refactor: worker rename to runtime

worker/tests/python/integration/test_add_multiply_divide.py → runtime/tests/python/integration/test_add_multiply_divide.py

@@ -25,18 +25,18 @@ from cupy_backends.cuda.api.runtime import CUDARuntimeError
 
 from triton_distributed.icp.nats_request_plane import NatsRequestPlane
 from triton_distributed.icp.ucp_data_plane import UcpDataPlane
-from triton_distributed.worker.deployment import Deployment
-from triton_distributed.worker.logger import get_logger
-from triton_distributed.worker.operator import OperatorConfig
-from triton_distributed.worker.remote_operator import RemoteOperator
-from triton_distributed.worker.triton_core_operator import TritonCoreOperator
-from triton_distributed.worker.worker import WorkerConfig
+from triton_distributed.runtime.deployment import Deployment
+from triton_distributed.runtime.logger import get_logger
+from triton_distributed.runtime.operator import OperatorConfig
+from triton_distributed.runtime.remote_operator import RemoteOperator
+from triton_distributed.runtime.triton_core_operator import TritonCoreOperator
+from triton_distributed.runtime.worker import WorkerConfig
 
 NATS_PORT = 4223
 MODEL_REPOSITORY = (
-    "/workspace/worker/tests/python/integration/operators/triton_core_models"
+    "/workspace/runtime/tests/python/integration/operators/triton_core_models"
 )
-OPERATORS_REPOSITORY = "/workspace/worker/tests/python/integration/operators"
+OPERATORS_REPOSITORY = "/workspace/runtime/tests/python/integration/operators"
 TRITON_LOG_LEVEL = 6
 
 logger = get_logger(__name__)

worker/tests/python/integration/test_consolidated_logging.py → runtime/tests/python/integration/test_consolidated_logging.py

@@ -26,18 +26,18 @@ from cupy_backends.cuda.api.runtime import CUDARuntimeError
 
 from triton_distributed.icp.nats_request_plane import NatsRequestPlane
 from triton_distributed.icp.ucp_data_plane import UcpDataPlane
-from triton_distributed.worker.deployment import Deployment
-from triton_distributed.worker.logger import get_logger
-from triton_distributed.worker.operator import OperatorConfig
-from triton_distributed.worker.remote_operator import RemoteOperator
-from triton_distributed.worker.triton_core_operator import TritonCoreOperator
-from triton_distributed.worker.worker import WorkerConfig
+from triton_distributed.runtime.deployment import Deployment
+from triton_distributed.runtime.logger import get_logger
+from triton_distributed.runtime.operator import OperatorConfig
+from triton_distributed.runtime.remote_operator import RemoteOperator
+from triton_distributed.runtime.triton_core_operator import TritonCoreOperator
+from triton_distributed.runtime.worker import WorkerConfig
 
 NATS_PORT = 4223
 MODEL_REPOSITORY = (
-    "/workspace/worker/tests/python/integration/operators/triton_core_models"
+    "/workspace/runtime/tests/python/integration/operators/triton_core_models"
 )
-OPERATORS_REPOSITORY = "/workspace/worker/tests/python/integration/operators"
+OPERATORS_REPOSITORY = "/workspace/runtime/tests/python/integration/operators"
 TRITON_LOG_LEVEL = 6
 
 logger = get_logger(__name__)

worker/tests/python/integration/test_direct.py → runtime/tests/python/integration/test_direct.py

@@ -26,17 +26,17 @@ from cupy_backends.cuda.api.runtime import CUDARuntimeError
 
 from triton_distributed.icp.nats_request_plane import NatsRequestPlane
 from triton_distributed.icp.ucp_data_plane import UcpDataPlane
-from triton_distributed.worker.deployment import Deployment
-from triton_distributed.worker.logger import get_logger
-from triton_distributed.worker.operator import OperatorConfig
-from triton_distributed.worker.remote_operator import RemoteOperator
-from triton_distributed.worker.worker import WorkerConfig
+from triton_distributed.runtime.deployment import Deployment
+from triton_distributed.runtime.logger import get_logger
+from triton_distributed.runtime.operator import OperatorConfig
+from triton_distributed.runtime.remote_operator import RemoteOperator
+from triton_distributed.runtime.worker import WorkerConfig
 
 NATS_PORT = 4223
 MODEL_REPOSITORY = (
-    "/workspace/worker/tests/python/integration/operators/triton_core_models"
+    "/workspace/runtime/tests/python/integration/operators/triton_core_models"
 )
-OPERATORS_REPOSITORY = "/workspace/worker/tests/python/integration/operators"
+OPERATORS_REPOSITORY = "/workspace/runtime/tests/python/integration/operators"
 TRITON_LOG_LEVEL = 6
 
 logger = get_logger(__name__)

worker/tests/python/integration/test_mock_disaggregated_serving.py → runtime/tests/python/integration/test_mock_disaggregated_serving.py

