torchx-nightly 2025.9.21__py3-none-any.whl → 2025.9.23__py3-none-any.whl

torchx/cli/cmd_list.py

@@ -33,8 +33,7 @@ class CmdList(SubCommand):
             type=str,
             default=get_default_scheduler_name(),
             choices=list(scheduler_names),
-            help=f"Name of the scheduler to use. One of: [{','.join(scheduler_names)}]."
-            " For listing app handles for ray scheduler, RAY_ADDRESS env variable should be set.",
+            help=f"Name of the scheduler to use. One of: [{','.join(scheduler_names)}].",
         )
 
     def run(self, args: argparse.Namespace) -> None:

torchx/components/integration_tests/component_provider.py

@@ -109,7 +109,7 @@ class CopyComponentProvider(ComponentProvider):
         self._dst_path = "<None>"
 
     def setUp(self) -> None:
-        if self._scheduler in ["local_cwd", "ray"]:
+        if self._scheduler in ["local_cwd"]:
             fname = "torchx_copy_test.txt"
             self._src_path: str = os.path.join(tempfile.gettempdir(), fname)
             self._dst_path: str = os.path.join(tempfile.gettempdir(), f"{fname}.copy")
@@ -126,7 +126,7 @@ class CopyComponentProvider(ComponentProvider):
     def tearDown(self) -> None:
         if os.path.exists(self._dst_path):
             os.remove(self._dst_path)
-        if self._scheduler in ["local_cwd", "ray"] and os.path.exists(self._dst_path):
+        if self._scheduler in ["local_cwd"] and os.path.exists(self._dst_path):
             os.remove(self._dst_path)
 
     def get_app_def(self) -> AppDef:

torchx/schedulers/__init__.py

@@ -21,7 +21,6 @@ DEFAULT_SCHEDULER_MODULES: Mapping[str, str] = {
     "kubernetes_mcad": "torchx.schedulers.kubernetes_mcad_scheduler",
     "aws_batch": "torchx.schedulers.aws_batch_scheduler",
     "aws_sagemaker": "torchx.schedulers.aws_sagemaker_scheduler",
-    "gcp_batch": "torchx.schedulers.gcp_batch_scheduler",
     "lsf": "torchx.schedulers.lsf_scheduler",
 }
 

torchx/util/strings.py

@@ -13,7 +13,7 @@ def normalize_str(data: str) -> str:
     """
     Invokes ``lower`` on thes string and removes all
     characters that do not satisfy ``[a-z0-9\\-]`` pattern.
-    This method is mostly used to make sure kubernetes and gcp_batch scheduler gets
+    This method is mostly used to make sure kubernetes scheduler gets
     the job name that does not violate its restrictions.
     """
     if data.startswith("-"):

{torchx_nightly-2025.9.21.dist-info → torchx_nightly-2025.9.23.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: torchx-nightly
-Version: 2025.9.21
+Version: 2025.9.23
 Summary: TorchX SDK and Components
 Home-page: https://github.com/pytorch/torchx
 Author: TorchX Devs
@@ -22,7 +22,6 @@ Requires-Dist: pyyaml
 Requires-Dist: docker
 Requires-Dist: filelock
 Requires-Dist: fsspec>=2023.10.0
-Requires-Dist: urllib3<1.27,>=1.21.1
 Requires-Dist: tabulate
 Provides-Extra: aws_batch
 Requires-Dist: boto3; extra == "aws-batch"
@@ -36,9 +35,6 @@ Requires-Dist: kubernetes==25.3.0; extra == "dev"
 Requires-Dist: flake8==3.9.0; extra == "dev"
 Requires-Dist: fsspec==2024.3.1; extra == "dev"
 Requires-Dist: s3fs==2024.3.1; extra == "dev"
-Requires-Dist: google-cloud-batch==0.17.14; extra == "dev"
-Requires-Dist: google-cloud-logging==3.10.0; extra == "dev"
-Requires-Dist: google-cloud-runtimeconfig==0.34.0; extra == "dev"
 Requires-Dist: hydra-core; extra == "dev"
 Requires-Dist: ipython; extra == "dev"
 Requires-Dist: mlflow-skinny; extra == "dev"
@@ -61,14 +57,6 @@ Requires-Dist: ts==0.5.1; extra == "dev"
 Requires-Dist: wheel; extra == "dev"
 Requires-Dist: lintrunner; extra == "dev"
 Requires-Dist: lintrunner-adapters; extra == "dev"
-Requires-Dist: grpcio==1.62.1; extra == "dev"
-Requires-Dist: grpcio-status==1.48.1; extra == "dev"
-Requires-Dist: googleapis-common-protos==1.63.0; extra == "dev"
-Requires-Dist: google-api-core==2.18.0; extra == "dev"
-Provides-Extra: gcp_batch
-Requires-Dist: google-cloud-batch>=0.5.0; extra == "gcp-batch"
-Requires-Dist: google-cloud-logging>=3.0.0; extra == "gcp-batch"
-Requires-Dist: google-cloud-runtimeconfig>=0.33.2; extra == "gcp-batch"
 Provides-Extra: kubernetes
 Requires-Dist: kubernetes>=11; extra == "kubernetes"
 
