From a64da507aee25b7953d4c4010744089b5fe90c18 Mon Sep 17 00:00:00 2001
From: John Chilton <jmchilton@gmail.com>
Date: Fri, 31 Oct 2025 18:17:33 -0400
Subject: [PATCH] Richer tracking of transient failures.

- Update error message to reflect the error is likely a known transient failure.
- Track explicit issue numbers with errors so we can more easily track information about the error over time.
---
 lib/galaxy/util/unittest_utils/__init__.py    | 36 +++++++++++++++++++
 lib/galaxy_test/api/test_jobs.py              |  2 ++
 .../selenium/test_history_sharing.py          |  2 ++
 .../selenium/test_tool_discovery_view.py      |  2 ++
 .../objectstore/test_swift_objectstore.py     |  2 ++
 5 files changed, 44 insertions(+)

diff --git a/lib/galaxy/util/unittest_utils/__init__.py b/lib/galaxy/util/unittest_utils/__init__.py
index 3618551cbc6..e3ccc8e46d8 100644
--- a/lib/galaxy/util/unittest_utils/__init__.py
+++ b/lib/galaxy/util/unittest_utils/__init__.py
@@ -57,3 +57,39 @@ def skip_unless_environ(env_var: str) -> Union[Callable[[Callable[P, T]], Callab
         return _identity
 
     return pytest.mark.skip(f"{env_var} must be set for this test")
+
+
+def transient_failure(issue: int) -> Callable[[Callable[P, T]], Callable[P, T]]:
+    """Mark test as known transient failure with GitHub issue tracking.
+
+    This decorator catches exceptions from tests and rewraps them with a marker
+    indicating this is a known transient failure. This allows automated tooling
+    to categorize failures and helps reviewers quickly identify flaky tests.
+
+    Args:
+        issue: GitHub issue number tracking this transient failure
+
+    Example:
+        @transient_failure(issue=12345)
+        def test_flaky_selenium(self):
+            # Test that sometimes fails due to race condition
+            ...
+    """
+
+    def decorator(func: Callable[P, T]) -> Callable[P, T]:
+        @wraps(func)
+        def wrapper(*args: P.args, **kwargs: P.kwargs) -> T:
+            try:
+                return func(*args, **kwargs)
+            except Exception as e:
+                msg = f"TRANSIENT FAILURE [Issue #{issue}]: {str(e)}"
+                # Try to preserve exception type, fallback to plain Exception
+                try:
+                    raise type(e)(msg) from e
+                except (TypeError, AttributeError):
+                    # type(e) doesn't accept single string arg
+                    raise Exception(msg) from e
+
+        return wrapper
+
+    return decorator
diff --git a/lib/galaxy_test/api/test_jobs.py b/lib/galaxy_test/api/test_jobs.py
index 52fb8a42f71..5639449b106 100644
--- a/lib/galaxy_test/api/test_jobs.py
+++ b/lib/galaxy_test/api/test_jobs.py
@@ -10,6 +10,7 @@ import pytest
 import requests
 from dateutil.parser import isoparse
 
+from galaxy.util.unittest_utils import transient_failure
 from galaxy_test.api.test_tools import TestsTools
 from galaxy_test.base.api_asserts import assert_status_code_is_ok
 from galaxy_test.base.populators import (
@@ -851,6 +852,7 @@ steps:
         search_payload = self._search_payload(history_id=history_id, tool_id="multi_data_param", inputs=inputs)
         self._search(search_payload, expected_search_count=0)
 
+    @transient_failure(issue=21230)
     @pytest.mark.require_new_history
     def test_search_delete_hdca_output(self, history_id):
         list_id_a = self.__history_with_ok_collection(collection_type="list", history_id=history_id)
diff --git a/lib/galaxy_test/selenium/test_history_sharing.py b/lib/galaxy_test/selenium/test_history_sharing.py
index 16371bcd68f..7c2673a38e3 100644
--- a/lib/galaxy_test/selenium/test_history_sharing.py
+++ b/lib/galaxy_test/selenium/test_history_sharing.py
@@ -1,3 +1,4 @@
+from galaxy.util.unittest_utils import transient_failure
 from .framework import (
     selenium_only,
     selenium_test,
@@ -157,6 +158,7 @@ class TestPrivateHistorySharingRequiresPermissionChanges(SeleniumTestCase):
     me and users this history is shared with" rather than "Make datasets public".
     """
 
+    @transient_failure(issue=21224)
     @selenium_test
     def test_sharing_private_history_default_permission(self):
         # Create two test users - one to own the history, one to share with
diff --git a/lib/galaxy_test/selenium/test_tool_discovery_view.py b/lib/galaxy_test/selenium/test_tool_discovery_view.py
index 43cecfd4d31..612eb91e973 100644
--- a/lib/galaxy_test/selenium/test_tool_discovery_view.py
+++ b/lib/galaxy_test/selenium/test_tool_discovery_view.py
@@ -1,3 +1,4 @@
+from galaxy.util.unittest_utils import transient_failure
 from .framework import (
     selenium_test,
     SeleniumTestCase,
@@ -15,6 +16,7 @@ class TestToolDiscoveryViewAnonymous(SeleniumTestCase):
     advanced search, and list vs grid view toggling.
     """
 
+    @transient_failure(issue=21225)
     @selenium_test
     def test_tool_discovery_landing(self):
         """Test navigation to the tool discovery view."""
diff --git a/test/integration/objectstore/test_swift_objectstore.py b/test/integration/objectstore/test_swift_objectstore.py
index 9802831f0a7..cf64b426f3c 100644
--- a/test/integration/objectstore/test_swift_objectstore.py
+++ b/test/integration/objectstore/test_swift_objectstore.py
@@ -18,6 +18,8 @@ TEST_TOOL_IDS = [
     "tool_provided_metadata_10",
     "tool_provided_metadata_11",
     "tool_provided_metadata_12",
+    # Transiently fails - see issue #21226 - I'd love to mark this with the decorator but
+    # the test framework here doesn't support it yet.
     "composite_output",
     "composite_output_tests",
     "metadata",