Merge pull request #4489 from codalab/rc1.7.0

Bump version to 1.7.0

codalab/common.py

@@ -29,7 +29,7 @@
 
 # Increment this on master when ready to cut a release.
 # http://semver.org/
-CODALAB_VERSION = '1.6.3'
+CODALAB_VERSION = '1.7.0'
 BINARY_PLACEHOLDER = '<binary>'
 URLOPEN_TIMEOUT_SECONDS = int(os.environ.get('CODALAB_URLOPEN_TIMEOUT_SECONDS', 5 * 60))
 

codalab/migration.py

@@ -106,6 +106,7 @@ def upload_to_azure_blob(self, bundle_uuid, bundle_location, is_dir=False):
         file_name = "contents.tar.gz" if is_dir else "contents.gz"
         target_location = f"{self.target_store_url}/{bundle_uuid}/{file_name}"
 
+        # TODO: This step might cause repeated upload. Can not check by checking size (Azure blob storage is zipped).
         if FileSystems.exists(target_location):
             path_util.remove(target_location)
 
@@ -202,33 +203,35 @@ def delete_original_bundle(self, bundle_uuid, bundle_location):
         description='Manages your local CodaLab Worksheets service deployment'
     )
     parser.add_argument(
-        '-w', '--worksheet', type=str, help='The worksheet uuid that needs migration'
+        '-a', '--all', help='Run migration on all worksheets and all bundles', action='store_true',
     )
-    parser.add_argument('--target_store_name', type=str, help='The destination bundle store name')
     parser.add_argument(
-        '-d',
-        '--dry-run',
-        help='Only upload the bundle to Azure, does not modify database',
-        action='store_true',
+        '-w', '--worksheet', type=str, help='The worksheet uuid that needs migration'
     )
+    parser.add_argument('--target_store_name', type=str, help='The destination bundle store name')
     parser.add_argument(
-        '-k', '--keep', help='Keep bundle content in origin bundle store', action='store_true'
+        '-c', '--change_db', help='Change the bundle location in the database', action='store_true',
     )
+    parser.add_argument('-d', '--delete', help='Delete the original database', action='store_true')
 
     args = parser.parse_args()
 
     worksheet_uuid = args.worksheet
     target_store_name = (
         "azure-store-default" if args.target_store_name is None else args.target_store_name
     )
-    if worksheet_uuid is not None and not spec_util.UUID_REGEX.match(worksheet_uuid):
-        raise Exception("Input worksheet uuid has wrong format. ")
 
     # TODO: write output to log / log files
     migration = Migration(target_store_name)
     migration.setUp()
 
-    bundle_uuids = migration.get_bundle_uuids(worksheet_uuid)
+    if args.all:
+        bundle_uuids = migration.get_bundle_uuids(worksheet_uuid=None)
+    else:
+        # Must specify worksheet uuid
+        if worksheet_uuid is not None and not spec_util.UUID_REGEX.match(worksheet_uuid):
+            raise Exception("Input worksheet uuid has wrong format. ")
+        bundle_uuids = migration.get_bundle_uuids(worksheet_uuid)
 
     for bundle_uuid in bundle_uuids:
         logging.info(bundle_uuid)
@@ -256,8 +259,8 @@ def delete_original_bundle(self, bundle_uuid, bundle_location):
         is_dir = bundle_info['type'] == 'directory'
         migration.upload_to_azure_blob(bundle_uuid, bundle_location, is_dir)
 
-        if not args.dry_run:  # If dry_run, only upload to new bundle location
+        if args.change_db:  # If need to change the database, continue to run
             migration.modify_bundle_data(bundle, bundle_uuid, is_dir)
             migration.sanity_check(bundle_uuid, bundle_location, bundle_info, is_dir)
-            if not args.keep:
+            if args.delete:
                 migration.delete_original_bundle(bundle_uuid, bundle_location)

