cgcsdk 1.0.5__py3-none-any.whl → 1.0.7__py3-none-any.whl

cgc/commands/jobs/jobs_cmd.py

@@ -0,0 +1,238 @@
+import click
+import json
+import sys
+
+from cgc.commands.jobs.job_utils import job_delete_payload, job_create_payload
+from cgc.commands.jobs.jobs_responses import (
+    job_delete_response,
+    job_list_response,
+    job_create_response,
+)
+from cgc.commands.compute.compute_models import GPUsList
+from cgc.utils.prepare_headers import get_api_url_and_prepare_headers
+from cgc.utils.response_utils import (
+    retrieve_and_validate_response_send_metric,
+)
+from cgc.utils.click_group import CustomGroup, CustomCommand
+from cgc.utils.requests_helper import call_api, EndpointTypes
+
+
+@click.group(name="job", cls=CustomGroup, hidden=False)
+def job_group():
+    """
+    Management of jobs.
+    """
+
+
+@job_group.command("delete", cls=CustomCommand)
+@click.argument("name", type=click.STRING)
+def job_delete(name: str):
+    """
+    Delete an job using backend endpoint.
+    \f
+    :param name: name of job to delete
+    :type name: str
+    """
+    api_url, headers = get_api_url_and_prepare_headers()
+    url = f"{api_url}/v1/api/job/delete"
+    metric = "job.delete"
+    __payload = job_delete_payload(name=name)
+    __res = call_api(
+        request=EndpointTypes.delete,
+        url=url,
+        headers=headers,
+        data=json.dumps(__payload).encode("utf-8"),
+    )
+    click.echo(
+        job_delete_response(retrieve_and_validate_response_send_metric(__res, metric))
+    )
+
+
+@job_group.command("list", cls=CustomCommand)
+def job_list():
+    """List all ports for a running resource"""
+    api_url, headers = get_api_url_and_prepare_headers()
+    url = f"{api_url}/v1/api/job/list"
+    metric = "job.list"
+    __res = call_api(
+        request=EndpointTypes.get,
+        url=url,
+        headers=headers,
+    )
+    click.echo(
+        job_list_response(retrieve_and_validate_response_send_metric(__res, metric))
+    )
+
+
+@job_group.command("create", cls=CustomCommand)
+@click.argument("startup_command", required=False)
+@click.option(
+    "-n", "--name", "name", type=click.STRING, required=True, help="Desired app name"
+)
+@click.option(
+    "-g",
+    "--gpu",
+    "gpu",
+    type=click.INT,
+    default=0,
+    help="How much GPU cards app will use",
+)
+@click.option(
+    "-gt",
+    "--gpu-type",
+    "gpu_type",
+    type=click.Choice(GPUsList.get_list(), case_sensitive=False),
+    default="A5000",
+    help="Graphic card used by the app",
+)
+@click.option(
+    "-c",
+    "--cpu",
+    "cpu",
+    type=click.INT,
+    default=1,
+    help="How much CPU cores app can use",
+)
+@click.option(
+    "-m",
+    "--memory",
+    "memory",
+    type=click.INT,
+    default=2,
+    help="How much Gi RAM app can use",
+)
+@click.option(
+    "-v",
+    "--volume",
+    "volumes",
+    multiple=True,
+    help="List of volume names to be mounted with default mount path",
+)
+@click.option(
+    "-fp",
+    "--full-path",
+    "volume_full_path",
+    type=click.STRING,
+    help="If set, full path will be used for volume mount. Valid for 1 volume.",
+)
+@click.option(
+    "-d",
+    "--resource-data",
+    "resource_data",
+    multiple=True,
+    help="List of optional arguments to be passed to the app, key=value format",
+)
+@click.option(
+    "--image",
+    "image_name",
+    type=click.STRING,
+    help="Image to be used by the app",
+)
+@click.option(
+    "--repository-secret",
+    "repository_secret",
+    type=click.STRING,
+    help="Use secret to pull image from private repository",
+)
+@click.option(
+    "-cm",
+    "--config-map",
+    "config_maps_data",
+    multiple=True,
+    help="List of optional arguments to be passed to the app, key=value format",
+)
+@click.option(
+    "--shm",
+    "shm_size",
+    type=click.IntRange(0, 1024, clamp=True),
+    default=0,
+    help="Size of shared memory in Gi",
+)
+@click.option(
+    "--ttl",
+    "ttl_seconds_after_finished",
+    type=click.INT,
+    default=None,
+    help="Time to live in seconds after app is finished",
+)
+def job_create(
+    gpu: int,
+    gpu_type: str,
+    cpu: int,
+    memory: int,
+    volumes: list[str],
+    volume_full_path: str,
+    resource_data: list[str],
+    config_maps_data: list[str],
+    name: str,
+    shm_size: int,
+    image_name: str,
+    startup_command: str,
+    repository_secret: str,
+    ttl_seconds_after_finished: int,
+):
+    """
+    Create job in user namespace.
+    \f
+    :param gpu: number of gpus to be used by app
+    :type gpu: int
+    :param cpu: number of cores to be used by app
+    :type cpu: int
+    :param memory: GB of memory to be used by app
+    :type memory: int
+    :param volumes: list of volumes to mount
+    :type volumes: list[str]
+    :param volume_full_path: if set, full path will be used for volume mount
+    :type volume_full_path: str
+    :param resource_data: list of optional arguments to be passed to the app
+    :type resource_data: list[str]
+    :param config_maps_data: list of optional arguments to be passed to the app
+    :type config_maps_data: list[str]
+    :param name: name of app
+    :type name: str
+    :param shm_size: size of shared memory
+    :type shm_size: int
+    :param image_name: name of image to be used by the app
+    :type image_name: str
+    :param startup_command: command to be executed on app startup; it is stdin input
+    :type startup_command: str
+    :param repository_secret: use secret to pull image from private repository
+    :type repository_secret: str
+    :param ttl_seconds_after_finished: time to live in seconds after app is finished
+    :type ttl_seconds_after_finished: int
+    """
+    api_url, headers = get_api_url_and_prepare_headers()
+    url = f"{api_url}/v1/api/job/create"
+    cleaned_data = ""
+    if not sys.stdin.isatty():
+        input_data = sys.stdin.read()
+        cleaned_data = input_data.replace("|", "")
+        startup_command = cleaned_data
+    elif startup_command:
+        cleaned_data = startup_command
+    metric = "job.create"
+    __payload = job_create_payload(
+        name=name,
+        cpu=cpu,
+        memory=memory,
+        gpu=gpu,
+        volumes=volumes,
+        volume_full_path=volume_full_path,
+        resource_data=resource_data,
+        config_maps_data=config_maps_data,
+        gpu_type=gpu_type,
+        shm_size=shm_size,
+        image_name=image_name,
+        startup_command=cleaned_data,
+        repository_secret=repository_secret,
+        ttl_seconds_after_finished=ttl_seconds_after_finished,
+    )
+    __res = call_api(
+        request=EndpointTypes.post,
+        url=url,
+        headers=headers,
+        data=json.dumps(__payload).encode("utf-8"),
+    )
+    click.echo(
+        job_create_response(retrieve_and_validate_response_send_metric(__res, metric))
+    )

