fix(test-studio): return partial result instead of raising when metrics not cached (#358)

CHANGELOG.md

@@ -32,6 +32,7 @@ SPDX-License-Identifier: MIT-0
 
 ### Fixed
 
+- **Test Studio results error for runs stuck in evaluation (#358)** — `getTestRun` (the `test_results_resolver` Lambda) raised an unhandled `ValueError` ("Test run … processing completed, evaluating results") when a run reached a terminal state but the evaluation-aggregation step never cached `testRunResult` — e.g. when aggregation is still running, timed out, or failed silently on a large run (the reporter hit this with 3 463 documents). The exception surfaced as an opaque error and the run spun on "Loading…" forever in Test Studio. The resolver now returns a structured partial `TestRun` (true status plus file counts and metadata, metric fields omitted) instead of raising, so the UI renders the in-progress/terminal state gracefully. This also stops a single not-yet-aggregated run from failing an entire `compareTestRuns` request. (The separate question of *why* aggregation can stall on very large runs is tracked as a follow-up.)
 - **Configuration version list silently truncated past the first page (#354)** — `ConfigurationManager.list_config_versions()` performed a single unpaginated `table.scan()` on the ConfigurationTable. Because a DynamoDB scan returns at most 1 MB per call, deployments with many config versions (e.g. 230+) only ever saw the ~58 that fit on the first page — uploaded-via-CLI and autotune-agent configs were invisible in the UI's View/Edit Configuration page and the upload-document config-version dropdown (the configs still worked when referenced by name). The method now paginates through `LastEvaluatedKey` so every version is returned. Fixes all callers (`update_configuration`, the AppSync `configuration_resolver`, `rules_discovery`, and the SDK).
 
 - **Build Info "update available" indicator broke against the public release bucket** — The `getLatestPublishedVersion` resolver discovered the newest published version by calling `ListObjectsV2` on the public artifacts bucket and parsing `idp-main_<version>.yaml` keys. That bucket grants `GetObject` only (no listing), so the check failed on real public deployments. `idp-cli publish` now writes a small pointer object — `<prefix>/idp-main-latest.json` (`{version, templateUrl}`) — at the version-stripped prefix on every release, and the resolver reads that one known key with a single `GetObject` (unsigned, falling back to signed), with a conventional `idp-main_<version>.yaml` URL fallback if the pointer omits one. No version parsing or `ListObjectsV2`. The check stays disabled when `PUBLIC_ARTIFACTS_BUCKET` is unset.

lib/idp_common_pkg/tests/unit/test_results_resolver.py

@@ -213,6 +213,51 @@ def test_build_config_comparison():
     assert "temperature" in [item["setting"] for item in config_diff]
 
 
+@pytest.mark.unit
+def test_get_test_results_missing_metrics_returns_partial_not_raises():
+    """When processing reached a terminal state but the evaluation aggregation
+    never cached testRunResult (timed out / failed silently on a large run),
+    get_test_results returns a structured partial TestRun instead of raising an
+    opaque ValueError that leaves the UI spinning on "Loading..." (issue #358)."""
+    test_run_id = "TEST-SET-ID"
+    metadata = {
+        "PK": f"testrun#{test_run_id}",
+        "SK": "metadata",
+        # Already terminal, so the status-refresh branch is skipped and we fall
+        # straight through to the "no cached metrics" else branch.
+        "Status": "COMPLETE",
+        "TestSetId": "set-1",
+        "TestSetName": "big-classification-set",
+        "FilesCount": 3463,
+        "CompletedFiles": 3460,
+        "FailedFiles": 3,
+        "CreatedAt": "2025-01-01T00:00:00Z",
+        "Context": "ctx",
+        "ConfigVersion": "v7",
+        # No "testRunResult" key -> aggregation hasn't written metrics yet.
+    }
+
+    mock_table = Mock()
+    mock_table.get_item.return_value = {"Item": metadata}
+
+    with (
+        patch.dict(os.environ, {"TRACKING_TABLE": "tracking"}),
+        patch.object(index.dynamodb, "Table", return_value=mock_table),
+    ):
+        result = index.get_test_results(test_run_id)
+
+    assert result["testRunId"] == test_run_id
+    # Reports the true terminal status rather than fabricating one.
+    assert result["status"] == "COMPLETE"
+    assert result["filesCount"] == 3463
+    assert result["completedFiles"] == 3460
+    assert result["failedFiles"] == 3
+    assert result["testSetId"] == "set-1"
+    assert result["configVersion"] == "v7"
+    # Metric fields are absent (not yet computed) but must not be required.
+    assert "overallAccuracy" not in result or result["overallAccuracy"] is None
+
+
 @pytest.mark.unit
 def test_handler_field_routing():
     """Test GraphQL field routing"""

nested/appsync/src/lambda/test_results_resolver/index.py

@@ -408,16 +408,36 @@ def get_test_results(test_run_id):
                 "config": _get_test_run_config(test_run_id),
             }
     else:
-        # Provide more specific message for ABORTED status
+        # No aggregate metrics have been cached yet. This happens when all
+        # files finished processing but the evaluation aggregation step hasn't
+        # written testRunResult (still running, or it timed out / failed on a
+        # large run). Don't raise — that surfaces as an opaque error and the UI
+        # spins on "Loading..." forever. Return a structured partial TestRun so
+        # the UI can render the in-progress status instead.
         if current_status == "ABORTED":
-            raise ValueError(
-                f"Test run {test_run_id} aborted, evaluating results for completed documents"
+            logger.info(
+                f"Test run {test_run_id} aborted; aggregate metrics not yet available"
             )
         else:
-            raise ValueError(
-                f"Test run {test_run_id} processing completed, evaluating results"
+            logger.info(
+                f"Test run {test_run_id} processing complete; "
+                "aggregate metrics not yet available (evaluation in progress)"
             )
 
+        return {
+            "testRunId": test_run_id,
+            "testSetId": metadata.get("TestSetId"),
+            "testSetName": metadata.get("TestSetName"),
+            "status": current_status,
+            "filesCount": metadata.get("FilesCount", 0),
+            "completedFiles": metadata.get("CompletedFiles", 0),
+            "failedFiles": metadata.get("FailedFiles", 0),
+            "createdAt": _format_datetime(metadata.get("CreatedAt")),
+            "completedAt": _format_datetime(metadata.get("CompletedAt")),
+            "context": metadata.get("Context"),
+            "configVersion": metadata.get("ConfigVersion"),
+        }
+
 
 def _query_test_runs_from_gsi(table, start_iso, end_iso):
     """Query test runs from TypeDateIndex GSI instead of scanning the full table.