Go back to randomized index names

Author: jhamon

.github/workflows/on-pr.yaml

@@ -48,7 +48,7 @@ jobs:
       - create-project
     with:
       encrypted_project_api_key: ${{ needs.create-project.outputs.encrypted_project_api_key }}
-      python_versions_json: '["3.9"]'
+      python_versions_json: '["3.13", "3.9"]'
 
   cleanup-project:
     if: ${{ always() }}

tests/integration/helpers/helpers.py

@@ -75,7 +75,7 @@ def poll_stats_for_namespace(
     idx: _Index,
     namespace: str,
     expected_count: int,
-    max_sleep: int = int(os.environ.get("FRESHNESS_TIMEOUT_SECONDS", 60)),
+    max_sleep: int = int(os.environ.get("FRESHNESS_TIMEOUT_SECONDS", 180)),
 ) -> None:
     delta_t = 5
     total_time = 0

tests/integration/helpers/names.py

@@ -1,4 +1,5 @@
-import hashlib
+import uuid
+# import hashlib
 
 
 def generate_name(test_name: str, label: str, max_length: int = 20) -> str:
@@ -11,4 +12,11 @@ def generate_name(test_name: str, label: str, max_length: int = 20) -> str:
     since the full length of 64 characters exceeds the allowed length of some fields
     in the API. For example, index names must be 45 characters or less.
     """
-    return hashlib.sha256(f"{test_name}-{label}".encode()).hexdigest()[:max_length]
+    # return hashlib.sha256(f"{test_name}-{label}".encode()).hexdigest()[:max_length]
+
+    # Having names be fully deterministic led to problems when multiple test builds
+    # are running in parallel, for example running the same tests in parallel for
+    # different python versions. We can solve this by incorporating more information
+    # into the name generation, but for now as a quick fix to unblock a release we
+    # will just fall back to using a random uuid.
+    return str(uuid.uuid4())