cgc/commands/jobs/jobs_responses.py

@@ -0,0 +1,52 @@
+from cgc.commands.jobs import NoJobsToList
+from cgc.commands.jobs.job_utils import get_job_list
+from cgc.telemetry.basic import change_gauge, setup_gauge
+from cgc.utils.config_utils import get_namespace
+from cgc.utils.message_utils import key_error_decorator_for_helpers
+from cgc.utils.response_utils import (
+    fill_missing_values_in_a_response,
+    tabulate_a_response,
+)
+
+
+@key_error_decorator_for_helpers
+def job_delete_response(data: dict) -> str:
+    """Create response string for job delete command.
+
+    :param response: dict object from API response.
+    :type response: requests.Response
+    :return: Response string.
+    :rtype: str
+    """
+    name = data.get("details", {}).get("job_deleted", {}).get("name")
+    change_gauge(f"{get_namespace()}.job.count", -1)
+    return f"Job {name} and its service successfully deleted."
+
+
+@key_error_decorator_for_helpers
+def job_list_response(data: dict) -> list:
+    job_pod_list = data.get("details", {}).get("job_pod_list", [])
+    job_list = data.get("details", {}).get("job_list", [])
+    setup_gauge(f"{get_namespace()}.job.count", len(job_list))
+
+    if not job_list:
+        raise NoJobsToList()
+
+    list_of_json_data = get_job_list(job_pod_list, job_list)
+    table = fill_missing_values_in_a_response(list_of_json_data)
+
+    return tabulate_a_response(table)
+
+
+@key_error_decorator_for_helpers
+def job_create_response(data: dict) -> str:
+    """Create response string for job create command.
+
+    :param response: dict object from API response.
+    :type response: requests.Response
+    :return: Response string.
+    :rtype: str
+    """
+    name = data.get("details", {}).get("job_created", {}).get("name")
+    change_gauge(f"{get_namespace()}.job.count", 1)
+    return f"Job {name} created successfully."