codalab/worker/docker_utils.py

@@ -152,7 +152,7 @@ def get_nvidia_devices(self, use_docker=True):
             docker.errors.ImageNotFound if the CUDA image cannot be pulled
             docker.errors.APIError if another server error occurs
         """
-        cuda_image = 'nvidia/cuda:12.2.0-devel-ubuntu22.04'
+        cuda_image = 'sulfurheron/nvidia-cuda:9.0-cudnn7-devel-ubuntu16.04-2018-06-08'
         nvidia_command = 'nvidia-smi --query-gpu=index,uuid --format=csv,noheader'
         if use_docker:
             self.client.images.pull(cuda_image)
@@ -164,20 +164,27 @@ def get_nvidia_devices(self, use_docker=True):
                 stdout=True,
                 remove=True,
             )
-            gpus = output.decode()
+            gpu_info = output.decode()
+            GPU_REGEX = r"(\d+), ((?:GPU-)[a-fA-F0-9-]+)"
+            gpus = {}
+            for line in gpu_info.splitlines():
+                match = re.match(GPU_REGEX, line)
+                if match:
+                    idx = match.group(1)
+                    uuid = match.group(2)
+                    gpus[idx] = uuid
+
         else:
             # use the singularity runtime to run nvidia-smi
             # img = Client.pull('docker://' + cuda_image, pull_folder='/tmp')
             # output = Client.execute(img, nvidia_command, options=['--nv'])
             # if output['return_code'] != 0:
             #     raise SingularityError
             # gpus = output['message']
-            gpus = ""
+            gpus = {}
         # Get newline delimited gpu-index, gpu-uuid list
-        logger.info("GPUs: " + str(gpus.split('\n')[:-1]))
-        return {
-            gpu.split(',')[0].strip(): gpu.split(',')[1].strip() for gpu in gpus.split('\n')[:-1]
-        }
+        logger.info("GPUs: " + str(gpus))
+        return gpus
 
     @wrap_exception('Unable to fetch Docker container ip')
     def get_container_ip(self, network_name: str, container_id: str):

codalab/worker/main.py

@@ -326,8 +326,11 @@ def main():
             # Create temp file to store kubernetes cert, as we need to pass in a file path.
             # TODO: Delete the file afterwards (upon CodaLab service stop?)
             with tempfile.NamedTemporaryFile(mode="w", delete=False) as f:
-                f.write(args.kubernetes_cert)
+                f.write(
+                    args.kubernetes_cert.replace(r'\n', '\n')
+                )  # Properly add newlines, which appear as "\n" if specified in the environment variable.
                 kubernetes_cert_path = f.name
+                logger.info('Temporarily writing kubernetes cert to: %s', kubernetes_cert_path)
         else:
             kubernetes_cert_path = args.kubernetes_cert_path
         bundle_runtime_class = KubernetesRuntime(

codalab/worker_manager/kubernetes_worker_manager.py

@@ -13,6 +13,7 @@
 from argparse import ArgumentParser
 from typing import Any, Dict, List, Optional
 from codalab.common import BundleRuntime
+import tempfile
 
 from urllib3.exceptions import MaxRetryError, NewConnectionError  # type: ignore
 
@@ -89,7 +90,18 @@ def __init__(self, args):
         configuration.api_key_prefix['authorization'] = 'Bearer'
         configuration.api_key['authorization'] = args.auth_token
         configuration.host = args.cluster_host
-        configuration.ssl_ca_cert = args.cert_path
+        if args.cert_path == "/dev/null" and args.cert != "/dev/null":
+            # Create temp file to store kubernetes cert, as we need to pass in a file path.
+            # TODO: Delete the file afterwards (upon CodaLab service stop?)
+            with tempfile.NamedTemporaryFile(mode="w", delete=False) as f:
+                f.write(
+                    args.cert.replace(r'\n', '\n')
+                )  # Properly add newlines, which appear as "\n" if specified in the environment variable.
+                cert_path = f.name
+                logger.info('Temporarily writing kubernetes cert to: %s', cert_path)
+        else:
+            cert_path = args.cert_path
+        configuration.ssl_ca_cert = cert_path
         if configuration.host == "https://codalab-control-plane:6443":
             # Don't verify SSL if we are connecting to a local cluster for testing / development.
             configuration.verify_ssl = False

docs/REST-API-Reference.md

@@ -1,6 +1,6 @@
 # REST API Reference
 
-_version 1.6.3_
+_version 1.7.0_
 
 This reference and the REST API itself is still under heavy development and is
 subject to change at any time. Feedback through our GitHub issues is appreciated!

frontend/src/constants.js

@@ -1,5 +1,5 @@
 // Should match codalab/common.py#CODALAB_VERSION
-export const CODALAB_VERSION = '1.6.3';
+export const CODALAB_VERSION = '1.7.0';
 
 // Name Regex to match the backend in spec_utils.py
 export const NAME_REGEX = /^[a-zA-Z_][a-zA-Z0-9_.-]*$/i;

setup.py

@@ -7,7 +7,7 @@
 
 
 # should match codalab/common.py#CODALAB_VERSION
-CODALAB_VERSION = "1.6.3"
+CODALAB_VERSION = "1.7.0"
 
 
 class Install(install):