Internal: message processing is now event-based

Problem: the pending message fetcher and processor use a polling loop to
look for messages to fetch/process. This leads to some latency when the
pending_messages table is empty as the task sleeps while waiting for new
pending messages.

Solution: add an exchange + queue in RabbitMQ to signal the arrival of
new messages. To avoid modifying the message processor too much and
avoid depending on coherency between the DB and RabbitMQ, the fetcher
and processor simply spawn a new task that looks for messages and sets
an asyncio Event object. The main fetching/processing loop waits on this
event (with a timeout).

Note that this system is not used for retries as this would require
another task that posts messages to the MQ on their next attempt.
Retried messages simply wait for the next iteration of the loop
(every second).

This solution has the following advantages and drawbacks:
+ No more arbitrary latency when processing new messages
+ No major modification of the pipeline, even if the MQ system fails
  for some reason the pending message processor will still process
  messages every second
+ No dependency on the state of the message queue, if the RabbitMQ queue
  is deleted for any reason the processor will keep on working
- RabbitMQ overhead (one more exchange + queue).

Author: odesenfans

src/aleph/commands.py

@@ -45,6 +45,8 @@
 __copyright__ = "Moshe Malawach"
 __license__ = "mit"
 
+from aleph.toolkit.rabbitmq import make_mq_conn
+
 LOGGER = logging.getLogger(__name__)
 
 
@@ -129,6 +131,9 @@ async def main(args: List[str]) -> None:
 
     setup_logging(args.loglevel)
 
+    mq_conn = await make_mq_conn(config)
+    mq_channel = await mq_conn.channel()
+
     node_cache = await init_node_cache(config)
     ipfs_service = IpfsService(ipfs_client=make_ipfs_client(config))
     storage_service = StorageService(
@@ -172,11 +177,12 @@ async def main(args: List[str]) -> None:
     LOGGER.debug("Initialized p2p")
 
     LOGGER.debug("Initializing listeners")
-    tasks += listener_tasks(
+    tasks += await listener_tasks(
         config=config,
         session_factory=session_factory,
         node_cache=node_cache,
         p2p_client=p2p_client,
+        mq_channel=mq_channel,
     )
     tasks.append(chain_connector.chain_event_loop(config))
     LOGGER.debug("Initialized listeners")

src/aleph/db/accessors/pending_messages.py

@@ -76,6 +76,13 @@ def get_pending_messages(
     return session.execute(select_stmt).scalars()
 
 
+def get_pending_message(session: DbSession, pending_message_id: int) -> Optional[PendingMessageDb]:
+    select_stmt = select(PendingMessageDb).where(
+        PendingMessageDb.id == pending_message_id
+    )
+    return session.execute(select_stmt).scalar_one_or_none()
+
+
 def count_pending_messages(session: DbSession, chain: Optional[Chain] = None) -> int:
     """
     Counts pending messages.

src/aleph/handlers/message_handler.py

@@ -2,6 +2,7 @@
 import logging
 from typing import Optional, Dict, Any, Mapping
 
+import aio_pika.abc
 import psycopg2
 import sqlalchemy.exc
 from aleph_message.models import MessageType, ItemType
@@ -175,12 +176,21 @@ def __init__(
         session_factory: DbSessionFactory,
         storage_service: StorageService,
         config: Config,
+        pending_message_exchange: aio_pika.abc.AbstractExchange,
     ):
         super().__init__(
             storage_service=storage_service,
             config=config,
         )
         self.session_factory = session_factory
+        self.pending_message_exchange = pending_message_exchange
+
+    async def _publish_pending_message(self, pending_message: PendingMessageDb) -> None:
+        mq_message = aio_pika.Message(body=f"{pending_message.id}".encode("utf-8"))
+        process_or_fetch = "process" if pending_message.fetched else "fetch"
+        await self.pending_message_exchange.publish(
+            mq_message, routing_key=f"{process_or_fetch}.{pending_message.item_hash}"
+        )
 
     async def add_pending_message(
         self,
@@ -241,7 +251,6 @@ async def add_pending_message(
                 session.execute(upsert_message_status_stmt)
                 session.execute(insert_pending_message_stmt)
                 session.commit()
-                return pending_message
 
             except (psycopg2.Error, sqlalchemy.exc.SQLAlchemyError) as e:
                 LOGGER.warning(
@@ -259,6 +268,9 @@ async def add_pending_message(
                 session.commit()
                 return None
 
