Remove config from task manager and stop killing it

azimuth/app.py

@@ -18,7 +18,6 @@
 from azimuth.startup import startup_tasks
 from azimuth.task_manager import TaskManager
 from azimuth.types import DatasetSplitName, ModuleOptions
-from azimuth.utils.cluster import default_cluster
 from azimuth.utils.conversion import JSONResponseIgnoreNan
 from azimuth.utils.logs import set_logger_config
 from azimuth.utils.project import load_dataset_split_managers_from_config, save_config
@@ -101,9 +100,7 @@ def start_app(config_path: Optional[str], load_config_history: bool, debug: bool
     if azimuth_config.dataset is None:
         raise ValueError("No dataset has been specified in the config.")
 
-    local_cluster = default_cluster(large=azimuth_config.large_dask_cluster)
-
-    run_startup_tasks(azimuth_config, local_cluster)
+    run_startup_tasks(azimuth_config)
     assert_not_none(_task_manager).client.run(set_logger_config, level)
 
     app = create_app()
@@ -240,25 +237,23 @@ def create_app() -> FastAPI:
     return app
 
 
-def initialize_managers(azimuth_config: AzimuthConfig, cluster: SpecCluster):
-    """Initialize DatasetSplitManagers and TaskManagers.
-
+def initialize_managers_and_config(
+    azimuth_config: AzimuthConfig, cluster: Optional[SpecCluster] = None
+):
+    """Initialize DatasetSplitManagers and Config.
 
     Args:
-        azimuth_config: Configuration
-        cluster: Dask cluster to use.
+        azimuth_config: Config
+        cluster: Dask cluster to use, if different than default.
     """
     global _task_manager, _dataset_split_managers, _azimuth_config
-    _azimuth_config = azimuth_config
-    if _task_manager is not None:
-        task_history = _task_manager.current_tasks
+    if _task_manager:
+        _task_manager.clear_worker_cache()
+        _task_manager.restart()
     else:
-        task_history = {}
-
-    _task_manager = TaskManager(azimuth_config, cluster=cluster)
-
-    _task_manager.current_tasks = task_history
+        _task_manager = TaskManager(cluster, azimuth_config.large_dask_cluster)
 
+    _azimuth_config = azimuth_config
     _dataset_split_managers = load_dataset_split_managers_from_config(azimuth_config)
 
 
@@ -295,15 +290,14 @@ def run_validation_module(pipeline_index=None):
     task_manager.restart()
 
 
-def run_startup_tasks(azimuth_config: AzimuthConfig, cluster: SpecCluster):
+def run_startup_tasks(azimuth_config: AzimuthConfig, cluster: Optional[SpecCluster] = None):
     """Initialize managers, run validation and startup tasks.
 
     Args:
         azimuth_config: Config
-        cluster: Cluster
-
+        cluster: Dask cluster to use, if different than default.
     """
-    initialize_managers(azimuth_config, cluster)
+    initialize_managers_and_config(azimuth_config, cluster)
 
     task_manager = assert_not_none(get_task_manager())
     # Validate that everything is in order **before** the startup tasks.
@@ -315,5 +309,5 @@ def run_startup_tasks(azimuth_config: AzimuthConfig, cluster: SpecCluster):
     save_config(azimuth_config)  # Save only after the validation modules ran successfully
 
     global _startup_tasks, _ready_flag
-    _startup_tasks = startup_tasks(_dataset_split_managers, task_manager)
+    _startup_tasks = startup_tasks(_dataset_split_managers, task_manager, azimuth_config)
     _ready_flag = Event()

azimuth/routers/app.py

@@ -84,15 +84,12 @@ def get_dataset_info(
         get_dataset_split_manager_mapping
     ),
     startup_tasks: Dict[str, Module] = Depends(get_startup_tasks),
-    task_manager: TaskManager = Depends(get_task_manager),
     config: AzimuthConfig = Depends(get_config),
 ):
     eval_dm = dataset_split_managers.get(DatasetSplitName.eval)
     training_dm = dataset_split_managers.get(DatasetSplitName.train)
     dm = assert_not_none(eval_dm or training_dm)
 
