Load Experiment without runners and metrics in the case where search …

…space and optimization config are immutable

Summary: `load_experiment` was previously failing when `skip_runners_and_metrics=True` for experiments with immutable search space and optimization config. See [Lena's comment](https://www.internalfb.com/diff/D46595953?dst_version_fbid=263061986396665&transaction_fbid=639853311380999) for more detail.

Differential Revision: D46595953

fbshipit-source-id: 7591b2d968fdf53bb86f96f6cdfe455d0e094217

ax/storage/sqa_store/load.py

@@ -28,6 +28,7 @@
 from ax.utils.common.constants import Keys
 from ax.utils.common.typeutils import not_none
 from sqlalchemy.orm import defaultload, lazyload, noload
+from sqlalchemy.orm.exc import DetachedInstanceError
 
 
 # ---------------------------- Loading `Experiment`. ---------------------------
@@ -135,10 +136,13 @@ def _load_experiment(
         for sqa_metric in experiment_sqa.metrics:
             sqa_metric.metric_type = 0
 
-        for sqa_trial in experiment_sqa.trials:
-            for sqa_generator_run in sqa_trial.generator_runs:
-                for sqa_metric in sqa_generator_run.metrics:
-                    sqa_metric.metric_type = 0
+        try:
+            for sqa_trial in experiment_sqa.trials:
+                for sqa_generator_run in sqa_trial.generator_runs:
+                    for sqa_metric in sqa_generator_run.metrics:
+                        sqa_metric.metric_type = 0
+        except DetachedInstanceError:
+            pass
 
     return decoder.experiment_from_sqa(
         experiment_sqa=experiment_sqa,