Change added USER_HASH in models name to full digits (8) instead of half (4) to prevent potential hash conflict (#4592)

* test: print

* feat: use 8 digits for user hash

* test: full user hash

* test: the content is shorten due to length limit

* feat: convert user hash from hex to base36

* chore: extend support lifetime of relied code

* fix: after merge, re-delete extra user hash splition in smoke test

* style: format and type hints

* Revert "feat: convert user hash from hex to base36"

This reverts commit 2f7ca14.

* Update tests/unit_tests/test_common_utils.py

---------

Co-authored-by: Zhanghao Wu <[email protected]>

Author: andylizf

sky/backends/backend_utils.py

@@ -997,7 +997,7 @@ def write_cluster_config(
 
     # Read the cluster name from the tmp yaml file, to take the backward
     # compatbility restortion above into account.
-    # TODO: remove this after 2 minor releases, 0.8.0.
+    # TODO: remove this after 2 minor releases, 0.10.0.
     yaml_config = common_utils.read_yaml(tmp_yaml_path)
     config_dict['cluster_name_on_cloud'] = yaml_config['cluster_name']
 

sky/utils/common_utils.py

@@ -28,7 +28,6 @@
 
 _USER_HASH_FILE = os.path.expanduser('~/.sky/user_hash')
 USER_HASH_LENGTH = 8
-USER_HASH_LENGTH_IN_CLUSTER_NAME = 4
 
 # We are using base36 to reduce the length of the hash. 2 chars -> 36^2 = 1296
 # possibilities. considering the final cluster name contains the prefix as well,
@@ -182,7 +181,7 @@ def make_cluster_name_on_cloud(display_name: str,
             f'on the cloud, we convert it to {cluster_name_on_cloud}.')
     user_hash = ''
     if add_user_hash:
-        user_hash = get_user_hash()[:USER_HASH_LENGTH_IN_CLUSTER_NAME]
+        user_hash = get_user_hash()
         user_hash = f'-{user_hash}'
     user_hash_length = len(user_hash)
 

tests/smoke_tests/smoke_tests_utils.py

@@ -4,7 +4,7 @@
 import subprocess
 import sys
 import tempfile
-from typing import Dict, List, NamedTuple, Optional, Tuple
+from typing import Dict, List, NamedTuple, Optional, Sequence
 import uuid
 
 import colorama
@@ -32,7 +32,7 @@
 
 STORAGE_SETUP_COMMANDS = [
     'touch ~/tmpfile', 'mkdir -p ~/tmp-workdir',
-    'touch ~/tmp-workdir/tmp\ file', 'touch ~/tmp-workdir/tmp\ file2',
+    r'touch ~/tmp-workdir/tmp\ file', r'touch ~/tmp-workdir/tmp\ file2',
     'touch ~/tmp-workdir/foo',
     '[ ! -e ~/tmp-workdir/circle-link ] && ln -s ~/tmp-workdir/ ~/tmp-workdir/circle-link || true',
     'touch ~/.ssh/id_rsa.pub'
@@ -55,7 +55,7 @@
     [status.value for status in sky.ManagedJobStatus])
 
 
-def _statuses_to_str(statuses: List[enum.Enum]):
+def _statuses_to_str(statuses: Sequence[enum.Enum]):
     """Convert a list of enums to a string with all the values separated by |."""
     assert len(statuses) > 0, 'statuses must not be empty'
     if len(statuses) > 1:
@@ -74,8 +74,8 @@ def _statuses_to_str(statuses: List[enum.Enum]):
     'fi; '
     'current_status=$(sky status {cluster_name} --refresh | '
     'awk "/^{cluster_name}/ '
-    '{{for (i=1; i<=NF; i++) if (\$i ~ /^(' + _ALL_CLUSTER_STATUSES +
-    ')$/) print \$i}}"); '
+    r'{{for (i=1; i<=NF; i++) if (\$i ~ /^(' + _ALL_CLUSTER_STATUSES +
+    r')$/) print \$i}}"); '
     'if [[ "$current_status" =~ {cluster_status} ]]; '
     'then echo "Target cluster status {cluster_status} reached."; break; fi; '
     'echo "Waiting for cluster status to become {cluster_status}, current status: $current_status"; '
@@ -136,8 +136,8 @@ def get_cmd_wait_until_cluster_is_not_found(cluster_name: str, timeout: int):
     'fi; '
     'current_status=$(sky queue {cluster_name} | '
     'awk "\\$1 == \\"{job_id}\\" '
-    '{{for (i=1; i<=NF; i++) if (\$i ~ /^(' + _ALL_JOB_STATUSES +
-    ')$/) print \$i}}"); '
+    r'{{for (i=1; i<=NF; i++) if (\$i ~ /^(' + _ALL_JOB_STATUSES +
+    r')$/) print \$i}}"); '
     'found=0; '  # Initialize found variable outside the loop
     'while read -r line; do '  # Read line by line
     '  if [[ "$line" =~ {job_status} ]]; then '  # Check each line
@@ -196,7 +196,8 @@ def get_cmd_wait_until_job_status_contains_matching_job_name(
 
