SG-43825 Improve pipeline stability

azure-pipelines-templates/run-tests.yml

@@ -102,6 +102,8 @@ jobs:
           --durations=0 \
           --nunit-xml=test-results.xml \
           --verbose \
+          --reruns 1 \
+          --reruns-delay 2 \
     env:
       # Tell Pytest that we're running in a CI environment
       CI: 1
@@ -123,20 +125,39 @@ jobs:
       # This will give a user name like 'something macOS 2.7'
       SG_HUMAN_NAME: $(python_api_human_name) ${{ parameters.os_name }} ${{ parameters.python_version }}
       SG_HUMAN_PASSWORD: $(python_api_human_password)
-      # So, first, we need to make sure that two builds running at the same time do not manipulate
-      # the same entities, so we're sandboxing build nodes based on their name.
-      SG_PROJECT_NAME: Python API CI - $(Agent.Name)
-      # The entities created and then reused between tests assume that the same user is always
-      # manipulating them. Because different builds will be assigned different agents and therefore
-      # different projects, it means each project needs to have an entity specific to a given user.
-      # Again, this would have been a lot simpler if we could simply have had a login based on the
-      # agent name, but alas, the agent name has a space in it which needs to be replaced to something
-      # else and string substitution can't be made on build variables, only template parameters.
-      SG_ASSET_CODE: CI-$(python_api_human_login)-${{ parameters.os_name }}-${{ parameters.python_version }}
-      SG_VERSION_CODE: CI-$(python_api_human_login)-${{ parameters.os_name }}-${{ parameters.python_version }}
-      SG_SHOT_CODE: CI-$(python_api_human_login)-${{ parameters.os_name }}-${{ parameters.python_version }}
-      SG_TASK_CONTENT: CI-$(python_api_human_login)-${{ parameters.os_name }}-${{ parameters.python_version }}
-      SG_PLAYLIST_CODE: CI-$(python_api_human_login)-${{ parameters.os_name }}-${{ parameters.python_version }}
+      # Each job gets its own ephemeral project, scoped to this build + OS + Python version.
+      # This eliminates state bleed between concurrent runs and across successive builds on the
+      # same agent. The project is retired in the "Cleanup test project" step below.
+      SG_PROJECT_NAME: Python API CI - $(Build.BuildId) - ${{ parameters.os_name }} - ${{ parameters.python_version }}
+
+  - task: PythonScript@0
+    displayName: Cleanup test project
+    condition: always()
+    inputs:
+      scriptSource: inline
+      workingDirectory: $(Build.SourcesDirectory)
+      script: |
+        import os, shotgun_api3
+        sg = shotgun_api3.Shotgun(
+            base_url=os.environ['SG_SERVER_URL'],
+            script_name=os.environ['SG_SCRIPT_NAME'],
+            api_key=os.environ['SG_API_KEY'],
+        )
+        project = sg.find_one(
+            entity_type='Project',
+            filters=[['name', 'is', os.environ['SG_PROJECT_NAME']]],
+        )
+        if not project:
+            print('Project not found, nothing to clean up.')
+            sys.exit(0)
+        sg.delete(entity_type='Project', entity_id=project['id'])
+        print('Retired project:', os.environ['SG_PROJECT_NAME'])
+    env:
+      PYTHONPATH: $(Build.SourcesDirectory)
+      SG_SERVER_URL: $(ci_site)
+      SG_SCRIPT_NAME: $(ci_site_script_name)
+      SG_API_KEY: $(ci_site_script_key)
+      SG_PROJECT_NAME: Python API CI - $(Build.BuildId) - ${{ parameters.os_name }} - ${{ parameters.python_version }}
 
   # Explicit call to PublishTestResults@2 and PublishCodeCoverageResults@2 here
   # instead of relying on pytest-azurepipelines because pytest-azurepipelines

tests/base.py

@@ -289,7 +289,7 @@ def _setup_db(cls, config, sg):
         cls.human_user = _find_or_create_entity(sg, "HumanUser", data)
 
         data = {"code": cls.config.asset_code, "project": cls.project}
-        keys = ["code"]
+        keys = ["code", "project"]
         cls.asset = _find_or_create_entity(sg, "Asset", data, keys)
 
         data = {

tests/requirements.txt

@@ -11,3 +11,4 @@
 pytest
 pytest-cov
 pytest-nunit
+pytest-rerunfailures

tests/test_api.py

@@ -3322,12 +3322,14 @@ def test_modify_visibility(self):
         if field_name not in schema:
             self.sg.schema_field_create("Asset", "text", "Project Visibility Test")
 
-        # Grab any two projects that we can use for toggling the visible property with.
-        projects = self.sg.find(
-            "Project", [], order=[{"field_name": "id", "direction": "asc"}]
+        # Hide/show on the ephemeral CI project so concurrent matrix jobs do not race.
+        # Use the oldest site project as the control (find_one defaults to sorting by "id" ascending)
+        project_1 = self.project
+        project_2 = self.sg.find_one("Project", [["id", "is_not", project_1["id"]]])
+        self.assertIsNotNone(
+            project_2,
+            "A second project is required to test per-project field visibility.",
         )
-        project_1 = projects[0]
-        project_2 = projects[1]
 
         # First, reset the field visibility in a known state, i.e. visible for both projects,
         # in case the last test run failed midway through.