wandb 0.17.6__py3-none-win32.whl → 0.17.8rc1__py3-none-win32.whl

wandb/sdk/launch/builder/kaniko_builder.py

@@ -63,6 +63,13 @@ else:
     NAMESPACE = "wandb"
 
 
+def get_pod_name_safe(job: client.V1Job):
+    try:
+        return job.spec.template.metadata.name
+    except AttributeError:
+        return None
+
+
 async def _wait_for_completion(
     batch_client: client.BatchV1Api, job_name: str, deadline_secs: Optional[int] = None
 ) -> bool:
@@ -319,17 +326,18 @@ class KanikoBuilder(AbstractBuilder):
                 await self._create_docker_ecr_config_map(
                     build_job_name, core_v1, repo_uri
                 )
-            await batch_v1.create_namespaced_job(NAMESPACE, build_job)
-
+            k8s_job = await batch_v1.create_namespaced_job(NAMESPACE, build_job)
             # wait for double the job deadline since it might take time to schedule
             if not await _wait_for_completion(
                 batch_v1, build_job_name, 3 * _DEFAULT_BUILD_TIMEOUT_SECS
             ):
                 if job_tracker:
                     job_tracker.set_err_stage("build")
-                raise Exception(
-                    f"Failed to build image in kaniko for job {run_id}. View logs with `kubectl logs -n {NAMESPACE} {build_job_name}`."
-                )
+                msg = f"Failed to build image in kaniko for job {run_id}."
+                pod_name = get_pod_name_safe(k8s_job)
+                if pod_name:
+                    msg += f" View logs with `kubectl logs -n {NAMESPACE} {pod_name}`."
+                raise Exception(msg)
             try:
                 pods_from_job = await core_v1.list_namespaced_pod(
                     namespace=NAMESPACE, label_selector=f"job-name={build_job_name}"

wandb/sdk/launch/create_job.py

@@ -114,6 +114,7 @@ def _create_job(
     git_hash: Optional[str] = None,
     build_context: Optional[str] = None,
     dockerfile: Optional[str] = None,
+    base_image: Optional[str] = None,
 ) -> Tuple[Optional[Artifact], str, List[str]]:
     wandb.termlog(f"Creating launch job of type: {job_type}...")
 
@@ -188,6 +189,7 @@ def _create_job(
         api.api,
         dockerfile=dockerfile,
         build_context=build_context,
+        base_image=base_image,
     )
     if not artifact:
         wandb.termerror("JobBuilder failed to build a job")

wandb/sdk/lib/apikey.py

@@ -107,7 +107,7 @@ def prompt_api_key(  # noqa: C901
 
     if jupyter and "google.colab" in sys.modules:
         log_string = term.LOG_STRING_NOCOLOR
-        key = wandb.jupyter.attempt_colab_login(app_url)
+        key = wandb.jupyter.attempt_colab_login(app_url)  # type: ignore
         if key is not None:
             write_key(settings, key, api=api)
             return key  # type: ignore

wandb/sdk/service/streams.py

@@ -81,9 +81,7 @@ class StreamRecord:
         self._wait_thread_active()
 
     def _wait_thread_active(self) -> None:
-        result = self._iface.communicate_status()
-        # TODO: using the default communicate timeout, is that enough? retries?
-        assert result
+        self._iface.deliver_status().wait(timeout=-1)
 
     def join(self) -> None:
         self._iface.join()
@@ -212,7 +210,7 @@ class StreamMux:
         # run_id = action.stream_id  # will want to fix if a streamid != runid
         settings = action._data
         thread = StreamThread(
-            target=wandb.wandb_sdk.internal.internal.wandb_internal,
+            target=wandb.wandb_sdk.internal.internal.wandb_internal,  # type: ignore
             kwargs=dict(
                 settings=settings,
                 record_q=stream._record_q,

wandb/sdk/wandb_config.py

@@ -61,7 +61,7 @@ class Config:
 
         Using absl flags
         ```
-        flags.DEFINE_string(‘model’, None, ‘model to run’) # name, default, help
+        flags.DEFINE_string("model", None, "model to run") # name, default, help
         wandb.config.update(flags.FLAGS) # adds all absl flags to config
         ```
 
@@ -129,6 +129,9 @@ class Config:
     def __getitem__(self, key):
         return self._items[key]
 
+    def __iter__(self):
+        return iter(self._items)
+
     def _check_locked(self, key, ignore_locked=False) -> bool:
         locked = self._locked.get(key)
         if locked is not None:

wandb/sdk/wandb_init.py

@@ -15,6 +15,7 @@ import os
 import platform
 import sys
 import tempfile
+import time
 from typing import TYPE_CHECKING, Any, Dict, List, Optional, Sequence, Union
 
 import wandb
@@ -174,7 +175,9 @@ class _WandbInit:
         # we add this logic to be backward compatible with the old behavior of disable
         # where it would disable the service if the mode was set to disabled
         mode = kwargs.get("mode")
-        settings_mode = (kwargs.get("settings") or {}).get("mode")
+        settings_mode = (kwargs.get("settings") or {}).get("mode") or os.environ.get(
+            "WANDB_MODE"
+        )
         _disable_service = mode == "disabled" or settings_mode == "disabled"
         setup_settings = {"_disable_service": _disable_service}
 