cgc/commands/resource/resource_cmd.py

@@ -72,7 +72,7 @@ def compute_restart(name: str):
         request=EndpointTypes.post,
         url=url,
         headers=headers,
-        data=json.dumps(__payload),
+        data=json.dumps(__payload).encode("utf-8"),
     )
     click.echo(
         compute_restart_response(
@@ -116,7 +116,7 @@ def resource_delete(name: str):
         request=EndpointTypes.delete,
         url=url,
         headers=headers,
-        data=json.dumps(__payload),
+        data=json.dumps(__payload).encode("utf-8"),
     )
     click.echo(
         compute_delete_response(

cgc/commands/volume/volume_cmd.py

@@ -94,7 +94,7 @@ def volume_create(name: str, size: int, disk_type: str, access: str):
     __res = call_api(
         request=EndpointTypes.post,
         url=url,
-        data=json.dumps(__payload),
+        data=json.dumps(__payload).encode("utf-8"),
         headers=headers,
     )
     click.echo(
@@ -132,7 +132,7 @@ def volume_delete(name: str, force_delete: bool):
     __res = call_api(
         request=EndpointTypes.delete,
         url=url,
-        data=json.dumps(__payload),
+        data=json.dumps(__payload).encode("utf-8"),
         headers=headers,
     )
     click.echo(
@@ -182,7 +182,7 @@ def volume_umount(name: str, target_template_names):
     __res = call_api(
         request=EndpointTypes.post,
         url=url,
-        data=json.dumps(__payload),
+        data=json.dumps(__payload).encode("utf-8"),
         headers=headers,
     )
     click.echo(
@@ -270,7 +270,7 @@ def volume_mount(
     __res = call_api(
         request=EndpointTypes.post,
         url=url,
-        data=json.dumps(__payload),
+        data=json.dumps(__payload).encode("utf-8"),
         headers=headers,
     )
     click.echo(

cgc/sdk/__init__.py

@@ -3,4 +3,5 @@ from cgc.sdk.redis import get_redis_access as redis_client
 from cgc.sdk.postgresql import get_postgresql_access as postgresql_client
 
 import cgc.sdk.resource as resource
+import cgc.sdk.job as job
 import cgc.sdk.exceptions as exceptions