@@ -93,7 +81,6 @@ TorchX currently supports:
 * AWS Batch
 * Docker
 * Local
-* GCP Batch (prototype)
 
 Need a scheduler not listed? [Let us know!](https://github.com/pytorch/torchx/issues?q=is%3Aopen+is%3Aissue+label%3Ascheduler-request)
 

{torchx_nightly-2025.9.21.dist-info → torchx_nightly-2025.9.23.dist-info}/RECORD

@@ -14,7 +14,7 @@ torchx/cli/cmd_base.py,sha256=SdqMtqi04CEqnzcgcS35DbDbsBeMxSgEhfynfpIkMGk,790
 torchx/cli/cmd_cancel.py,sha256=NKfOCu_44Lch9vliGSQ0Uv6BVqpUqj7Tob652TI-ua4,835
 torchx/cli/cmd_configure.py,sha256=1kTv0qbsbV44So74plAySwWu56pQrqjhfW_kbfdC3Rw,1722
 torchx/cli/cmd_describe.py,sha256=E5disbHoKTsqYKp2s3DaFW9GDLCCOgdOc3pQoHKoyCs,1283
-torchx/cli/cmd_list.py,sha256=4Y1ZOq-kqJbztoBt56hAW_InJEaJuDAjpKWgMhBw4II,1507
+torchx/cli/cmd_list.py,sha256=alkS9aIaDI8lX3W8uj8Vtr3IU3G2VeCuokKSd3zOFug,1409
 torchx/cli/cmd_log.py,sha256=v-EZYUDOcG95rEgTnrsmPJMUyxM9Mk8YFAJtUxtgViE,5475
 torchx/cli/cmd_run.py,sha256=TshvEMTxMRj5O0KhetzHepZUaAFq8R5nFgY8GC_Gl6g,18576
 torchx/cli/cmd_runopts.py,sha256=NWZiP8XpQjfTDJgays2c6MgL_8wxFoeDge6NstaZdKk,1302
@@ -32,7 +32,7 @@ torchx/components/structured_arg.py,sha256=8jMcd0rtUmzCKEQKJ_JYzxSkMMK9q0fYjkwAs
 torchx/components/train.py,sha256=vtrQXRcD7bIcbb3lSeyD9BBlIe1mv1WNW6rnLK9R0Mw,1259
 torchx/components/utils.py,sha256=QRBxBm1OnNhOhpPs0lKdbJ8_mNhWYMklY6cl1gPIw9A,9363
 torchx/components/integration_tests/__init__.py,sha256=Md3cCHD7Ano9kV15PqGbicgUO-RMdh4aVy1yKiDt_xE,208
-torchx/components/integration_tests/component_provider.py,sha256=cFNGqmclcZTJlOW_YGf5XEuGeWloTmcJEAh02Aob_PQ,3995
+torchx/components/integration_tests/component_provider.py,sha256=g-4ig1vtd5Vzgug0VAKRAFUt6KAV3TgQrBCrwRSJ7ZY,3981
 torchx/components/integration_tests/integ_tests.py,sha256=O8jd8Jq5O0mns7xzIFsHexBDHkIIAIfELQkWCzNPzRw,5165