+            await self._publish_pending_message(pending_message)
+            return pending_message
+
 
 class MessageHandler(BaseMessageHandler):
     """
@@ -299,7 +311,6 @@ async def confirm_existing_message(
                 )
             )
 
-
     async def insert_message(
         self, session: DbSession, pending_message: PendingMessageDb, message: MessageDb
     ):

src/aleph/jobs/fetch_pending_messages.py

@@ -13,17 +13,19 @@
     NewType,
 )
 
+import aio_pika.abc
 from configmanager import Config
 from setproctitle import setproctitle
 
-from ..chains.signature_verifier import SignatureVerifier
+from aleph.chains.signature_verifier import SignatureVerifier
 from aleph.db.accessors.pending_messages import (
     make_pending_message_fetched_statement,
     get_next_pending_messages,
 )
 from aleph.db.connection import make_engine, make_session_factory
-from aleph.db.models import PendingMessageDb, MessageDb
+from aleph.db.models import MessageDb, PendingMessageDb
 from aleph.handlers.message_handler import MessageHandler
+from aleph.services.cache.node_cache import NodeCache
 from aleph.services.ipfs import IpfsService
 from aleph.services.ipfs.common import make_ipfs_client
 from aleph.services.storage.fileystem_engine import FileSystemStorageEngine
@@ -32,8 +34,8 @@
 from aleph.toolkit.monitoring import setup_sentry
 from aleph.toolkit.timestamp import utc_now
 from aleph.types.db_session import DbSessionFactory
-from .job_utils import prepare_loop, MessageJob
-from ..services.cache.node_cache import NodeCache
+from .job_utils import prepare_loop, MessageJob, make_pending_message_queue
+from ..toolkit.rabbitmq import make_mq_conn
 
 LOGGER = getLogger(__name__)
 
@@ -47,12 +49,15 @@ def __init__(
         session_factory: DbSessionFactory,
         message_handler: MessageHandler,
         max_retries: int,
+        pending_message_queue: aio_pika.abc.AbstractQueue,
     ):
         super().__init__(
             session_factory=session_factory,
             message_handler=message_handler,
             max_retries=max_retries,
+            pending_message_queue=pending_message_queue,
         )
+        self.pending_message_queue = pending_message_queue
 
     async def fetch_pending_message(self, pending_message: PendingMessageDb):
         with self.session_factory() as session:
@@ -76,6 +81,7 @@ async def fetch_pending_message(self, pending_message: PendingMessageDb):
                     exception=e,
                 )
                 session.commit()
+                return None
 
     async def fetch_pending_messages(
         self, config: Config, node_cache: NodeCache, loop: bool = True
@@ -140,8 +146,11 @@ async def fetch_pending_messages(
                         break
                     # If we are done, wait a few seconds until retrying
                     if not fetch_tasks:
-                        LOGGER.info("waiting 1 second(s) for new pending messages...")
-                        await asyncio.sleep(1)
+                        LOGGER.info("waiting for new pending messages...")
+                        try:
+                            await asyncio.wait_for(self.ready(), 1)
+                        except TimeoutError:
+                            pass
 
     def make_pipeline(
         self,
@@ -156,12 +165,16 @@ def make_pipeline(
 
 
 async def fetch_messages_task(config: Config):
-    # TODO: this sleep can probably be removed
-    await asyncio.sleep(4)
-
     engine = make_engine(config=config, application_name="aleph-fetch")
     session_factory = make_session_factory(engine)
 
+    mq_conn = await make_mq_conn(config=config)
+    mq_channel = await mq_conn.channel()
+
+    pending_message_queue = await make_pending_message_queue(
+        config=config, routing_key="fetch.*", channel=mq_channel
+    )
+
     node_cache = NodeCache(
         redis_host=config.redis.host.value, redis_port=config.redis.port.value
     )
@@ -182,10 +195,11 @@ async def fetch_messages_task(config: Config):
         session_factory=session_factory,
         message_handler=message_handler,
         max_retries=config.aleph.jobs.pending_messages.max_retries.value,
+        pending_message_queue=pending_message_queue,
     )
 