cgc/sdk/job.py

@@ -0,0 +1,147 @@
+import json as _json
+from typing import Optional
+import cgc.sdk.exceptions as _exceptions
+from cgc.utils.custom_exceptions import CUSTOM_EXCEPTIONS
+import cgc.utils.prepare_headers as _prepare_headers
+import cgc.commands.jobs.job_utils as _job_utils
+import cgc.utils.requests_helper as _requests_helper
+import cgc.utils.response_utils as _response_utils
+from cgc.commands.compute.compute_models import GPUsList
+
+
+def job_create(
+    name: str,
+    image_name: str,
+    cpu: int = 1,
+    memory: int = 2,
+    shm_size: int = 0,
+    gpu: int = 0,
+    gpu_type: str = "A5000",
+    volumes: list = [],
+    volume_full_path: str = "",
+    startup_command: str = "",
+    repository_secret: str = "",
+    resource_data: list = [],
+    config_maps_data: list = [],
+    ttl_seconds_after_finished: Optional[int] = None,
+):
+    """
+    Create a custom compute resource.
+
+    :param name: The name of the compute resource.
+    :type name: str
+    :param image_name: The name of the image to use for the compute resource.
+    :type image_name: str,
+    :param cpu: The number of CPUs for the compute resource, defaults to 1.
+    :type cpu: int, optional
+    :param memory: The amount of memory (in GB) for the compute resource, defaults to 2.
+    :type memory: int, optional
+    :param shm_size: The size of the shared memory (in GB) for the compute resource, defaults to 0.
+    :type shm_size: int, optional
+    :param gpu: The number of GPUs for the compute resource, defaults to 0.
+    :type gpu: int, optional
+    :param gpu_type: The type of GPU for the compute resource, defaults to "A5000".
+    :type gpu_type: str, optional
+    :param volumes: The list of volumes to attach to the compute resource, defaults to [].
+    :type volumes: list, optional
+    :param volume_full_path: The full path of the volume, defaults to "".
+    :type volume_full_path: str, optional
+    :param startup_command: The startup command for the compute resource, defaults to "".
+    :type startup_command: str, optional
+    :param repository_secret: The secret for accessing the repository, defaults to "".
+    :type repository_secret: str, optional
+    :param resource_data: The additional resource data, defaults to [].
+    :type resource_data: list, optional
+    :param config_maps_data: The additional config maps data, defaults to [].
+    :type config_maps_data: list, optional
+    :param ttl_seconds_after_finished: The time to live in seconds after the app is finished, defaults to None.
+    :type ttl_seconds_after_finished: int, optional
+    :raises _SDKException: If the image name is not provided.
+    :raises _SDKException: If an invalid GPU type is specified.
+    :return: The response from the API call.
+    :rtype: _type_
+    """
+    if not image_name:
+        raise _exceptions.SDKException(-2, "Image name is required")
+    api_url, headers = _prepare_headers.get_api_url_and_prepare_headers()
+    url = f"{api_url}/v1/api/job/create"
+    metric = "job.create"
+    gpu_type = gpu_type.upper()
+    if gpu_type not in GPUsList.get_list():
+        raise _exceptions.SDKException(-3, f"Invalid GPU type: {gpu_type}")
+    __payload = _job_utils.job_create_payload(
+        name=name,
+        cpu=cpu,
+        memory=memory,
+        gpu=gpu,
+        gpu_type=gpu_type,
+        volumes=volumes,
+        volume_full_path=volume_full_path,
+        resource_data=resource_data,
+        config_maps_data=config_maps_data,
+        shm_size=shm_size,
+        image_name=image_name,
+        startup_command=startup_command,
+        repository_secret=repository_secret,
+        ttl_seconds_after_finished=ttl_seconds_after_finished,
+    )
+
+    __res = _requests_helper.call_api(
+        request=_requests_helper.EndpointTypes.post,
+        url=url,
+        headers=headers,
+        data=_json.dumps(__payload).encode("utf-8"),
+    )
+
+    return _response_utils.retrieve_and_validate_response_send_metric_for_sdk(
+        __res, metric
+    )
+
+
+def job_list():
+    """
+    List jobs using backend endpoint.
+
+    :return: response from the API call
+    :rtype: dict
+    """
+    api_url, headers = _prepare_headers.get_api_url_and_prepare_headers()
+    url = f"{api_url}/v1/api/job/list"
+    metric = "job.list"
+    __res = _requests_helper.call_api(
+        request=_requests_helper.EndpointTypes.get,
+        url=url,
+        headers=headers,
+    )
+
+    return _response_utils.retrieve_and_validate_response_send_metric_for_sdk(
+        __res, metric
+    )
+    # job_pod_list = _response.get("details", {}).get("job_pod_list", [])
+    # job_list = _response.get("details", {}).get("job_list", [])
+    # return _job_utils.get_job_list(job_pod_list, job_list)
+
+
+def job_delete(name: str):
+    """
+    Delete job using backend endpoint.
+
+    :param name: name of job to delete
+    :type name: str
+    :return: response from the API call
+    :rtype: dict
+    """
+    api_url, headers = _prepare_headers.get_api_url_and_prepare_headers()
+    url = f"{api_url}/v1/api/job/delete"
+    metric = "job.delete"
+    __payload = _job_utils.job_delete_payload(name=name)
+    __res = _requests_helper.call_api(
+        request=_requests_helper.EndpointTypes.delete,
+        url=url,
+        headers=headers,
+        data=_json.dumps(__payload).encode("utf-8"),
+    )
+
+    return _response_utils.retrieve_and_validate_response_send_metric_for_sdk(
+        __res, metric
+    )

cgc/sdk/resource.py

@@ -1,6 +1,6 @@
 import json as _json
-import re
 import cgc.sdk.exceptions as _exceptions
+from cgc.utils.custom_exceptions import CUSTOM_EXCEPTIONS
 import cgc.utils.prepare_headers as _prepare_headers
 import cgc.commands.compute.compute_utills as _compute_utills
 import cgc.utils.requests_helper as _requests_helper
@@ -20,41 +20,36 @@ def start_function_loop(function, infinite: bool = True, *args, **kwargs):
         **kwargs: Keyword arguments to be passed to the function.
 
     Raises:
-        _SDKException: If the app fails to start within 10 iterations (when infinite is False).
+        _SDKException: If the response code is not 200 after 5 attempts (if finite) or if the response code is 409.
 