-    model_contract = task_manager.config.model_contract
-
     return DatasetInfoResponse(
         project_name=config.name,
         class_names=dm.get_class_names(),
@@ -105,19 +102,16 @@ def get_dataset_info(
         if training_dm is not None
         else [],
         startup_tasks={k: v.status() for k, v in startup_tasks.items()},
-        model_contract=model_contract,
-        prediction_available=predictions_available(task_manager.config),
-        perturbation_testing_available=perturbation_testing_available(task_manager.config),
+        model_contract=config.model_contract,
+        prediction_available=predictions_available(config),
+        perturbation_testing_available=perturbation_testing_available(config),
         available_dataset_splits=AvailableDatasetSplits(
             eval=eval_dm is not None, train=training_dm is not None
         ),
-        similarity_available=similarity_available(task_manager.config),
+        similarity_available=similarity_available(config),
         postprocessing_editable=None
         if config.pipelines is None
-        else [
-            postprocessing_editable(task_manager.config, idx)
-            for idx in range(len(config.pipelines))
-        ],
+        else [postprocessing_editable(config, idx) for idx in range(len(config.pipelines))],
     )
 
 
@@ -171,6 +165,7 @@ def get_perturbation_testing_summary(
             SupportedModule.PerturbationTestingMerged,
             dataset_split_name=DatasetSplitName.all,
             task_manager=task_manager,
+            config=config,
             last_update=last_update,
             mod_options=ModuleOptions(pipeline_index=pipeline_index),
         )[0]
@@ -186,6 +181,7 @@ def get_perturbation_testing_summary(
         SupportedModule.PerturbationTestingSummary,
         dataset_split_name=DatasetSplitName.all,
         task_manager=task_manager,
+        config=config,
         mod_options=ModuleOptions(pipeline_index=pipeline_index),
     )[0]
     return PerturbationTestingSummary(

azimuth/routers/class_overlap.py

@@ -72,6 +72,7 @@ def get_class_overlap_plot(
         SupportedModule.ClassOverlap,
         dataset_split_name=dataset_split_name,
         task_manager=task_manager,
+        config=config,
         last_update=-1,
     )[0]
     class_overlap_plot_response: ClassOverlapPlotResponse = make_sankey_plot(
@@ -94,6 +95,7 @@ def get_class_overlap_plot(
 def get_class_overlap(
     dataset_split_name: DatasetSplitName,
     task_manager: TaskManager = Depends(get_task_manager),
+    config: AzimuthConfig = Depends(get_config),
     dataset_split_manager: DatasetSplitManager = Depends(get_dataset_split_manager),
     dataset_split_managers: Dict[DatasetSplitName, DatasetSplitManager] = Depends(
         get_dataset_split_manager_mapping
@@ -106,6 +108,7 @@ def get_class_overlap(
         SupportedModule.ClassOverlap,
         dataset_split_name=dataset_split_name,
         task_manager=task_manager,
+        config=config,
         last_update=-1,
     )[0]
     dataset_class_count = class_overlap_result.s_matrix.shape[0]
@@ -121,6 +124,7 @@ def get_class_overlap(
             SupportedModule.ConfusionMatrix,
             DatasetSplitName.eval,
             task_manager=task_manager,
+            config=config,
             mod_options=ModuleOptions(
                 pipeline_index=pipeline_index, cf_normalize=False, cf_reorder_classes=False
             ),

azimuth/routers/config.py

@@ -11,7 +11,7 @@
 from azimuth.app import (
     get_config,
     get_task_manager,
-    initialize_managers,
+    initialize_managers_and_config,
     require_editable_config,
     run_startup_tasks,
 )
@@ -81,11 +81,11 @@ def patch_config(
 
     try:
         new_config = update_config(old_config=config, partial_config=partial_config)
-        run_startup_tasks(new_config, task_manager.cluster)
+        run_startup_tasks(new_config)
     except Exception as e:
         log.error("Rollback config update due to error", exc_info=e)
         new_config = config
-        initialize_managers(new_config, task_manager.cluster)
+        initialize_managers_and_config(new_config)
         if isinstance(e, AzimuthValidationError):
             raise HTTPException(HTTP_400_BAD_REQUEST, detail=str(e))
         elif isinstance(e, ValidationError):