 torchx/distributed/__init__.py,sha256=lobebigfujmRTe_SJw07_a9iohBxDhq2iiPsV1YcKjw,10247
 torchx/examples/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -56,13 +56,12 @@ torchx/runner/events/handlers.py,sha256=ThHCIJW21BfBgB7b6ftyjASJmD1KdizpjuTtsyqn
 torchx/runtime/__init__.py,sha256=Wxje2BryzeQneFu5r6P9JJiEKG-_C9W1CcZ_JNrKT6g,593
 torchx/runtime/tracking/__init__.py,sha256=dYnAPnrXYREfPXkpHhdOFkcYIODWEbA13PdD-wLQYBo,3055
 torchx/runtime/tracking/api.py,sha256=SmUQyUKZqG3KlAhT7CJOGqRz1O274E4m63wQeOVq3CU,5472
-torchx/schedulers/__init__.py,sha256=hliMsZHZNOKue0uTHUWxvO0V7xsKApBxN4Wb_9L0Mz4,2253
+torchx/schedulers/__init__.py,sha256=_Wx6-X3FNh8RJR82UGgUwKg7V_VQYsAkrveDoSSk2xU,2195
 torchx/schedulers/api.py,sha256=lfxNhrEO6eYYqVuQzzj9sTXrZShuZkyYxJ1jPE-Lvpo,14561
 torchx/schedulers/aws_batch_scheduler.py,sha256=hFxYzSZEK2SVS5sEyQC5YvNI0JJUJUQsWORlYpj_h3M,28105
 torchx/schedulers/aws_sagemaker_scheduler.py,sha256=flN8GumKE2Dz4X_foAt6Jnvt-ZVojWs6pcyrHwB0hz0,20921
 torchx/schedulers/devices.py,sha256=RjVcu22ZRl_9OKtOtmA1A3vNXgu2qD6A9ST0L0Hsg4I,1734
 torchx/schedulers/docker_scheduler.py,sha256=xuK00-dB6o8TV1YaZox7O5P09LHB2KeQ6t4eiNtqMYQ,16781
-torchx/schedulers/gcp_batch_scheduler.py,sha256=JQuaEJVL_7NSa9AeUc_0Qo74XZNJk_kp6XwgunvlUKI,16281
 torchx/schedulers/ids.py,sha256=3E-_vwVYC-8Tv8kjuY9-W7TbOe_-Laqd8a65uIN3hQY,1798
 torchx/schedulers/kubernetes_mcad_scheduler.py,sha256=1tuzq3OutCMdSPqg_dNmCHt_wyuSFKG0-ywLc3qITJo,42949
 torchx/schedulers/kubernetes_scheduler.py,sha256=0_loGJ7WnxEr9dhgFt3Gw-7nVLirMDVN-MAFTCq7erE,28217
@@ -97,15 +96,15 @@ torchx/util/log_tee_helpers.py,sha256=wPyozmh9BOt_2d3Gxa0iNogwnjzwFitIIMBJOJ1arI
 torchx/util/modules.py,sha256=o4y_d07gTpJ4nIVBcoUVJ0JtXIHEsEC5kbgBM6NGpgA,2135
 torchx/util/session.py,sha256=r6M_nyzXgcbk1GgYGZ324F_ehRGCqjjdVk4YgKxMj8M,1214
