBUFFERED is used to disable python output buffering.
     script = [
         textwrap.dedent(f"""\
@@ -271,6 +274,7 @@ def make_task_bash_script(codegen: str,
             set -a
             . $(conda info --base 2> /dev/null)/etc/profile.d/conda.sh > /dev/null 2>&1 || true
             set +a
+            {constants.DEACTIVATE_SKY_REMOTE_PYTHON_ENV}
             export PYTHONUNBUFFERED=1
             cd {constants.SKY_REMOTE_WORKDIR}"""),
     ]

tests/test_smoke.py

@@ -55,6 +55,7 @@
 from sky.data import data_utils
 from sky.data import storage as storage_lib
 from sky.data.data_utils import Rclone
+from sky.skylet import constants
 from sky.skylet import events
 from sky.utils import common_utils
 from sky.utils import resources_utils
@@ -362,6 +363,9 @@ def test_aws_region():
             f'sky status --all | grep {name} | grep us-east-2',  # Ensure the region is correct.
             f'sky exec {name} \'echo $SKYPILOT_CLUSTER_INFO | jq .region | grep us-east-2\'',
             f'sky logs {name} 2 --status',  # Ensure the job succeeded.
+            # A user program should not access SkyPilot runtime env python by default.
+            f'sky exec {name} \'which python | grep {constants.SKY_REMOTE_PYTHON_ENV_NAME} || exit 1\'',
+            f'sky logs {name} 3 --status',  # Ensure the job succeeded.
         ],
         f'sky down -y {name}',
     )
@@ -382,6 +386,9 @@ def test_gcp_region_and_service_account():
             f'sky status --all | grep {name} | grep us-central1',  # Ensure the region is correct.
             f'sky exec {name} \'echo $SKYPILOT_CLUSTER_INFO | jq .region | grep us-central1\'',
             f'sky logs {name} 3 --status',  # Ensure the job succeeded.
+            # A user program should not access SkyPilot runtime env python by default.
+            f'sky exec {name} \'which python | grep {constants.SKY_REMOTE_PYTHON_ENV_NAME} || exit 1\'',
+            f'sky logs {name} 4 --status',  # Ensure the job succeeded.
         ],
         f'sky down -y {name}',
     )
@@ -419,6 +426,9 @@ def test_azure_region():
             f'sky logs {name} 2 --status',  # Ensure the job succeeded.
             f'sky exec {name} \'echo $SKYPILOT_CLUSTER_INFO | jq .zone | grep null\'',
             f'sky logs {name} 3 --status',  # Ensure the job succeeded.
+            # A user program should not access SkyPilot runtime env python by default.
+            f'sky exec {name} \'which python | grep {constants.SKY_REMOTE_PYTHON_ENV_NAME} || exit 1\'',
+            f'sky logs {name} 4 --status',  # Ensure the job succeeded.
         ],
         f'sky down -y {name}',
     )