ci: run Python SDK CI on CodeBuild runners

.github/workflows/ci.yml

@@ -16,7 +16,10 @@ permissions:
 jobs:
   lint:
     name: Lint and Format
-    runs-on: ubuntu-latest
+    # Distinct job labels keep each ephemeral runner assigned to its intended job.
+    runs-on:
+      - codebuild-agentcore-sdk-python-ci-${{ github.run_id }}-${{ github.run_attempt }}
+      - ci-lint
     steps:
       - uses: actions/checkout@v6
 
@@ -43,7 +46,11 @@ jobs:
 
   test:
     name: Test Python ${{ matrix.python-version }}
-    runs-on: ubuntu-latest
+    env:
+      UV_PYTHON: ${{ matrix.python-version }}
+    runs-on:
+      - codebuild-agentcore-sdk-python-ci-${{ github.run_id }}-${{ github.run_attempt }}
+      - ci-test-${{ matrix.python-version }}
     strategy:
       matrix:
         python-version: ['3.10', '3.11', '3.12', '3.13']
@@ -83,7 +90,9 @@ jobs:
 
   build:
     name: Build Distribution
-    runs-on: ubuntu-latest
+    runs-on:
+      - codebuild-agentcore-sdk-python-ci-${{ github.run_id }}-${{ github.run_attempt }}
+      - ci-build
     needs: [lint, test]
     steps:
       - uses: actions/checkout@v6
@@ -118,7 +127,9 @@ jobs:
 
   test-install:
     name: Test Package Installation
-    runs-on: ubuntu-latest
+    runs-on:
+      - codebuild-agentcore-sdk-python-ci-${{ github.run_id }}-${{ github.run_attempt }}
+      - ci-install-${{ matrix.python-version }}
     needs: build
     strategy:
       matrix:

tests/bedrock_agentcore/evaluation/runner/batch/test_batch_evaluation_runner.py

@@ -6,6 +6,7 @@
 import pytest
 from pydantic import ValidationError
 
+from bedrock_agentcore._utils.endpoints import DEFAULT_REGION
 from bedrock_agentcore.evaluation.runner.batch.batch_evaluation_models import (
     BatchEvaluationRunConfig,
     BatchEvaluatorConfig,
@@ -39,7 +40,7 @@ class TestRegionName:
             ({"region": "us-west-2", "region_name": "eu-west-1"}, "us-east-1", "us-west-2"),
             ({"region": "", "region_name": "eu-west-1"}, "us-east-1", "eu-west-1"),
             ({}, "us-east-1", "us-east-1"),
-            ({}, None, "us-west-2"),
+            ({}, None, DEFAULT_REGION),
         ],
     )
     def test_region_precedence(self, kwargs, session_region, expected):

tests/bedrock_agentcore/memory/integrations/strands/memorystore/test_sender.py

@@ -86,7 +86,8 @@ async def test_chunks_batch_at_max_turns() -> None:
     client.create_event.return_value = {}
     await make_sender(client, max_turns_per_event=2).send_batch([user_message(value) for value in "abcde"])
     assert client.create_event.call_count == 3
-    assert [[turn["text"] for turn in turns(call)] for call in client.create_event.call_args_list] == [
+    # Events are sent concurrently, but turns within each event retain their order.
+    assert sorted([turn["text"] for turn in turns(call)] for call in client.create_event.call_args_list) == [
         ["a", "b"],
         ["c", "d"],
         ["e"],
@@ -133,13 +134,14 @@ async def test_sequence_range_token_is_stable_and_chunk_specific() -> None:
     sender = make_sender(client, max_turns_per_event=2)
     batch = [user_message("a"), user_message("b"), user_message("c")]
     await sender.send_batch(batch, [1, 2, 3])
-    assert [call.kwargs["clientToken"] for call in client.create_event.call_args_list] == [
+    expected_tokens = [
         "mem-1-actor-1-run-1-1-2",
         "mem-1-actor-1-run-1-3-3",
     ]
+    assert sorted(call.kwargs["clientToken"] for call in client.create_event.call_args_list) == expected_tokens
     client.reset_mock()
     await sender.send_batch(batch, [1, 2, 3])
-    assert client.create_event.call_args_list[0].kwargs["clientToken"] == "mem-1-actor-1-run-1-1-2"
+    assert sorted(call.kwargs["clientToken"] for call in client.create_event.call_args_list) == expected_tokens
 
