diff --git a/datadog_lambda/metric.py b/datadog_lambda/metric.py
index 115686af..3bc9955f 100644
--- a/datadog_lambda/metric.py
+++ b/datadog_lambda/metric.py
@@ -7,6 +7,7 @@
 import time
 import logging
 import ujson as json
+from datetime import datetime, timedelta
 
 from datadog_lambda.extension import should_use_extension
 from datadog_lambda.tags import get_enhanced_metrics_tags, dd_lambda_layer_tag
@@ -61,6 +62,16 @@ def lambda_metric(metric_name, value, timestamp=None, tags=None, force_async=Fal
     if should_use_extension and timestamp is not None:
         # The extension does not support timestamps for distributions so we create a
         # a thread stats writer to submit metrics with timestamps to the API
+        timestamp_ceiling = int(
+            (datetime.now() - timedelta(hours=4)).timestamp()
+        )  # 4 hours ago
+        if timestamp_ceiling > timestamp:
+            logger.warning(
+                "Timestamp %s is older than 4 hours, not submitting metric %s",
+                timestamp,
+                metric_name,
+            )
+            return
         global extension_thread_stats
         if extension_thread_stats is None:
             from datadog_lambda.thread_stats_writer import ThreadStatsWriter
@@ -108,11 +119,19 @@ def write_metric_point_to_stdout(metric_name, value, timestamp=None, tags=[]):
     )
 
 
-def flush_stats():
+def flush_stats(lambda_context=None):
     lambda_stats.flush()
 
     if extension_thread_stats is not None:
-        extension_thread_stats.flush()
+        if lambda_context is not None:
+            tags = get_enhanced_metrics_tags(lambda_context)
+            split_arn = lambda_context.invoked_function_arn.split(":")
+            if len(split_arn) > 7:
+                # Get rid of the alias
+                split_arn.pop()
+            arn = ":".join(split_arn)
+            tags.append("function_arn:" + arn)
+        extension_thread_stats.flush(tags)
 
 
 def submit_enhanced_metric(metric_name, lambda_context):
diff --git a/datadog_lambda/thread_stats_writer.py b/datadog_lambda/thread_stats_writer.py
index bfcf3c99..367b8b21 100644
--- a/datadog_lambda/thread_stats_writer.py
+++ b/datadog_lambda/thread_stats_writer.py
@@ -22,11 +22,13 @@ def distribution(self, metric_name, value, tags=[], timestamp=None):
             metric_name, value, tags=tags, timestamp=timestamp
         )
 
-    def flush(self):
+    def flush(self, tags=None):
         """ "Flush distributions from ThreadStats to Datadog.
         Modified based on `datadog.threadstats.base.ThreadStats.flush()`,
         to gain better control over exception handling.
         """
+        if tags:
+            self.thread_stats.constant_tags = self.thread_stats.constant_tags + tags
         _, dists = self.thread_stats._get_aggregate_metrics_and_dists(float("inf"))
         count_dists = len(dists)
         if not count_dists:
diff --git a/datadog_lambda/wrapper.py b/datadog_lambda/wrapper.py
index ba31f2be..9025bfef 100644
--- a/datadog_lambda/wrapper.py
+++ b/datadog_lambda/wrapper.py
@@ -366,7 +366,7 @@ def _after(self, event, context):
                     logger.debug("Failed to create cold start spans. %s", e)
 
             if not self.flush_to_log or should_use_extension:
-                flush_stats()
+                flush_stats(context)
             if should_use_extension and self.local_testing_mode:
                 # when testing locally, the extension does not know when an
                 # invocation completes because it does not have access to the
diff --git a/tests/test_metric.py b/tests/test_metric.py
index f07a4c6a..031b1180 100644
--- a/tests/test_metric.py
+++ b/tests/test_metric.py
@@ -5,7 +5,7 @@
 
 from botocore.exceptions import ClientError as BotocoreClientError
 from datadog.api.exceptions import ClientError
-
+from datetime import datetime, timedelta
 
 from datadog_lambda.metric import lambda_metric
 from datadog_lambda.api import decrypt_kms_api_key, KMS_ENCRYPTION_CONTEXT_KEY
@@ -49,12 +49,28 @@ def test_lambda_metric_timestamp_with_extension(self):
         self.mock_metric_extension_thread_stats = patcher.start()
         self.addCleanup(patcher.stop)
 
-        lambda_metric("test_timestamp", 1, 123)
+        delta = timedelta(minutes=1)
+        timestamp = int((datetime.now() - delta).timestamp())
+
+        lambda_metric("test_timestamp", 1, timestamp)
         self.mock_metric_lambda_stats.distribution.assert_not_called()
         self.mock_metric_extension_thread_stats.distribution.assert_called_with(
-            "test_timestamp", 1, timestamp=123, tags=[dd_lambda_layer_tag]
+            "test_timestamp", 1, timestamp=timestamp, tags=[dd_lambda_layer_tag]
         )
 
+    @patch("datadog_lambda.metric.should_use_extension", True)
+    def test_lambda_metric_invalid_timestamp_with_extension(self):
+        patcher = patch("datadog_lambda.metric.extension_thread_stats")
+        self.mock_metric_extension_thread_stats = patcher.start()
+        self.addCleanup(patcher.stop)
+
+        delta = timedelta(hours=5)
+        timestamp = int((datetime.now() - delta).timestamp())
+
+        lambda_metric("test_timestamp", 1, timestamp)
+        self.mock_metric_lambda_stats.distribution.assert_not_called()
+        self.mock_metric_extension_thread_stats.distribution.assert_not_called()
+
     def test_lambda_metric_flush_to_log(self):
         os.environ["DD_FLUSH_TO_LOG"] = "True"
 
@@ -84,6 +100,16 @@ def test_retry_on_remote_disconnected(self):
         lambda_stats.flush()
         self.assertEqual(self.mock_threadstats_flush_distributions.call_count, 2)
 
+    def test_flush_stats_with_tags(self):
+        lambda_stats = ThreadStatsWriter(True)
+        tags = ["tag1:value1", "tag2:value2"]
+        lambda_stats.flush(tags)
+        self.mock_threadstats_flush_distributions.assert_called_once_with(
+            lambda_stats.thread_stats._get_aggregate_metrics_and_dists(float("inf"))[1]
+        )
+        for tag in tags:
+            self.assertTrue(tag in lambda_stats.thread_stats.constant_tags)
+
 
 MOCK_FUNCTION_NAME = "myFunction"