 torchx/util/shlex.py,sha256=eXEKu8KC3zIcd8tEy9_s8Ds5oma8BORr-0VGWNpG2dk,463
-torchx/util/strings.py,sha256=GkLWCmYS89Uv6bWc5hH0XwvHy7oQmprv2U7axC4A2e8,678
+torchx/util/strings.py,sha256=7Ef1loz2IYMrzeJ6Lewywi5cBIc3X3g7lSPbT1Tn_z4,664
 torchx/util/types.py,sha256=E9dxAWQnsJkIDuHtg-poeOJ4etucSI_xP_Z5kNJX8uI,9229
 torchx/workspace/__init__.py,sha256=cZsKVvUWwDYcGhe6SCXQGBQfbk_yTnKEImOkI6xmu30,809
 torchx/workspace/api.py,sha256=Ct_75VU94fsH9Rf1WRe-wJGpVgl5O05S_Dq_t2ArJWA,11348
 torchx/workspace/dir_workspace.py,sha256=npNW_IjUZm_yS5r-8hrRkH46ndDd9a_eApT64m1S1T4,2268
 torchx/workspace/docker_workspace.py,sha256=PFu2KQNVC-0p2aKJ-W_BKA9ZOmXdCY2ABEkCExp3udQ,10269
-torchx_nightly-2025.9.21.dist-info/LICENSE,sha256=WVHfXhFC0Ia8LTKt_nJVYobdqTJVg_4J3Crrfm2A8KQ,1721
-torchx_nightly-2025.9.21.dist-info/METADATA,sha256=0KvbX8m2uQZVgOBL_JiKB8nVyXDWBCVFRAypM61NWQU,5693
-torchx_nightly-2025.9.21.dist-info/WHEEL,sha256=tZoeGjtWxWRfdplE7E3d45VPlLNQnvbKiYnx7gwAy8A,92
-torchx_nightly-2025.9.21.dist-info/entry_points.txt,sha256=T328AMXeKI3JZnnxfkEew2ZcMN1oQDtkXjMz7lkV-P4,169
-torchx_nightly-2025.9.21.dist-info/top_level.txt,sha256=pxew3bc2gsiViS0zADs0jb6kC5v8o_Yy_85fhHj_J1A,7
-torchx_nightly-2025.9.21.dist-info/RECORD,,
+torchx_nightly-2025.9.23.dist-info/LICENSE,sha256=WVHfXhFC0Ia8LTKt_nJVYobdqTJVg_4J3Crrfm2A8KQ,1721
+torchx_nightly-2025.9.23.dist-info/METADATA,sha256=Dsh27u65MAyHzYM5eRSPfYEQZmnh9qfsvs1_0vWkhCo,5003
+torchx_nightly-2025.9.23.dist-info/WHEEL,sha256=tZoeGjtWxWRfdplE7E3d45VPlLNQnvbKiYnx7gwAy8A,92
+torchx_nightly-2025.9.23.dist-info/entry_points.txt,sha256=T328AMXeKI3JZnnxfkEew2ZcMN1oQDtkXjMz7lkV-P4,169
+torchx_nightly-2025.9.23.dist-info/top_level.txt,sha256=pxew3bc2gsiViS0zADs0jb6kC5v8o_Yy_85fhHj_J1A,7
+torchx_nightly-2025.9.23.dist-info/RECORD,,