 
 async def test_explicit_empty_run_id_is_preserved() -> None:
@@ -179,13 +181,14 @@ def provider(message: Message) -> dict[str, MetadataValue]:
         [user_message(value) for value in ["A", "A", "B", "C", "B"]]
     )
     assert client.create_event.call_count == 4
-    assert [[turn["text"] for turn in turns(call)] for call in client.create_event.call_args_list] == [
+    assert sorted([turn["text"] for turn in turns(call)] for call in client.create_event.call_args_list) == [
         ["A", "A"],
         ["B"],
-        ["C"],
         ["B"],
+        ["C"],
     ]
-    assert client.create_event.call_args_list[0].kwargs["metadata"] == {"topic": {"stringValue": "A"}}
+    for call in client.create_event.call_args_list:
+        assert call.kwargs["metadata"] == {"topic": {"stringValue": turns(call)[0]["text"]}}
 
 
 async def test_constant_metadata_is_mapped_and_empty_bag_omitted() -> None:

tests/bedrock_agentcore/memory/integrations/strands/memorystore/test_static_typing.py

@@ -8,8 +8,19 @@
 
 
 def _run_mypy(fixture: str) -> subprocess.CompletedProcess[str]:
+    # Installed dependency stubs target this interpreter; Python 3.10 is
+    # checked by its own CI matrix entry.
     return subprocess.run(
-        [sys.executable, "-m", "mypy", "--no-incremental", "--show-error-codes", str(FIXTURES / fixture)],
+        [
+            sys.executable,
+            "-m",
+            "mypy",
+            "--python-version",
+            f"{sys.version_info.major}.{sys.version_info.minor}",
+            "--no-incremental",
+            "--show-error-codes",
+            str(FIXTURES / fixture),
+        ],
         check=False,
         capture_output=True,
         text=True,

tests/bedrock_agentcore/memory/integrations/strands/memorystore/test_types.py

@@ -28,7 +28,9 @@ def test_connection_config_runtime_typed_dict_keys_and_hints() -> None:
             "client",
         }
     )
-    assert get_args(get_type_hints(_AgentCoreMemoryConnectionConfig)["boto3_session"]) == (boto3.Session,)
+    assert get_args(get_type_hints(_AgentCoreMemoryConnectionConfig, include_extras=True)["boto3_session"]) == (
+        boto3.Session,
+    )
 
 
 def test_public_factory_typed_dict_keys_are_correct_at_runtime() -> None:
@@ -39,7 +41,7 @@ def test_public_factory_typed_dict_keys_are_correct_at_runtime() -> None:
     assert CreateAgentCoreMemoryStoresInput.__optional_keys__ == frozenset(
         {"extraction", "metadata_provider", "max_turns_per_event", "region_name", "boto3_session", "client"}
     )
-    hints = get_type_hints(CreateAgentCoreMemoryStoresInput)
+    hints = get_type_hints(CreateAgentCoreMemoryStoresInput, include_extras=True)
     assert get_args(hints["boto3_session"]) == (boto3.Session,)
     assert "extraction_mode" not in hints
 

tests/bedrock_agentcore/runtime/test_app.py

@@ -231,8 +231,10 @@ def test_serve_with_dockerenv_file(self, mock_uvicorn, mock_exists):
             bedrock_agentcore, host="0.0.0.0", port=8080, access_log=False, log_level="warning"
         )
 
+    @patch.dict(os.environ, {"DOCKER_CONTAINER": ""})
+    @patch("os.path.exists", return_value=False)
     @patch("uvicorn.run")
-    def test_serve_localhost(self, mock_uvicorn):
+    def test_serve_localhost(self, mock_uvicorn, mock_exists):
         """Test serve method uses localhost when not in Docker."""
         bedrock_agentcore = BedrockAgentCoreApp()
         bedrock_agentcore.run(port=8080)