 
 def get_cmd_wait_until_managed_job_status_contains_matching_job_name(
-        job_name: str, job_status: List[sky.JobStatus], timeout: int):
+        job_name: str, job_status: Sequence[sky.ManagedJobStatus],
+        timeout: int):
     return _WAIT_UNTIL_MANAGED_JOB_STATUS_CONTAINS_MATCHING_JOB_NAME.format(
         job_name=job_name,
         job_status=_statuses_to_str(job_status),
@@ -215,7 +216,7 @@ class Test(NamedTuple):
     # Timeout for each command in seconds.
     timeout: int = DEFAULT_CMD_TIMEOUT
     # Environment variables to set for each command.
-    env: Dict[str, str] = None
+    env: Optional[Dict[str, str]] = None
 
     def echo(self, message: str):
         # pytest's xdist plugin captures stdout; print to stderr so that the
@@ -254,7 +255,7 @@ def terminate_gcp_replica(name: str, zone: str, replica_id: int) -> str:
             f' --quiet $({query_cmd})')
 
 
-def run_one_test(test: Test) -> Tuple[int, str, str]:
+def run_one_test(test: Test) -> None:
     # Fail fast if `sky` CLI somehow errors out.
     subprocess.run(['sky', 'status'], stdout=subprocess.DEVNULL, check=True)
     log_to_stdout = os.environ.get('LOG_TO_STDOUT', None)