torchx/schedulers/gcp_batch_scheduler.py

@@ -1,497 +0,0 @@
-#!/usr/bin/env python3
-# Copyright (c) Meta Platforms, Inc. and affiliates.
-# All rights reserved.
-#
-# This source code is licensed under the BSD-style license found in the
-# LICENSE file in the root directory of this source tree.
-
-# pyre-strict
-
-"""
-
-This contains the TorchX GCP Batch scheduler which can be used to run TorchX
-components directly on GCP Batch.
-
-This scheduler is in prototype stage and may change without notice.
-
-Prerequisites
-==============
-
-You need to have a GCP project configured to use Batch by enabling and setting it up.
-See https://cloud.google.com/batch/docs/get-started#prerequisites
-
-"""
-
-from dataclasses import dataclass
-from datetime import datetime
-from typing import Any, Dict, Iterable, List, Optional, TYPE_CHECKING, TypedDict
-
-import torchx
-import yaml
-
-from torchx.schedulers.api import (
-    DescribeAppResponse,
-    ListAppResponse,
-    Scheduler,
-    Stream,
-)
-from torchx.schedulers.ids import make_unique
-from torchx.specs.api import (
-    AppDef,
-    AppDryRunInfo,
-    AppState,
-    macros,
-    Resource,
-    Role,
-    runopts,
-)
-from torchx.util.strings import normalize_str
-
-
-if TYPE_CHECKING:
-    from google.cloud import batch_v1
-
-
-JOB_STATE: Dict[str, AppState] = {
-    "STATE_UNSPECIFIED": AppState.UNKNOWN,
-    "QUEUED": AppState.SUBMITTED,
-    "SCHEDULED": AppState.PENDING,
-    "RUNNING": AppState.RUNNING,
-    "SUCCEEDED": AppState.SUCCEEDED,
-    "FAILED": AppState.FAILED,
-    "DELETION_IN_PROGRESS": AppState.UNKNOWN,
-}
-
-GPU_COUNT_TO_TYPE: Dict[int, str] = {
-    1: "a2-highgpu-1g",
-    2: "a2-highgpu-2g",
-    4: "a2-highgpu-4g",
-    8: "a2-highgpu-8g",
-    16: "a2-highgpu-16g",
-}
-
-GPU_TYPE_TO_COUNT: Dict[str, int] = {v: k for k, v in GPU_COUNT_TO_TYPE.items()}
-
-LABEL_VERSION: str = "torchx_version"
-LABEL_APP_NAME: str = "torchx_app_name"
-
-DEFAULT_LOC: str = "us-central1"
-
-# TODO Remove LOCATIONS list once Batch supports all locations
-# or when there is an API to query locations supported by Batch
-LOCATIONS: List[str] = [
-    DEFAULT_LOC,
-    "us-west1",
-    "us-east1",
-    "asia-southeast1",
-    "europe-north1",
-    "europe-west6",
-]
-
-BATCH_LOGGER_NAME = "batch_task_logs"
-
-
-@dataclass
-class GCPBatchJob:
-    name: str
-    project: str
-    location: str
-    job_def: "batch_v1.Job"
-
-    def __str__(self) -> str:
-        return yaml.dump(self.job_def)
-
-    def __repr__(self) -> str:
-        return str(self)
-
-
-class GCPBatchOpts(TypedDict, total=False):
-    project: Optional[str]
-    location: Optional[str]
-
-
-class GCPBatchScheduler(Scheduler[GCPBatchOpts, AppDef, AppDryRunInfo[GCPBatchJob]]):
-    """
-    GCPBatchScheduler is a TorchX scheduling interface to GCP Batch.
-
-    .. code-block:: bash
-
-        $ pip install torchx[gcp_batch]
-        $ torchx run --scheduler gcp_batch utils.echo --msg hello
-        # This launches a job with app handle like gcp_batch://torchx/project:location:app_id1234 and prints it
-        $ torchx status gcp_batch://torchx/project:location:app_id1234
-        ...
-
-    Authentication is loaded from the environment using the gcloud credential handling.
-
-    **Config Options**
-
-    .. runopts::
-        class: torchx.schedulers.gcp_batch_scheduler.create_scheduler
-
-    **Compatibility**
-
-    .. compatibility::
-        type: scheduler
-        features:
-            cancel: true
-            logs: true
-            describe: true
-            distributed: true
-            workspaces: false
-            mounts: false
-            elasticity: false
-
-    """
-
-    def __init__(
-        self,
-        session_name: str,
-        # pyre-fixme[2]: Parameter annotation cannot be `Any`.
-        client: Optional[Any] = None,
-    ) -> None:
-        # NOTE: make sure any new init options are supported in create_scheduler(...)
-        Scheduler.__init__(self, "gcp_batch", session_name)
-        # pyre-fixme[4]: Attribute annotation cannot be `Any`.
-        self.__client = client
-
-    @property
-    # pyre-fixme[3]: Return annotation cannot be `Any`.
-    def _client(self) -> Any:
