Merge branch 'azure-provisioner' of github.com:skypilot-org/skypilot …

…into azure-provisioner

sky/provision/azure/instance.py

@@ -4,6 +4,7 @@
 import json
 import logging
 from multiprocessing import pool
+import os
 import pathlib
 import time
 import typing
@@ -16,6 +17,7 @@
 from sky.adaptors import azure
 from sky.provision import common
 from sky.provision import constants
+from sky.skylet import constants as skylet_constants
 from sky.utils import common_utils
 from sky.utils import ux_utils
 
@@ -503,6 +505,15 @@ def get_cluster_info(
         provider_config: Optional[Dict[str, Any]] = None) -> common.ClusterInfo:
     """See sky/provision/__init__.py"""
     del region
+    # Add the executable path to the PATH environment variable to make sure, az
+    # command is available in the subprocess. Thisis useful for a controller to
+    # query statuses of old Azure instances that was provisioned with ray
+    # autoscaler.
+    os.environ['PATH'] = os.pathsep.join([
+        os.environ.get('PATH', ''),
+        os.path.expanduser(
+            os.path.join(skylet_constants.SKY_REMOTE_PYTHON_ENV, 'bin'))
+    ])
     filters = {constants.TAG_RAY_CLUSTER_NAME: cluster_name_on_cloud}
     assert provider_config is not None, (cluster_name_on_cloud, provider_config)
     resource_group = provider_config['resource_group']

sky/utils/controller_utils.py

@@ -238,7 +238,8 @@ def _get_cloud_dependencies_installation_commands(
                 '! command -v curl &> /dev/null || '
                 '! command -v socat &> /dev/null || '
                 '! command -v netcat &> /dev/null; '
-                'then apt update && apt install curl socat netcat -y; '
+                'then apt update && apt install curl socat netcat -y '
+                '&> /dev/null; '
                 'fi" && '
                 # Install kubectl
                 '(command -v kubectl &>/dev/null || '
@@ -253,7 +254,7 @@ def _get_cloud_dependencies_installation_commands(
                 'pip list | grep cudo-compute > /dev/null 2>&1 || '
                 'pip install "cudo-compute>=0.1.10" > /dev/null 2>&1 && '
                 'wget https://download.cudo.org/compute/cudoctl-0.3.2-amd64.deb -O ~/cudoctl.deb > /dev/null 2>&1 && '  # pylint: disable=line-too-long
-                'sudo dpkg -i ~/cudoctl.deb > /dev/null 2>&1')
+                'sudo dpkg -i ~/cudoctl.deb 2>&1 > /dev/null')
         if controller == Controllers.JOBS_CONTROLLER:
             if isinstance(cloud, clouds.IBM):
                 commands.append(

tests/backward_compatibility_tests.sh

@@ -168,7 +168,7 @@ if [ "$start_from" -le 7 ]; then
 conda activate sky-back-compat-master
 rm -r  ~/.sky/wheels || true
 sky spot launch -d --cloud ${CLOUD} -y --cpus 2 --num-nodes 2 -n ${MANAGED_JOB_JOB_NAME}-7-0 "echo hi; sleep 1000"
-sky spot launch -d --cloud ${CLOUD} -y --cpus 2 --num-nodes 2 -n ${MANAGED_JOB_JOB_NAME}-7-1 "echo hi; sleep 300"
+sky spot launch -d --cloud ${CLOUD} -y --cpus 2 --num-nodes 2 -n ${MANAGED_JOB_JOB_NAME}-7-1 "echo hi; sleep 400"
 conda activate sky-back-compat-current
 rm -r  ~/.sky/wheels || true
 s=$(sky jobs queue | grep ${MANAGED_JOB_JOB_NAME}-7 | grep "RUNNING" | wc -l)
@@ -183,7 +183,7 @@ s=$(sky jobs queue | grep ${MANAGED_JOB_JOB_NAME}-7)
 echo "$s"
 echo "$s" | grep "RUNNING" | wc -l | grep 3 || exit 1
 sky jobs cancel -y -n ${MANAGED_JOB_JOB_NAME}-7-0
-sky jobs logs -n "${MANAGED_JOB_JOB_NAME}-7-1"
+sky jobs logs -n "${MANAGED_JOB_JOB_NAME}-7-1" || exit 1
 s=$(sky jobs queue | grep ${MANAGED_JOB_JOB_NAME}-7)
 echo "$s"
 echo "$s" | grep "SUCCEEDED" | wc -l | grep 2 || exit 1