     Returns:
         None
     """
-    from time import sleep
+    import logging
 
     counter = 0
-    try:
-        response = function(*args, **kwargs)
-        if type(response) is bool:
-            while not response:
-                counter += 1
-                if not infinite and counter > 10:
-                    raise _exceptions.SDKException(-1, response)
-                sleep(5)
-                response = function(*args, **kwargs)
-        elif type(response) is dict:
-            while response["code"] != 200:
-                counter += 1
-                if not infinite and counter > 10:
-                    raise _exceptions.SDKException(-1, response)
-                sleep(5)
-                response = function(*args, **kwargs)
-        else:
-            raise _exceptions.SDKException(-1, response)
-    except _exceptions.SDKException as e:
-        import logging
-
-        if e.code == 409:
-            logging.warning(e)
-        else:
-            raise e
-    else:
-        return response
+    while True:
+        try:
+            counter += 1
+            response = function(*args, **kwargs)
+
+            if type(response) is bool:
+                if response:
+                    break
+            elif type(response) is dict:
+                if response["code"] == 409:
+                    raise _exceptions.SDKException(409, response)
+                return response
+        except _exceptions.SDKException as e:
+            if e.code == 409:
+                logging.warning(e)
+                return
+            logging.error(e)
+        finally:
+            if not infinite and counter > 4:
+                raise _exceptions.SDKException(
+                    -1, "Response code not 200 after 5 attempts."
+                )
 
 
 def stop_function_loop(function, infinite: bool = True, *args, **kwargs):
@@ -68,32 +63,35 @@ def stop_function_loop(function, infinite: bool = True, *args, **kwargs):
         **kwargs: Arbitrary keyword arguments to be passed to the function.
 
     Raises:
-        _SDKException: If the app fails to stop within 10 attempts (when infinite is False).
+        _SDKException: If the response code is not 200 after 5 attempts (if finite) or if the response code is 404.
 
     Returns:
         None
     """
-    from time import sleep
+    import logging
 
     counter = 0
-    response = function(*args, **kwargs)
-    if type(response) is bool:
-        while not response:
-            counter += 1
-            if not infinite and counter > 10:
-                raise _exceptions.SDKException(-1, response)
-            sleep(5)
-            response = function(*args, **kwargs)
-    elif type(response) is dict:
-        while response["code"] != 200:
+    while True:
+        try:
             counter += 1
-            if not infinite and counter > 10:
-                raise _exceptions.SDKException(-1, response)
-            sleep(5)
             response = function(*args, **kwargs)
-    else:
-        raise _exceptions.SDKException(-1, response)
-    return response
+            if type(response) is bool:
+                if not response:
+                    break
+            elif type(response) is dict:
+                if response["code"] == 404:
+                    raise _exceptions.SDKException(404, response)
+                return response
+        except _exceptions.SDKException as e:
+            if e.code == 404:
+                logging.warning(e)
+                return
+            logging.error(e)
+        finally:
+            if not infinite and counter > 4:
+                raise _exceptions.SDKException(
+                    -1, "Response code not 200 after 5 attempts."
+                )
 
 
 def compute_list():
@@ -116,6 +114,7 @@ def compute_list():
 class ResourceTypes(_Enum):
     compute = "compute"
     db = "db"
+    job = "job"
 
 
 def _resource_status_ready(name: str, response: dict):
@@ -266,7 +265,7 @@ def compute_create_custom(
         request=_requests_helper.EndpointTypes.post,
         url=url,
         headers=headers,
-        data=_json.dumps(__payload),
+        data=_json.dumps(__payload).encode("utf-8"),
     )
 
     return _response_utils.retrieve_and_validate_response_send_metric_for_sdk(
@@ -302,7 +301,7 @@ def resource_update_port(
         request=_requests_helper.EndpointTypes.post,
         url=url,
         headers=headers,
-        data=_json.dumps(__payload),
+        data=_json.dumps(__payload).encode("utf-8"),
     )
 
     return _response_utils.retrieve_and_validate_response_send_metric_for_sdk(
@@ -336,7 +335,7 @@ def resource_add_port(name: str, port_name: str, new_port: int, ingress: bool =
         request=_requests_helper.EndpointTypes.post,
         url=url,
         headers=headers,
-        data=_json.dumps(__payload),
+        data=_json.dumps(__payload).encode("utf-8"),
     )
 
     return _response_utils.retrieve_and_validate_response_send_metric_for_sdk(
@@ -369,7 +368,7 @@ def resource_delete_port(
         request=_requests_helper.EndpointTypes.post,
         url=url,
         headers=headers,
-        data=_json.dumps(__payload),
+        data=_json.dumps(__payload).encode("utf-8"),
     )
 
     return _response_utils.retrieve_and_validate_response_send_metric_for_sdk(
@@ -417,7 +416,7 @@ def resource_delete(name: str):
         request=_requests_helper.EndpointTypes.delete,
         url=url,
         headers=headers,
-        data=_json.dumps(__payload),
+        data=_json.dumps(__payload).encode("utf-8"),
     )
 
     return _response_utils.retrieve_and_validate_response_send_metric_for_sdk(