vellum-workflow-server 1.9.1.post2__tar.gz → 1.9.6.post2__tar.gz

{vellum_workflow_server-1.9.1.post2 → vellum_workflow_server-1.9.6.post2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: vellum-workflow-server
-Version: 1.9.1.post2
+Version: 1.9.6.post2
 Summary: 
 License: AGPL
 Requires-Python: >=3.9.0,<4
@@ -29,7 +29,7 @@ Requires-Dist: pyjwt (==2.10.0)
 Requires-Dist: python-dotenv (==1.0.1)
 Requires-Dist: retrying (==1.3.4)
 Requires-Dist: sentry-sdk[flask] (==2.20.0)
-Requires-Dist: vellum-ai (==1.9.1)
+Requires-Dist: vellum-ai (==1.9.6)
 Description-Content-Type: text/markdown
 
 # Vellum Workflow Runner Server

{vellum_workflow_server-1.9.1.post2 → vellum_workflow_server-1.9.6.post2}/pyproject.toml

@@ -3,7 +3,7 @@ name = "vellum-workflow-server"
 
 [tool.poetry]
 name = "vellum-workflow-server"
-version = "1.9.1.post2"
+version = "1.9.6.post2"
 description = ""
 readme = "README.md"
 authors = []
@@ -45,7 +45,7 @@ flask = "2.3.3"
 orderly-set = "5.2.2"
 pebble = "5.0.7"
 gunicorn = "23.0.0"
-vellum-ai = "1.9.1"
+vellum-ai = "1.9.6"
 python-dotenv = "1.0.1"
 retrying = "1.3.4"
 sentry-sdk = {extras = ["flask"], version = "2.20.0"}

{vellum_workflow_server-1.9.1.post2 → vellum_workflow_server-1.9.6.post2}/src/workflow_server/api/tests/test_workflow_view_stream_workflow_route.py

@@ -5,6 +5,7 @@ import io
 import json
 from queue import Empty
 import re
+import time
 from unittest import mock
 from uuid import uuid4
 
@@ -133,6 +134,8 @@ class Workflow(BaseWorkflow):
 
     with mock.patch("builtins.open", mock.mock_open(read_data="104857600")):
         # WHEN we call the stream route
+        ts_ns = time.time_ns()
+        request_body["vembda_service_initiated_timestamp"] = ts_ns
         status_code, events = both_stream_types(request_body)
 
     # THEN we get a 200 response
@@ -177,6 +180,15 @@ class Workflow(BaseWorkflow):
     assert "is_new_server" in server_metadata
     assert server_metadata["is_new_server"] is False
 
+    # AND the initiated event should have initiated_latency within a reasonable range
+    assert "initiated_latency" in server_metadata, "initiated_latency should be present in server_metadata"
+    initiated_latency = server_metadata["initiated_latency"]
+    assert isinstance(initiated_latency, int), "initiated_latency should be an integer (nanoseconds)"
+    # Latency should be positive and less than 60 seconds (60_000_000_000 nanoseconds) for CI
+    assert (
+        0 < initiated_latency < 60_000_000_000
+    ), f"initiated_latency should be between 0 and 60 seconds, got {initiated_latency} ns"
+
     assert events[2]["name"] == "workflow.execution.fulfilled", events[2]
     assert events[2]["body"]["workflow_definition"]["module"] == ["test", "workflow"]
 

{vellum_workflow_server-1.9.1.post2 → vellum_workflow_server-1.9.6.post2}/src/workflow_server/core/executor.py

@@ -1,4 +1,4 @@
-from datetime import datetime
+from datetime import datetime, timezone
 from io import StringIO
 import json
 import logging
@@ -191,6 +191,7 @@ def stream_workflow(
             previous_execution_id=executor_context.previous_execution_id,
             timeout=executor_context.timeout,
             trigger=trigger,
+            execution_id=executor_context.workflow_span_id,
         )
     except WorkflowInitializationException as e:
         cancel_watcher_kill_switch.set()
@@ -449,6 +450,14 @@ def _enrich_event(event: WorkflowEvent, executor_context: Optional[BaseExecutorC
 