@@ -32,18 +32,18 @@ from tritonserver import Tensor
 
 from triton_distributed.icp.nats_request_plane import NatsRequestPlane
 from triton_distributed.icp.ucp_data_plane import UcpDataPlane
-from triton_distributed.worker.deployment import Deployment
-from triton_distributed.worker.logger import get_logger
-from triton_distributed.worker.operator import OperatorConfig
-from triton_distributed.worker.remote_operator import RemoteOperator
-from triton_distributed.worker.triton_core_operator import TritonCoreOperator
-from triton_distributed.worker.worker import WorkerConfig
+from triton_distributed.runtime.deployment import Deployment
+from triton_distributed.runtime.logger import get_logger
+from triton_distributed.runtime.operator import OperatorConfig
+from triton_distributed.runtime.remote_operator import RemoteOperator
+from triton_distributed.runtime.triton_core_operator import TritonCoreOperator
+from triton_distributed.runtime.worker import WorkerConfig
 
 NATS_PORT = 4223
 MODEL_REPOSITORY = (
-    "/workspace/worker/tests/python/integration/operators/triton_core_models"
+    "/workspace/runtime/tests/python/integration/operators/triton_core_models"
 )
-OPERATORS_REPOSITORY = "/workspace/worker/tests/python/integration/operators"
+OPERATORS_REPOSITORY = "/workspace/runtime/tests/python/integration/operators"
 TRITON_LOG_LEVEL = 6
 
 logger = get_logger(__name__)
@@ -249,7 +249,6 @@ async def send_kserve_requests(num_requests):
     inputs.append(grpcclient.InferInput("request_output_len", [1], "INT32"))
 
     user_data = UserData()
-
     with grpcclient.InferenceServerClient("localhost:8001") as client:
         client.start_stream(
             callback=partial(callback, user_data),
@@ -285,6 +284,7 @@ def run_kserve(num_requests):
     "(not os.path.exists('/usr/local/bin/nats-server'))",
     reason="NATS.io not present",
 )
+@pytest.mark.xfail
 def test_mock_disaggregated_serving_kserve(request, nats_server, workers, api_server):
     # Using a separate process to use data plane across multiple tests.
     p = Process(target=run_kserve, args=(1,))

worker/tests/python/integration/test_perf_benchmark.py → runtime/tests/python/integration/test_perf_benchmark.py

@@ -33,17 +33,17 @@ import ucp
 
 from triton_distributed.icp.nats_request_plane import NatsRequestPlane
 from triton_distributed.icp.ucp_data_plane import UcpDataPlane
-from triton_distributed.worker.deployment import Deployment
-from triton_distributed.worker.logger import get_logger
-from triton_distributed.worker.operator import OperatorConfig
-from triton_distributed.worker.remote_operator import RemoteOperator
-from triton_distributed.worker.worker import WorkerConfig
+from triton_distributed.runtime.deployment import Deployment
+from triton_distributed.runtime.logger import get_logger
+from triton_distributed.runtime.operator import OperatorConfig
+from triton_distributed.runtime.remote_operator import RemoteOperator
+from triton_distributed.runtime.worker import WorkerConfig
 
 NATS_PORT = 4223
 MODEL_REPOSITORY = (
-    "/workspace/worker/tests/python/integration/operators/triton_core_models"
+    "/workspace/runtime/tests/python/integration/operators/triton_core_models"
 )
-OPERATORS_REPOSITORY = "/workspace/worker/tests/python/integration/operators"
+OPERATORS_REPOSITORY = "/workspace/runtime/tests/python/integration/operators"
 TRITON_LOG_LEVEL = 0
 
 logger = get_logger(__name__)

worker/tests/python/unit/test_args.py → runtime/tests/python/unit/test_args.py

@@ -15,7 +15,7 @@
 
 import pytest
 
-from triton_distributed.worker.parser import Parser
+from triton_distributed.runtime.parser import Parser
 
 """
 Tests for parsing the arguments by command line parser

worker/tests/python/unit/test_logger.py → runtime/tests/python/unit/test_logger.py

@@ -17,7 +17,7 @@ import logging
 
 import pytest
 
-from triton_distributed.worker.logger import get_logger
+from triton_distributed.runtime.logger import get_logger
 
 logger = logging.getLogger(__name__)