@@ -262,7 +265,7 @@ class _WandbInit:
 
         monitor_gym = kwargs.pop("monitor_gym", None)
         if monitor_gym and len(wandb.patched["gym"]) == 0:
-            wandb.gym.monitor()
+            wandb.gym.monitor()  # type: ignore
 
         if wandb.patched["tensorboard"]:
             with telemetry.context(obj=self._init_telemetry_obj) as tel:
@@ -271,7 +274,7 @@ class _WandbInit:
         tensorboard = kwargs.pop("tensorboard", None)
         sync_tensorboard = kwargs.pop("sync_tensorboard", None)
         if tensorboard or sync_tensorboard and len(wandb.patched["tensorboard"]) == 0:
-            wandb.tensorboard.patch()
+            wandb.tensorboard.patch()  # type: ignore
             with telemetry.context(obj=self._init_telemetry_obj) as tel:
                 tel.feature.tensorboard_sync = True
 
@@ -459,7 +462,7 @@ class _WandbInit:
 
     def _jupyter_setup(self, settings: Settings) -> None:
         """Add hooks, and session history saving."""
-        self.notebook = wandb.jupyter.Notebook(settings)
+        self.notebook = wandb.jupyter.Notebook(settings)  # type: ignore
         ipython = self.notebook.shell
 
         # Monkey patch ipython publish to capture displayed outputs
@@ -522,17 +525,62 @@ class _WandbInit:
         logger.info(f"Logging internal logs to {settings.log_internal}")
 
     def _make_run_disabled(self) -> Run:
+        """Returns a Run-like object where all methods are no-ops.
+
+        This method is used when wandb.init(mode="disabled") is called or WANDB_MODE=disabled
+        is set. It creates a Run object that mimics the behavior of a normal Run but doesn't
+        communicate with the W&B servers.
+
+        The returned Run object has all expected attributes and methods, but they are
+        no-op versions that don't perform any actual logging or communication.
+        """
         drun = Run(settings=Settings(mode="disabled", files_dir=tempfile.gettempdir()))
-        drun._config = wandb.wandb_sdk.wandb_config.Config()
+        # config and summary objects
+        drun._config = wandb.sdk.wandb_config.Config()
         drun._config.update(self.sweep_config)
         drun._config.update(self.config)
         drun.summary = SummaryDisabled()  # type: ignore
+        # methods
         drun.log = lambda data, *_, **__: drun.summary.update(data)  # type: ignore
         drun.finish = lambda *_, **__: module.unset_globals()  # type: ignore
+        drun.join = drun.finish  # type: ignore
+        drun.define_metric = lambda *_, **__: wandb.sdk.wandb_metric.Metric("dummy")  # type: ignore
+        drun.save = lambda *_, **__: False  # type: ignore
+        for symbol in (
+            "alert",
+            "finish_artifact",
+            "get_project_url",
+            "get_sweep_url",
+            "get_url",
+            "link_artifact",
+            "link_model",
+            "use_artifact",
+            "log_artifact",
+            "log_code",
+            "log_model",
+            "use_model",
+            "mark_preempting",
+            "plot_table",
+            "restore",
+            "status",
+            "watch",
+            "unwatch",
+            "upsert_artifact",
+        ):
+            setattr(drun, symbol, lambda *_, **__: None)  # type: ignore
+        # attributes
         drun._step = 0
+        drun._attach_id = None
         drun._run_obj = None
         drun._run_id = runid.generate_id()
         drun._name = "dummy-" + drun.id
+        drun._project = "dummy"
+        drun._entity = "dummy"
+        drun._tags = tuple()
+        drun._notes = None
+        drun._group = None
+        drun._start_time = time.time()
+        drun._starting_step = 0
         module.set_global(
             run=drun,
             config=drun.config,
@@ -868,7 +916,7 @@ def _attach(
         raise UsageError(
             "Either `attach_id` or `run_id` must be specified or `run` must have `_attach_id`"
         )
-    wandb._assert_is_user_process()
+    wandb._assert_is_user_process()  # type: ignore
 
     _wl = wandb_setup._setup()
     assert _wl
@@ -1157,7 +1205,7 @@ def init(
     Returns:
         A `Run` object.
     """
-    wandb._assert_is_user_process()
+    wandb._assert_is_user_process()  # type: ignore
 
     kwargs = dict(locals())