tests/smoke_tests/test_managed_job.py

@@ -77,7 +77,7 @@ def test_managed_jobs_basic(generic_cloud: str):
             # Test the functionality for logging.
             f's=$(sky jobs logs -n {name}-2 --no-follow); echo "$s"; echo "$s" | grep "start counting"',
             f's=$(sky jobs logs --controller -n {name}-2 --no-follow); echo "$s"; echo "$s" | grep "Cluster launched:"',
-            f'{smoke_tests_utils.GET_JOB_QUEUE} | grep {name}-2 | head -n1 | grep "RUNNING\|SUCCEEDED"',
+            rf'{smoke_tests_utils.GET_JOB_QUEUE} | grep {name}-2 | head -n1 | grep "RUNNING\|SUCCEEDED"',
         ],
         # TODO(zhwu): Change to f'sky jobs cancel -y -n {name}-1 -n {name}-2' when
         # canceling multiple job names is supported.
@@ -105,19 +105,19 @@ def test_job_pipeline(generic_cloud: str):
         [
             f'sky jobs launch -n {name} tests/test_yamls/pipeline.yaml --cloud {generic_cloud} -y -d',
             'sleep 5',
-            f'{smoke_tests_utils.GET_JOB_QUEUE} | grep {name} | head -n1 | grep "STARTING\|RUNNING"',
+            rf'{smoke_tests_utils.GET_JOB_QUEUE} | grep {name} | head -n1 | grep "STARTING\|RUNNING"',
             # `grep -A 4 {name}` finds the job with {name} and the 4 lines
             # after it, i.e. the 4 tasks within the job.
             # `sed -n 2p` gets the second line of the 4 lines, i.e. the first
             # task within the job.
-            f'{smoke_tests_utils.GET_JOB_QUEUE} | grep -A 4 {name}| sed -n 2p | grep "STARTING\|RUNNING"',
+            rf'{smoke_tests_utils.GET_JOB_QUEUE} | grep -A 4 {name}| sed -n 2p | grep "STARTING\|RUNNING"',
             f'{smoke_tests_utils.GET_JOB_QUEUE} | grep -A 4 {name}| sed -n 3p | grep "PENDING"',
             f'sky jobs cancel -y -n {name}',
             'sleep 5',
-            f'{smoke_tests_utils.GET_JOB_QUEUE} | grep -A 4 {name}| sed -n 2p | grep "CANCELLING\|CANCELLED"',
-            f'{smoke_tests_utils.GET_JOB_QUEUE} | grep -A 4 {name}| sed -n 3p | grep "CANCELLING\|CANCELLED"',
-            f'{smoke_tests_utils.GET_JOB_QUEUE} | grep -A 4 {name}| sed -n 4p | grep "CANCELLING\|CANCELLED"',
-            f'{smoke_tests_utils.GET_JOB_QUEUE} | grep -A 4 {name}| sed -n 5p | grep "CANCELLING\|CANCELLED"',
+            rf'{smoke_tests_utils.GET_JOB_QUEUE} | grep -A 4 {name}| sed -n 2p | grep "CANCELLING\|CANCELLED"',
+            rf'{smoke_tests_utils.GET_JOB_QUEUE} | grep -A 4 {name}| sed -n 3p | grep "CANCELLING\|CANCELLED"',
+            rf'{smoke_tests_utils.GET_JOB_QUEUE} | grep -A 4 {name}| sed -n 4p | grep "CANCELLING\|CANCELLED"',
+            rf'{smoke_tests_utils.GET_JOB_QUEUE} | grep -A 4 {name}| sed -n 5p | grep "CANCELLING\|CANCELLED"',
             'sleep 200',
             f'{smoke_tests_utils.GET_JOB_QUEUE} | grep -A 4 {name}| sed -n 2p | grep "CANCELLED"',
             f'{smoke_tests_utils.GET_JOB_QUEUE} | grep -A 4 {name}| sed -n 3p | grep "CANCELLED"',
@@ -212,7 +212,7 @@ def test_managed_jobs_recovery_aws(aws_config_region):
     test = smoke_tests_utils.Test(
         'managed_jobs_recovery_aws',
         [
-            f'sky jobs launch --cloud aws --region {region} --use-spot -n {name} "echo SKYPILOT_TASK_ID: \$SKYPILOT_TASK_ID; sleep 1800"  -y -d',
+            rf'sky jobs launch --cloud aws --region {region} --use-spot -n {name} "echo SKYPILOT_TASK_ID: \$SKYPILOT_TASK_ID; sleep 1800" -y -d',
             smoke_tests_utils.
             get_cmd_wait_until_managed_job_status_contains_matching_job_name(
                 job_name=name,
@@ -258,7 +258,7 @@ def test_managed_jobs_recovery_gcp():
     test = smoke_tests_utils.Test(
         'managed_jobs_recovery_gcp',
         [
-            f'sky jobs launch --cloud gcp --zone {zone} -n {name} --use-spot --cpus 2 "echo SKYPILOT_TASK_ID: \$SKYPILOT_TASK_ID; sleep 1800"  -y -d',
+            rf'sky jobs launch --cloud gcp --zone {zone} -n {name} --use-spot --cpus 2 "echo SKYPILOT_TASK_ID: \$SKYPILOT_TASK_ID; sleep 1800" -y -d',
             smoke_tests_utils.
             get_cmd_wait_until_managed_job_status_contains_matching_job_name(
                 job_name=name,
@@ -288,14 +288,13 @@ def test_managed_jobs_pipeline_recovery_aws(aws_config_region):
     """Test managed job recovery for a pipeline."""
     name = smoke_tests_utils.get_cluster_name()
     user_hash = common_utils.get_user_hash()
-    user_hash = user_hash[:common_utils.USER_HASH_LENGTH_IN_CLUSTER_NAME]
     region = aws_config_region
     if region != 'us-east-2':
         pytest.skip('Only run spot pipeline recovery test in us-east-2')
     test = smoke_tests_utils.Test(
         'managed_jobs_pipeline_recovery_aws',
         [
-            f'sky jobs launch -n {name} tests/test_yamls/pipeline_aws.yaml  -y -d',
+            f'sky jobs launch -n {name} tests/test_yamls/pipeline_aws.yaml -y -d',
             smoke_tests_utils.
             get_cmd_wait_until_managed_job_status_contains_matching_job_name(
                 job_name=name,
@@ -342,7 +341,6 @@ def test_managed_jobs_pipeline_recovery_gcp():
     name = smoke_tests_utils.get_cluster_name()
     zone = 'us-east4-b'
     user_hash = common_utils.get_user_hash()
-    user_hash = user_hash[:common_utils.USER_HASH_LENGTH_IN_CLUSTER_NAME]
     query_cmd = (
         'gcloud compute instances list --filter='
         f'"(labels.ray-cluster-name:*-${{MANAGED_JOB_ID}}-{user_hash})" '
@@ -352,7 +350,7 @@ def test_managed_jobs_pipeline_recovery_gcp():
     test = smoke_tests_utils.Test(
         'managed_jobs_pipeline_recovery_gcp',
         [
-            f'sky jobs launch -n {name} tests/test_yamls/pipeline_gcp.yaml  -y -d',
+            f'sky jobs launch -n {name} tests/test_yamls/pipeline_gcp.yaml -y -d',
             smoke_tests_utils.
             get_cmd_wait_until_managed_job_status_contains_matching_job_name(
                 job_name=name,
@@ -426,7 +424,7 @@ def test_managed_jobs_recovery_multi_node_aws(aws_config_region):