-        from google.api_core import gapic_v1
-        from google.cloud import batch_v1
-
-        c = self.__client
-        if c is None:
-            client_info = gapic_v1.client_info.ClientInfo(
-                user_agent=f"TorchX/{torchx.__version__}"
-            )
-            c = self.__client = batch_v1.BatchServiceClient(client_info=client_info)
-        return c
-
-    def schedule(self, dryrun_info: AppDryRunInfo[GCPBatchJob]) -> str:
-        from google.cloud import batch_v1
-
-        req = dryrun_info.request
-        assert req is not None, f"{dryrun_info} missing request"
-
-        request = batch_v1.CreateJobRequest(
-            parent=f"projects/{req.project}/locations/{req.location}",
-            job=req.job_def,
-            job_id=req.name,
-        )
-
-        response = self._client.create_job(request=request)
-        return f"{req.project}:{req.location}:{req.name}"
-
-    def _app_to_job(self, app: AppDef) -> "batch_v1.Job":
-        from google.cloud import batch_v1
-
-        name = normalize_str(make_unique(app.name))
-
-        taskGroups = []
-        allocationPolicy = None
-
-        # 1. Convert role to task
-        # TODO implement retry_policy, mount conversion
-        # NOTE: Supports only one role for now as GCP Batch supports only one TaskGroup
-        # which is ok to start with as most components have only one role
-        for role_idx, role in enumerate(app.roles):
-            values = macros.Values(
-                img_root="",
-                app_id=name,
-                replica_id=str(0),
-                rank0_env=("BATCH_MAIN_NODE_HOSTNAME"),
-            )
-            role_dict = values.apply(role)
-            role_dict.env["TORCHX_ROLE_IDX"] = str(role_idx)
-            role_dict.env["TORCHX_ROLE_NAME"] = str(role.name)
-
-            resource = role_dict.resource
-            res = batch_v1.ComputeResource()
-            cpu = resource.cpu
-            if cpu <= 0:
-                cpu = 1
-            MILLI = 1000
-            res.cpu_milli = cpu * MILLI
-            memMB = resource.memMB
-            if memMB < 0:
-                raise ValueError(
-                    f"memMB should to be set to a positive value, got {memMB}"
-                )
-            res.memory_mib = memMB
-
-            # TODO support named resources
-            # Using v100 as default GPU type as a100 does not allow changing count for now
-            # TODO See if there is a better default GPU type
-            if resource.gpu > 0:
-                if resource.gpu not in GPU_COUNT_TO_TYPE:
-                    raise ValueError(
-                        f"gpu should to be set to one of these values: {GPU_COUNT_TO_TYPE.keys()}"
-                    )
-                machineType = GPU_COUNT_TO_TYPE[resource.gpu]
-                allocationPolicy = batch_v1.AllocationPolicy(
-                    instances=[
-                        batch_v1.AllocationPolicy.InstancePolicyOrTemplate(
-                            install_gpu_drivers=True,
-                            policy=batch_v1.AllocationPolicy.InstancePolicy(
-                                machine_type=machineType,
-                            ),
-                        )
-                    ],
-                )
-                print(f"Using GPUs of type: {machineType}")
-
-            # Configure host firewall rules to accept ingress communication
-            config_network_runnable = batch_v1.Runnable(
-                script=batch_v1.Runnable.Script(
-                    text="/sbin/iptables -A INPUT -j ACCEPT"
-                )
-            )
-
-            runnable = batch_v1.Runnable(
-                container=batch_v1.Runnable.Container(
-                    image_uri=role_dict.image,
-                    commands=[role_dict.entrypoint] + role_dict.args,
-                    entrypoint="",
-                    # Configure docker to use the host network stack to communicate with containers/other hosts in the same network
-                    options="--net host",
-                )
-            )
-
-            ts = batch_v1.TaskSpec(
-                runnables=[config_network_runnable, runnable],