             if executor_context is not None:
                 metadata["is_new_server"] = executor_context.is_new_server
+
+                if executor_context.vembda_service_initiated_timestamp is not None and event.timestamp is not None:
+                    event_ts = event.timestamp
+                    if event_ts.tzinfo is None:
+                        event_ts = event_ts.replace(tzinfo=timezone.utc)
+                    event_ts_ns = int(event_ts.timestamp() * 1_000_000_000)
+                    initiated_latency = event_ts_ns - executor_context.vembda_service_initiated_timestamp
+                    metadata["initiated_latency"] = initiated_latency
         elif event.name == "workflow.execution.fulfilled" and is_deployment:
             metadata = {}
             memory_mb = get_memory_in_use_mb()

{vellum_workflow_server-1.9.1.post2 → vellum_workflow_server-1.9.6.post2}/src/workflow_server/core/workflow_executor_context.py

@@ -40,6 +40,7 @@ class BaseExecutorContext(UniversalBaseModel):
     # The actual 'execution id' of the workflow that we pass into the workflow
     # when running in async mode.
     workflow_span_id: Optional[UUID] = None
+    vembda_service_initiated_timestamp: Optional[int] = None
 
     @field_validator("inputs", mode="before")
     @classmethod

vellum_workflow_server-1.9.6.post2/src/workflow_server/utils/exit_handler.py

@@ -0,0 +1,56 @@
+from datetime import datetime
+import logging
+import multiprocessing
+import signal
+from time import sleep
+from typing import Any
+
+from workflow_server.config import IS_ASYNC_MODE, is_development
+from workflow_server.utils.system_utils import get_active_process_count
+
+logger = logging.getLogger(__name__)
+process_killed_switch = multiprocessing.Event()
+
+
+def _wait_for_workers() -> None:
+    # Would be annoying to have this on for dev since would prevent reload restarts. Also disabling this
+    # for non async mode for now since it shouldn't be needed anyway cus we keep the requests open.
+    if is_development() and not IS_ASYNC_MODE:
+        return
+
+    start_time = datetime.now()
+    loops = 0
+
+    while get_active_process_count() > 0:
+        if loops % 30 == 0:
+            logger.info("Waiting for workflow processes to finish...")
+
+        # TODO needa pass in max workflow time here for VPC
+        if (datetime.now() - start_time).total_seconds() > 1800:
+            logger.warning("Max elapsed time waiting for workflow processes to complete exceeded, shutting down")
+            exit(1)
+
+        sleep(1)
+        loops += 1
+
+
+def gunicorn_exit_handler(_worker: Any) -> None:
+    logger.info("Received gunicorn kill signal")
+    process_killed_switch.set()
+    _wait_for_workers()
+
+
+def exit_handler(_signal: int, _frame: Any) -> None:
+    """
+    Gunicorn overrides this signal handler but theres periods where the gunicorn server
+    hasn't initialized or for local dev where this will get called.
+    """
+    process_killed_switch.set()
+    logger.warning("Received kill signal")
+    _wait_for_workers()
+    exit(1)
+
+
+def init_signal_handlers() -> None:
+    signal.signal(signal.SIGTERM, exit_handler)
+    signal.signal(signal.SIGINT, exit_handler)

vellum_workflow_server-1.9.1.post2/src/workflow_server/utils/exit_handler.py

@@ -1,27 +0,0 @@
-import logging
-import multiprocessing
-import signal
-from typing import Any
-
-logger = logging.getLogger(__name__)
-process_killed_switch = multiprocessing.Event()
-
-
-def gunicorn_exit_handler(_worker: Any) -> None:
-    process_killed_switch.set()
-    logger.warning("Received gunicorn kill signal")
-
-
-def exit_handler(_signal: int, _frame: Any) -> None:
-    """
-    Gunicorn overrides this signal handler but theres periods where the gunicorn server
-    hasn't initialized or for local dev where this will get called.
-    """
-    process_killed_switch.set()
-    logger.warning("Received kill signal")
-    exit(1)
-
-
-def init_signal_handlers() -> None:
-    signal.signal(signal.SIGTERM, exit_handler)
-    signal.signal(signal.SIGINT, exit_handler)