azimuth/routers/custom_utterances.py

@@ -95,11 +95,13 @@ def get_saliency(
     utterances: List[str] = Query([], title="Utterances"),
     pipeline_index: int = Depends(require_pipeline_index),
     task_manager: TaskManager = Depends(get_task_manager),
+    config: AzimuthConfig = Depends(get_config),
 ) -> List[SaliencyResponse]:
     task_result: List[SaliencyResponse] = get_custom_task_result(
         SupportedMethod.Saliency,
         task_manager=task_manager,
-        custom_query={task_manager.config.columns.text_input: utterances},
+        config=config,
+        custom_query={config.columns.text_input: utterances},
         mod_options=ModuleOptions(pipeline_index=pipeline_index),
     )
 

azimuth/routers/dataset_warnings.py

@@ -6,7 +6,8 @@
 
 from fastapi import APIRouter, Depends
 
-from azimuth.app import get_dataset_split_manager_mapping, get_task_manager
+from azimuth.app import get_config, get_dataset_split_manager_mapping, get_task_manager
+from azimuth.config import AzimuthConfig
 from azimuth.dataset_split_manager import DatasetSplitManager
 from azimuth.task_manager import TaskManager
 from azimuth.types import DatasetSplitName, SupportedModule
@@ -25,6 +26,7 @@
 )
 def get_dataset_warnings(
     task_manager: TaskManager = Depends(get_task_manager),
+    config: AzimuthConfig = Depends(get_config),
     dataset_split_managers: Dict[DatasetSplitName, DatasetSplitManager] = Depends(
         get_dataset_split_manager_mapping
     ),
@@ -35,6 +37,7 @@ def get_dataset_warnings(
         dataset_split_name=DatasetSplitName.all,
         task_manager=task_manager,
         last_update=get_last_update(list(dataset_split_managers.values())),
+        config=config
     )[0]
 
     return task_result.warning_groups

azimuth/routers/export.py

@@ -108,18 +108,18 @@ def get_export_perturbation_testing_summary(
         SupportedModule.PerturbationTestingSummary,
         DatasetSplitName.all,
         task_manager=task_manager,
+        config=config,
         last_update=last_update,
         mod_options=ModuleOptions(pipeline_index=pipeline_index),
     )[0].all_tests_summary
 
-    cfg = task_manager.config
     df = pd.DataFrame.from_records([t.dict() for t in task_result])
     df["example"] = df["example"].apply(lambda i: i["perturbedUtterance"])
     file_label = time.strftime("%Y%m%d_%H%M%S", time.localtime())
 
-    filename = f"azimuth_export_behavioral_testing_summary_{cfg.name}_{file_label}.csv"
+    filename = f"azimuth_export_behavioral_testing_summary_{config.name}_{file_label}.csv"
 
-    path = pjoin(cfg.get_artifact_path(), filename)
+    path = pjoin(config.get_artifact_path(), filename)
 
     df.to_csv(path, index=False)
 
@@ -143,15 +143,15 @@ def get_export_perturbed_set(
 ) -> FileResponse:
     pipeline_index_not_null = assert_not_none(pipeline_index)
     file_label = time.strftime("%Y%m%d_%H%M%S", time.localtime())
-    cfg = task_manager.config
 
-    filename = f"azimuth_export_modified_set_{cfg.name}_{dataset_split_name}_{file_label}.json"
-    path = pjoin(cfg.get_artifact_path(), filename)
+    filename = f"azimuth_export_modified_set_{config.name}_{dataset_split_name}_{file_label}.json"
+    path = pjoin(config.get_artifact_path(), filename)
 
     task_result: List[List[PerturbedUtteranceResult]] = get_standard_task_result(
         SupportedModule.PerturbationTesting,
         dataset_split_name,
         task_manager,
+        config=config,
         mod_options=ModuleOptions(pipeline_index=pipeline_index_not_null),
     )
 