-                environment=batch_v1.Environment(variables=role_dict.env),
-                max_retry_count=role_dict.max_retries,
-                compute_resource=res,
-            )
-
-            task_env = [
-                batch_v1.Environment(variables={"TORCHX_REPLICA_IDX": str(i)})
-                for i in range(role_dict.num_replicas)
-            ]
-
-            tg = batch_v1.TaskGroup(
-                task_spec=ts,
-                task_count=role_dict.num_replicas,
-                task_count_per_node=1,
-                task_environments=task_env,
-                require_hosts_file=True,
-            )
-            taskGroups.append(tg)
-
-        # 2. Convert AppDef to Job
-        job = batch_v1.Job(
-            name=name,
-            task_groups=taskGroups,
-            allocation_policy=allocationPolicy,
-            logs_policy=batch_v1.LogsPolicy(
-                destination=batch_v1.LogsPolicy.Destination.CLOUD_LOGGING,
-            ),
-            # NOTE: GCP Batch does not allow label names with "."
-            labels={
-                LABEL_VERSION: torchx.__version__.replace(".", "-"),
-                LABEL_APP_NAME: name,
-            },
-        )
-        return job
-
-    def _get_project(self) -> str:
-        from google.cloud import runtimeconfig
-
-        return runtimeconfig.Client().project
-
-    def _submit_dryrun(
-        self, app: AppDef, cfg: GCPBatchOpts
-    ) -> AppDryRunInfo[GCPBatchJob]:
-        proj = cfg.get("project")
-        if proj is None:
-            proj = self._get_project()
-        assert proj is not None and isinstance(proj, str), "project must be a str"
-
-        loc = cfg.get("location")
-        assert loc is not None and isinstance(loc, str), "location must be a str"
-
-        job = self._app_to_job(app)
-
-        # Convert JobDef + BatchOpts to GCPBatchJob
-        req = GCPBatchJob(
-            name=str(job.name),
-            project=proj,
-            location=loc,
-            job_def=job,
-        )
-
-        return AppDryRunInfo(req, repr)
-
-    def run_opts(self) -> runopts:
-        opts = runopts()
-        opts.add(
-            "project",
-            type_=str,
-            help="Name of the GCP project. Defaults to the configured GCP project in the environment",
-        )
-        opts.add(
-            "location",
-            type_=str,
-            default=DEFAULT_LOC,
-            help=f"Name of the location to schedule the job in. Defaults to {DEFAULT_LOC}",
-        )
-        return opts
-
-    def _app_id_to_job_full_name(self, app_id: str) -> str:
-        """
-        app_id format: f"{project}:{location}:{name}"
-        job_full_name format: f"projects/{project}/locations/{location}/jobs/{name}"
-        where 'name' was created uniquely for the job from the app name
-        """
-        app_id_splits = app_id.split(":")
-        if len(app_id_splits) != 3:
-            raise ValueError(f"app_id not in expected format: {app_id}")
-        return f"projects/{app_id_splits[0]}/locations/{app_id_splits[1]}/jobs/{app_id_splits[2]}"
-
-    def _get_job(self, app_id: str) -> "batch_v1.Job":
-        from google.cloud import batch_v1
-
-        job_name = self._app_id_to_job_full_name(app_id)
-        request = batch_v1.GetJobRequest(
-            name=job_name,
-        )
-        return self._client.get_job(request=request)
-
-    def describe(self, app_id: str) -> Optional[DescribeAppResponse]:
-        job = self._get_job(app_id)
-        if job is None:
-            print(f"app not found: {app_id}")
-            return None
-
-        gpu = 0
-        if len(job.allocation_policy.instances) != 0:
-            gpu_type = job.allocation_policy.instances[0].policy.machine_type
-            gpu = GPU_TYPE_TO_COUNT[gpu_type]
-
-        roles = {}
-        for tg in job.task_groups:
-            env = tg.task_spec.environment.variables
-            role = env["TORCHX_ROLE_NAME"]
-            container = tg.task_spec.runnables[1].container
-            roles[role] = Role(
-                name=role,
-                num_replicas=tg.task_count,
-                image=container.image_uri,
-                entrypoint=container.commands[0],
