feat: DIA-977: Get streaming job status

docker-compose.yml

@@ -48,6 +48,7 @@ services:
       - AWS_SESSION_TOKEN=${AWS_SESSION_TOKEN}
       - MODULE_NAME=process_file.app
       - KAFKA_BOOTSTRAP_SERVERS=kafka:9093
+      - C_FORCE_ROOT=true # needed when using pickle serializer in celery + running as root - remove when we dont run as root
     command:
       'sh -c "cd tasks && poetry run celery -A $$MODULE_NAME worker --loglevel=info"'
   redis:

pyproject.toml

@@ -35,6 +35,7 @@ fastapi = "^0.104.1"
 celery = {version = "^5.3.6", extras = ["redis"]}
 uvicorn = "*"
 pydantic-settings = "^2.2.1"
+label-studio-sdk = "^0.0.32"
 
 [tool.poetry.dev-dependencies]
 pytest = "^7.4.3"

server/app.py

@@ -23,6 +23,7 @@
     process_file,
     process_file_streaming,
     process_streaming_output,
+    streaming_parent_task,
 )
 from utils import get_input_topic, Settings
 from server.handlers.result_handlers import ResultHandler
@@ -207,24 +208,13 @@ async def submit_streaming(request: SubmitStreamingRequest):
         Response[JobCreated]: The response model for a job created.
     """
 
-    # TODO: get task by name, e.g. request.task_name
-    task = process_file_streaming
-    agent = request.agent
-
-    logger.info(f"Submitting task {task.name} with agent {agent}")
-    input_result = task.delay(agent=agent)
-    input_job_id = input_result.id
-    logger.debug(f"Task {task.name} submitted with job_id {input_job_id}")
-
-    task = process_streaming_output
-    logger.debug(f"Submitting task {task.name}")
-    output_result = task.delay(
-        job_id=input_job_id, result_handler=request.result_handler
+    task = streaming_parent_task
+    result = task.apply_async(
+        kwargs={"agent": request.agent, "result_handler": request.result_handler}
     )
-    output_job_id = output_result.id
-    logger.debug(f"Task {task.name} submitted with job_id {output_job_id}")
+    logger.info(f"Submitted {task.name} with ID {result.id}")
 
-    return Response[JobCreated](data=JobCreated(job_id=input_job_id))
+    return Response[JobCreated](data=JobCreated(job_id=result.id))
 
 
 @app.post("/jobs/submit-batch", response_model=Response)
@@ -256,6 +246,24 @@ async def submit_batch(batch: BatchData):
     return Response[BatchSubmitted](data=BatchSubmitted(job_id=batch.job_id))
 
 
+def aggregate_statuses(input_job_id: str, output_job_id: str):
+    input_job_status = process_file_streaming.AsyncResult(input_job_id).status
+    output_job_status = process_streaming_output.AsyncResult(output_job_id).status
+
+    statuses = [input_job_status, output_job_status]
+
+    if "PENDING" in statuses:
+        return "PENDING"
+    if "FAILURE" in statuses:
+        return "FAILURE"
+    if "REVOKED" in statuses:
+        return "REVOKED"
+    if "STARTED" in statuses or "RETRY" in statuses:
+        return "STARTED"
+
+    return "SUCCESS"
+
+
 @app.get("/jobs/{job_id}", response_model=Response[JobStatusResponse])
 def get_status(job_id):
     """
@@ -275,9 +283,23 @@ def get_status(job_id):
         "REVOKED": Status.CANCELED,
         "RETRY": Status.INPROGRESS,
     }
-    job = process_file.AsyncResult(job_id)
+    job = streaming_parent_task.AsyncResult(job_id)
+    logger.info(f"\n\nParent task meta : {job.info}\n\n")
+
+    # If parent task meta does not contain input/output job IDs - return FAILED
+    if job.info is None or "input_job_id" not in job.info or "output_job_id" not in job.info:
+        logger.error(
+            "Parent task does not contain input job ID and/or output_job_id - unable to return proper status"
+        )
+        return Response[JobStatusResponse](data=JobStatusResponse(status=Status.FAILED))
+
+    input_job_id = job.info["input_job_id"]
+    output_job_id = job.info["output_job_id"]
+
     try:
-        status: Status = celery_status_map[job.status]
+        status: Status = celery_status_map[
+            aggregate_statuses(input_job_id, output_job_id)
+        ]
     except Exception as e:
         logger.error(f"Error getting job status: {e}")
         status = Status.FAILED

server/tasks/process_file.py

@@ -3,6 +3,9 @@
 import pickle
 import os
 import logging
+import time
+
+from adala.agents import Agent
 
 from adala.agents import Agent
 
@@ -33,25 +36,92 @@ def process_file(agent: Agent):
     asyncio.run(agent.environment.finalize())
 