     test = smoke_tests_utils.Test(
         'managed_jobs_recovery_multi_node_aws',
         [
-            f'sky jobs launch --cloud aws --region {region} -n {name} --use-spot --num-nodes 2 "echo SKYPILOT_TASK_ID: \$SKYPILOT_TASK_ID; sleep 1800"  -y -d',
+            rf'sky jobs launch --cloud aws --region {region} -n {name} --use-spot --num-nodes 2 "echo SKYPILOT_TASK_ID: \$SKYPILOT_TASK_ID; sleep 1800" -y -d',
             smoke_tests_utils.
             get_cmd_wait_until_managed_job_status_contains_matching_job_name(
                 job_name=name,
@@ -473,7 +471,7 @@ def test_managed_jobs_recovery_multi_node_gcp():
     test = smoke_tests_utils.Test(
         'managed_jobs_recovery_multi_node_gcp',
         [
-            f'sky jobs launch --cloud gcp --zone {zone} -n {name} --use-spot --num-nodes 2 "echo SKYPILOT_TASK_ID: \$SKYPILOT_TASK_ID; sleep 1800"  -y -d',
+            rf'sky jobs launch --cloud gcp --zone {zone} -n {name} --use-spot --num-nodes 2 "echo SKYPILOT_TASK_ID: \$SKYPILOT_TASK_ID; sleep 1800" -y -d',
             smoke_tests_utils.
             get_cmd_wait_until_managed_job_status_contains_matching_job_name(
                 job_name=name,
@@ -512,7 +510,7 @@ def test_managed_jobs_cancellation_aws(aws_config_region):
         'managed_jobs_cancellation_aws',
         [
             # Test cancellation during spot cluster being launched.
-            f'sky jobs launch --cloud aws --region {region} -n {name} --use-spot "sleep 1000"  -y -d',
+            f'sky jobs launch --cloud aws --region {region} -n {name} --use-spot "sleep 1000" -y -d',
             smoke_tests_utils.
             get_cmd_wait_until_managed_job_status_contains_matching_job_name(
                 job_name=name,
@@ -532,7 +530,7 @@ def test_managed_jobs_cancellation_aws(aws_config_region):
              '--output text) && echo "$s" && echo; [[ -z "$s" ]] || [[ "$s" = "terminated" ]] || [[ "$s" = "shutting-down" ]]'
             ),
             # Test cancelling the spot cluster during spot job being setup.
-            f'sky jobs launch --cloud aws --region {region} -n {name}-2 --use-spot tests/test_yamls/test_long_setup.yaml  -y -d',
+            f'sky jobs launch --cloud aws --region {region} -n {name}-2 --use-spot tests/test_yamls/test_long_setup.yaml -y -d',
             # The job is set up in the cluster, will shown as RUNNING.
             smoke_tests_utils.
             get_cmd_wait_until_managed_job_status_contains_matching_job_name(
@@ -551,7 +549,7 @@ def test_managed_jobs_cancellation_aws(aws_config_region):
              '--output text) && echo "$s" && echo; [[ -z "$s" ]] || [[ "$s" = "terminated" ]] || [[ "$s" = "shutting-down" ]]'
             ),
             # Test cancellation during spot job is recovering.
-            f'sky jobs launch --cloud aws --region {region} -n {name}-3 --use-spot "sleep 1000"  -y -d',
+            f'sky jobs launch --cloud aws --region {region} -n {name}-3 --use-spot "sleep 1000" -y -d',
             # The job is running in the cluster, will shown as RUNNING.
             smoke_tests_utils.
             get_cmd_wait_until_managed_job_status_contains_matching_job_name(
@@ -605,7 +603,7 @@ def test_managed_jobs_cancellation_gcp():
         'managed_jobs_cancellation_gcp',
         [
             # Test cancellation during spot cluster being launched.
-            f'sky jobs launch --cloud gcp --zone {zone} -n {name} --use-spot "sleep 1000"  -y -d',
+            f'sky jobs launch --cloud gcp --zone {zone} -n {name} --use-spot "sleep 1000" -y -d',
             smoke_tests_utils.
             get_cmd_wait_until_managed_job_status_contains_matching_job_name(
                 job_name=name,
@@ -618,7 +616,7 @@ def test_managed_jobs_cancellation_gcp():
                 job_status=[sky.ManagedJobStatus.CANCELLED],
                 timeout=155),
             # Test cancelling the spot cluster during spot job being setup.
-            f'sky jobs launch --cloud gcp --zone {zone} -n {name}-2 --use-spot tests/test_yamls/test_long_setup.yaml  -y -d',
+            f'sky jobs launch --cloud gcp --zone {zone} -n {name}-2 --use-spot tests/test_yamls/test_long_setup.yaml -y -d',
             # The job is set up in the cluster, will shown as RUNNING.
             smoke_tests_utils.
             get_cmd_wait_until_managed_job_status_contains_matching_job_name(
@@ -632,7 +630,7 @@ def test_managed_jobs_cancellation_gcp():
                 job_status=[sky.ManagedJobStatus.CANCELLED],
                 timeout=155),
             # Test cancellation during spot job is recovering.
-            f'sky jobs launch --cloud gcp --zone {zone} -n {name}-3 --use-spot "sleep 1000"  -y -d',
+            f'sky jobs launch --cloud gcp --zone {zone} -n {name}-3 --use-spot "sleep 1000" -y -d',
             smoke_tests_utils.
             get_cmd_wait_until_managed_job_status_contains_matching_job_name(
                 job_name=f'{name}-3',
@@ -885,7 +883,7 @@ def test_managed_jobs_inline_env(generic_cloud: str):
     test = smoke_tests_utils.Test(
         'test-managed-jobs-inline-env',
         [
-            f'sky jobs launch -n {name} -y --cloud {generic_cloud} --env TEST_ENV="hello world" -- "echo "\\$TEST_ENV"; ([[ ! -z \\"\$TEST_ENV\\" ]] && [[ ! -z \\"\${constants.SKYPILOT_NODE_IPS}\\" ]] && [[ ! -z \\"\${constants.SKYPILOT_NODE_RANK}\\" ]] && [[ ! -z \\"\${constants.SKYPILOT_NUM_NODES}\\" ]]) || exit 1"',
+            rf'sky jobs launch -n {name} -y --cloud {generic_cloud} --env TEST_ENV="hello world" -- "echo "\$TEST_ENV"; ([[ ! -z \"\$TEST_ENV\" ]] && [[ ! -z \"\${constants.SKYPILOT_NODE_IPS}\" ]] && [[ ! -z \"\${constants.SKYPILOT_NODE_RANK}\" ]] && [[ ! -z \"\${constants.SKYPILOT_NUM_NODES}\" ]]) || exit 1"',
             smoke_tests_utils.
             get_cmd_wait_until_managed_job_status_contains_matching_job_name(
                 job_name=name,