-                args=list(container.commands[1:]),
-                resource=Resource(
-                    cpu=int(tg.task_spec.compute_resource.cpu_milli / 1000),
-                    memMB=tg.task_spec.compute_resource.memory_mib,
-                    gpu=gpu,
-                ),
-                env=dict(env),
-                max_retries=tg.task_spec.max_retry_count,
-            )
-
-        # Map job -> DescribeAppResponse
-        # TODO map role/replica status
-        desc = DescribeAppResponse(
-            app_id=app_id,
-            state=JOB_STATE[job.status.state.name],
-            roles=list(roles.values()),
-        )
-        return desc
-
-    def log_iter(
-        self,
-        app_id: str,
-        role_name: str = "",
-        k: int = 0,
-        regex: Optional[str] = None,
-        since: Optional[datetime] = None,
-        until: Optional[datetime] = None,
-        should_tail: bool = False,
-        streams: Optional[Stream] = None,
-    ) -> Iterable[str]:
-        if streams not in (None, Stream.COMBINED):
-            raise ValueError("GCPBatchScheduler only supports COMBINED log stream")
-
-        job = self._get_job(app_id)
-        if not job:
-            raise ValueError(f"app not found: {app_id}")
-
-        job_uid = job.uid
-        filters = [
-            f"labels.job_uid={job_uid}",
-            f"labels.task_id:{job_uid}-group0-{k}",
-        ]
-
-        if since is not None:
-            filters.append(f'timestamp>="{str(since.isoformat())}"')
-        else:
-            # gcloud logger.list by default only returns logs in the last 24 hours
-            # Since many ML jobs can run longer add timestamp filter to get all logs
-            filters.append(f'timestamp>="{str(datetime.fromtimestamp(0).isoformat())}"')
-
-        if until is not None:
-            filters.append(f'timestamp<="{str(until.isoformat())}"')
-        if regex is not None:
-            filters.append(f'textPayload =~ "{regex}"')
-        filter = " AND ".join(filters)
-        return self._batch_log_iter(filter)
-
-    def _batch_log_iter(self, filter: str) -> Iterable[str]:
-        from google.cloud import logging
-
-        logger = logging.Client().logger(BATCH_LOGGER_NAME)
-        for entry in logger.list_entries(filter_=filter):
-            yield entry.payload + "\n"
-
-    def _job_full_name_to_app_id(self, job_full_name: str) -> str:
-        """
-        job_full_name format: f"projects/{project}/locations/{location}/jobs/{name}"
-        app_id format: f"{project}:{location}:{name}"
-        where 'name' was created uniquely for the job from the app name
-        """
-        job_name_splits = job_full_name.split("/")
-        if len(job_name_splits) != 6:
-            raise ValueError(f"job full name not in expected format: {job_full_name}")
-        return f"{job_name_splits[1]}:{job_name_splits[3]}:{job_name_splits[5]}"
-
-    def list(self) -> List[ListAppResponse]:
-        all_jobs = []
-        proj = self._get_project()
-        for loc in LOCATIONS:
-            jobs = self._client.list_jobs(parent=f"projects/{proj}/locations/{loc}")
-            all_jobs += jobs
-        all_jobs.sort(key=lambda job: job.create_time.timestamp(), reverse=True)
-        return [
-            ListAppResponse(
-                app_id=self._job_full_name_to_app_id(job.name),
-                state=JOB_STATE[job.status.state.name],
-            )
-            for job in all_jobs
-        ]
-
-    def _validate(self, app: AppDef, scheduler: str, cfg: GCPBatchOpts) -> None:
-        # Skip validation step
-        pass
-
-    def _cancel_existing(self, app_id: str) -> None:
-        from google.cloud import batch_v1
-
-        job_name = self._app_id_to_job_full_name(app_id)
-        request = batch_v1.DeleteJobRequest(
-            name=job_name,
-            reason="Killed via TorchX",
-        )
-        self._client.delete_job(request=request)
-
-
-def create_scheduler(
-    session_name: str,
-    # pyre-fixme[2]: Parameter annotation cannot be `Any`.
-    client: Optional[Any] = None,
-    **kwargs: object,
-) -> GCPBatchScheduler:
-    return GCPBatchScheduler(
-        session_name=session_name,
-        client=client,
-    )