azimuth/routers/model_performance/confidence_histogram.py

@@ -4,7 +4,8 @@
 
 from fastapi import APIRouter, Depends, Query
 
-from azimuth.app import get_dataset_split_manager, get_task_manager
+from azimuth.app import get_config, get_dataset_split_manager, get_task_manager
+from azimuth.config import AzimuthConfig
 from azimuth.dataset_split_manager import DatasetSplitManager
 from azimuth.task_manager import TaskManager
 from azimuth.types import (
@@ -33,6 +34,7 @@ def get_confidence_histogram(
     dataset_split_name: DatasetSplitName,
     named_filters: NamedDatasetFilters = Depends(build_named_dataset_filters),
     task_manager: TaskManager = Depends(get_task_manager),
+    config: AzimuthConfig = Depends(get_config),
     dataset_split_manager: DatasetSplitManager = Depends(get_dataset_split_manager),
     pipeline_index: int = Depends(require_pipeline_index),
     without_postprocessing: bool = Query(False, title="Without Postprocessing"),
@@ -47,6 +49,7 @@ def get_confidence_histogram(
         task_name=SupportedModule.ConfidenceHistogram,
         dataset_split_name=dataset_split_name,
         task_manager=task_manager,
+        config=config,
         mod_options=mod_options,
         last_update=dataset_split_manager.last_update,
     )[0]

azimuth/routers/model_performance/confusion_matrix.py

@@ -4,7 +4,8 @@
 
 from fastapi import APIRouter, Depends, Query
 
-from azimuth.app import get_dataset_split_manager, get_task_manager
+from azimuth.app import get_config, get_dataset_split_manager, get_task_manager
+from azimuth.config import AzimuthConfig
 from azimuth.dataset_split_manager import DatasetSplitManager
 from azimuth.task_manager import TaskManager
 from azimuth.types import (
@@ -33,6 +34,7 @@ def get_confusion_matrix(
     dataset_split_name: DatasetSplitName,
     named_filters: NamedDatasetFilters = Depends(build_named_dataset_filters),
     task_manager: TaskManager = Depends(get_task_manager),
+    config: AzimuthConfig = Depends(get_config),
     dataset_split_manager: DatasetSplitManager = Depends(get_dataset_split_manager),
     pipeline_index: int = Depends(require_pipeline_index),
     without_postprocessing: bool = Query(False, title="Without Postprocessing"),
@@ -51,6 +53,7 @@ def get_confusion_matrix(
         SupportedModule.ConfusionMatrix,
         dataset_split_name,
         task_manager=task_manager,
+        config=config,
         mod_options=mod_options,
         last_update=dataset_split_manager.last_update,
     )[0]

azimuth/routers/model_performance/metrics.py

@@ -5,7 +5,8 @@
 
 from fastapi import APIRouter, Depends, Query
 
-from azimuth.app import get_dataset_split_manager, get_task_manager
+from azimuth.app import get_config, get_dataset_split_manager, get_task_manager
+from azimuth.config import AzimuthConfig
 from azimuth.dataset_split_manager import DatasetSplitManager
 from azimuth.modules.model_performance.metrics import MetricsModule
 from azimuth.task_manager import TaskManager
@@ -41,6 +42,7 @@ def get_metrics(
     dataset_split_name: DatasetSplitName,
     named_filters: NamedDatasetFilters = Depends(build_named_dataset_filters),
     task_manager: TaskManager = Depends(get_task_manager),
+    config: AzimuthConfig = Depends(get_config),
     dataset_split_manager: DatasetSplitManager = Depends(get_dataset_split_manager),
     pipeline_index: int = Depends(require_pipeline_index),
     without_postprocessing: bool = Query(False, title="Without Postprocessing"),
@@ -55,6 +57,7 @@ def get_metrics(
         SupportedModule.Metrics,
         dataset_split_name,
         task_manager,
+        config=config,
         mod_options=mod_options,
         last_update=dataset_split_manager.last_update,
     )
@@ -73,6 +76,7 @@ def get_metrics(
 def get_metrics_per_filter(
     dataset_split_name: DatasetSplitName,
     task_manager: TaskManager = Depends(get_task_manager),
+    config: AzimuthConfig = Depends(get_config),
     dataset_split_manager: DatasetSplitManager = Depends(get_dataset_split_manager),
     pipeline_index: int = Depends(require_pipeline_index),
 ) -> MetricsPerFilterAPIResponse:
@@ -81,6 +85,7 @@ def get_metrics_per_filter(
         SupportedModule.MetricsPerFilter,
         dataset_split_name,
         task_manager,
+        config=config,
         mod_options=mod_options,
         last_update=dataset_split_manager.last_update,
     )[0]
@@ -89,6 +94,7 @@ def get_metrics_per_filter(
         SupportedModule.Metrics,
         dataset_split_name,
         task_manager,
+        config=config,
         mod_options=mod_options,
         last_update=dataset_split_manager.last_update,
     )[0]