tests/unit_tests/test_common_utils.py

@@ -36,16 +36,19 @@ class TestMakeClusterNameOnCloud:
     @mock.patch('sky.utils.common_utils.get_user_hash')
     def test_make(self, mock_get_user_hash):
         mock_get_user_hash.return_value = MOCKED_USER_HASH
-        assert "lora-ab12" == common_utils.make_cluster_name_on_cloud("lora")
+        assert "lora-ab12cd34" == common_utils.make_cluster_name_on_cloud(
+            "lora")
 
     @mock.patch('sky.utils.common_utils.get_user_hash')
     def test_make_with_hyphen(self, mock_get_user_hash):
         mock_get_user_hash.return_value = MOCKED_USER_HASH
-        assert "seed-1-ab12" == common_utils.make_cluster_name_on_cloud(
+        assert "seed-1-ab12cd34" == common_utils.make_cluster_name_on_cloud(
             "seed-1")
 
     @mock.patch('sky.utils.common_utils.get_user_hash')
     def test_make_with_characters_to_transform(self, mock_get_user_hash):
         mock_get_user_hash.return_value = MOCKED_USER_HASH
-        assert "cuda-11-8-ab12" == common_utils.make_cluster_name_on_cloud(
+        assert "cud-73-ab12cd34" == common_utils.make_cluster_name_on_cloud(
             "Cuda_11.8")
+        assert "cuda-11-8-ab12cd34" == common_utils.make_cluster_name_on_cloud(
+            "Cuda_11.8", max_length=20)