Add Orbit beacon metric and retention policy.

PiperOrigin-RevId: 569314229

Author: tensorflower-gardener

orbit/controller.py

@@ -26,6 +26,14 @@
 
 import tensorflow as tf
 
+# pylint: disable=g-direct-tensorflow-import
+from tensorflow.python.eager import monitoring
+# pylint: enable=g-direct-tensorflow-import
+
+_orbit_api_gauge = monitoring.BoolGauge(
+    "/tensorflow/api/orbit", "orbit api usage"
+)
+
 
 def _log(message: str):
   """Logs `message` to the `info` log, and also prints to stdout."""
@@ -243,6 +251,9 @@ def __init__(
       if restored_path:
         _log(f"restored from checkpoint: {restored_path}")
 
+    # Set Orbit framework gauge to True value
+    _orbit_api_gauge.get_cell().set(True)
+
   def train(self, steps: int, checkpoint_at_completion: bool = True):
     """Runs training until the specified global step count has been reached.
 

orbit/controller_test.py

@@ -281,6 +281,7 @@ def test_no_checkpoint(self):
     test_controller.train_and_evaluate(
         train_steps=10, eval_steps=2, eval_interval=6)
     self.assertEqual(test_runner.global_step, 10)
+    self.assertTrue(controller._orbit_api_gauge.get_cell().value())
 
   def test_no_checkpoint_and_summaries(self):
     test_runner = TestRunner()
@@ -293,6 +294,7 @@ def test_no_checkpoint_and_summaries(self):
     test_controller.train_and_evaluate(
         train_steps=10, eval_steps=2, eval_interval=6)
     self.assertEqual(test_runner.global_step, 10)
+    self.assertTrue(controller._orbit_api_gauge.get_cell().value())
 
   @parameterized.named_parameters(
       ("_sync_checkpoint_saving", False),
@@ -317,6 +319,7 @@ def test_has_checkpoint_no_summaries(self, enable_async_checkpoint_saving):
     test_controller.train_and_evaluate(
         train_steps=10, eval_steps=2, eval_interval=6)
     self.assertEqual(test_runner.global_step, 10)
+    self.assertTrue(controller._orbit_api_gauge.get_cell().value())
 
     # No summaries are saved.
     self.assertEmpty(tf.io.gfile.glob(