 
+@app.task(
+<<<<<<< HEAD
+    name="streaming_parent_task", track_started=True, bind=True, serializer="pickle"
+)
+def streaming_parent_task(
+    self, agent: Agent, result_handler: ResultHandler, batch_size: int = 2
+):
+    """
+    This task is used to launch the two tasks that are doing the real work, so that
+    we store those two job IDs as metadata of this parent task, and be able to get
+    the status of the entire job from one task ID
+    """
+
+    # Parent job ID is used for input/output topic names
+    parent_job_id = self.request.id
+
+    inference_task = process_file_streaming
+    logger.info(f"Submitting task {inference_task.name} with agent {agent}")
+    input_result = inference_task.delay(agent=agent, parent_job_id=parent_job_id)
+    input_job_id = input_result.id
+    logger.info(f"Task {inference_task.name} submitted with job_id {input_job_id}")
+
+    result_handler_task = process_streaming_output
+    logger.info(f"Submitting task {result_handler_task.name}")
+    output_result = result_handler_task.delay(
+        input_job_id=input_job_id,
+        parent_job_id=parent_job_id,
+        result_handler=result_handler,
+        batch_size=batch_size,
+    )
+    output_job_id = output_result.id
+    logger.info(
+        f"Task {result_handler_task.name} submitted with job_id {output_job_id}"
+    )
+
+    # Store input and output job IDs in parent task metadata
+    # Need to pass state as well otherwise its overwritten to None
+    self.update_state(
+        state=states.STARTED,
+        meta={"input_job_id": input_job_id, "output_job_id": output_job_id},
+    )
+
+    input_job = process_file_streaming.AsyncResult(input_job_id)
+    output_job = process_streaming_output.AsyncResult(output_job_id)
+
+    terminal_statuses = ["SUCCESS", "FAILURE", "REVOKED"]
+
+    while (
+        input_job.status not in terminal_statuses
+        or output_job.status not in terminal_statuses
+    ):
+        time.sleep(1)
+
+    logger.info("Both input and output jobs complete")
+
+    # Update parent task status to SUCCESS and pass metadata again
+    # otherwise its overwritten to None
+    self.update_state(
+        state=states.SUCCESS,
+        meta={"input_job_id": input_job_id, "output_job_id": output_job_id},
+    )
+    raise Ignore()
+
+
 @app.task(
     name="process_file_streaming", track_started=True, bind=True, serializer="pickle"
 )
-def process_file_streaming(self, agent: Agent):
-    # Get own job ID to set Consumer topic accordingly
-    job_id = self.request.id
-    agent.environment.kafka_input_topic = get_input_topic(job_id)
-    agent.environment.kafka_output_topic = get_output_topic(job_id)
+def process_file_streaming(self, agent: Agent, parent_job_id: str):
+    # Set input and output topics using parent job ID
+    agent.environment.kafka_input_topic = get_input_topic(parent_job_id)
+    agent.environment.kafka_output_topic = get_output_topic(parent_job_id)
 
     # Run the agent
     asyncio.run(agent.arun())
 
 
 async def async_process_streaming_output(
-    input_job_id: str, result_handler: ResultHandler, batch_size: int
+    input_job_id: str,
+    parent_job_id: str,
+    result_handler: ResultHandler,
+    batch_size: int,
 ):
-    logger.info(f"Polling for results {input_job_id=}")
+    logger.info(f"Polling for results {parent_job_id=}")
+
+    topic = get_output_topic(parent_job_id)
 
-    topic = get_output_topic(input_job_id)
     settings = Settings()
 
     consumer = AIOKafkaConsumer(
@@ -61,7 +131,7 @@ async def async_process_streaming_output(
         auto_offset_reset="earliest",
     )
     await consumer.start()
-    logger.info(f"consumer started {input_job_id=}")
+    logger.info(f"consumer started {parent_job_id=}")
 
     input_job_running = True
 
@@ -70,14 +140,13 @@ async def async_process_streaming_output(
             data = await consumer.getmany(timeout_ms=3000, max_records=batch_size)
             for tp, messages in data.items():
                 if messages:
-                    logger.debug(f"Handling {messages=} in topic {tp.topic}")
+                    logger.info(f"Handling {messages=} in topic {tp.topic}")
                     data = [msg.value for msg in messages]
                     result_handler(data)
-                    logger.debug(
-                        f"Handled {len(messages)} messages in topic {tp.topic}"
-                    )
+                    logger.info(f"Handled {len(messages)} messages in topic {tp.topic}")
                 else:
-                    logger.debug(f"No messages in topic {tp.topic}")
+                    logger.info(f"No messages in topic {tp.topic}")
+
             if not data:
                 logger.info(f"No messages in any topic")
         finally:
@@ -96,16 +165,23 @@ async def async_process_streaming_output(
     name="process_streaming_output", track_started=True, bind=True, serializer="pickle"
 )
 def process_streaming_output(
-    self, job_id: str, result_handler: ResultHandler, batch_size: int = 2
+    self,
+    input_job_id: str,
+    parent_job_id: str,
+    result_handler: ResultHandler,
+    batch_size: int,
 ):
     try:
-        asyncio.run(async_process_streaming_output(job_id, result_handler, batch_size))
+        asyncio.run(
+            async_process_streaming_output(
+                input_job_id, parent_job_id, result_handler, batch_size
+            )
+        )
     except Exception as e:
         # Set own status to failure
         self.update_state(state=states.FAILURE)
 
-        logger.error(msg=e)
+        logger.log(level=logging.ERROR, msg=e)
 
         # Ignore the task so no other state is recorded
-        # TODO check if this updates state to Ignored, or keeps Failed
         raise Ignore()