-    while True:
-        with session_factory() as session:
+    async with fetcher:
+        while True:
             try:
                 fetch_pipeline = fetcher.make_pipeline(
                     config=config, node_cache=node_cache
@@ -197,11 +211,10 @@ async def fetch_messages_task(config: Config):
                         )
 
             except Exception:
-                LOGGER.exception("Error in pending messages job")
-                session.rollback()
+                LOGGER.exception("Unexpected error in pending messages fetch job")
 
-        LOGGER.debug("Waiting 1 second(s) for new pending messages...")
-        await asyncio.sleep(1)
+            LOGGER.debug("Waiting 1 second(s) for new pending messages...")
+            await asyncio.sleep(1)
 
 
 def fetch_pending_messages_subprocess(config_values: Dict):

src/aleph/jobs/job_utils.py

@@ -1,10 +1,10 @@
 import asyncio
 import datetime as dt
 import logging
-from typing import Dict, Union
+from typing import Dict, Union, Optional
 from typing import Tuple
 
-import aio_pika.abc
+import aio_pika
 from configmanager import Config
 from sqlalchemy import update
 
@@ -29,6 +29,58 @@
 MAX_RETRY_INTERVAL: int = 300
 
 
+async def _make_pending_queue(
+    config: Config,
+    exchange_name: str,
+    queue_name: str,
+    routing_key: str,
+    channel: Optional[aio_pika.abc.AbstractChannel] = None,
+) -> aio_pika.abc.AbstractQueue:
+    if not channel:
+        mq_conn = await aio_pika.connect_robust(
+            host=config.p2p.mq_host.value,
+            port=config.rabbitmq.port.value,
+            login=config.rabbitmq.username.value,
+            password=config.rabbitmq.password.value,
+        )
+        channel = await mq_conn.channel()
+
+    exchange = await channel.declare_exchange(
+        name=exchange_name,
+        type=aio_pika.ExchangeType.TOPIC,
+        auto_delete=False,
+    )
+    queue = await channel.declare_queue(name=queue_name, auto_delete=False)
+    await queue.bind(exchange, routing_key=routing_key)
+    return queue
+
+
+async def make_pending_tx_queue(
+    config: Config, channel: aio_pika.abc.AbstractChannel
+) -> aio_pika.abc.AbstractQueue:
+    return await _make_pending_queue(
+        config=config,
+        exchange_name=config.rabbitmq.pending_tx_exchange.value,
+        queue_name="pending-tx-queue",
+        routing_key="#",
+        channel=channel,
+    )
+
+
+async def make_pending_message_queue(
+    config: Config,
+    routing_key: str,
+    channel: Optional[aio_pika.abc.AbstractChannel] = None,
+) -> aio_pika.abc.AbstractQueue:
+    return await _make_pending_queue(
+        config=config,
+        exchange_name=config.rabbitmq.pending_message_exchange.value,
+        queue_name="pending_message_queue",
+        routing_key=routing_key,
+        channel=channel,
+    )
+
+
 def compute_next_retry_interval(attempts: int) -> dt.timedelta:
     """
     Computes the time interval for the next attempt/retry of a message.
@@ -67,6 +119,8 @@ def schedule_next_attempt(
     set_next_retry(
         session=session, pending_message=pending_message, next_attempt=next_attempt
     )
+    pending_message.next_attempt = next_attempt
+    pending_message.retries += 1
 
 
 def prepare_loop(config_values: Dict) -> Tuple[asyncio.AbstractEventLoop, Config]:
@@ -125,19 +179,22 @@ async def ready(self):
         self._event.clear()
 
 
-class MessageJob:
+class MessageJob(MqWatcher):
     def __init__(
         self,
         session_factory: DbSessionFactory,
         message_handler: MessageHandler,
         max_retries: int,
+        pending_message_queue: aio_pika.abc.AbstractQueue,
     ):
+        super().__init__(mq_queue=pending_message_queue)
+
         self.session_factory = session_factory
         self.message_handler = message_handler
         self.max_retries = max_retries
 
+    @staticmethod
     def _handle_rejection(
-        self,
         session: DbSession,
         pending_message: PendingMessageDb,
         exception: BaseException,
@@ -158,7 +215,7 @@ def _handle_rejection(
 
         return RejectedMessage(pending_message=pending_message, error_code=error_code)
 
-    def _handle_retry(
+    async def _handle_retry(
         self,
         session: DbSession,
         pending_message: PendingMessageDb,
@@ -222,6 +279,6 @@ async def handle_processing_error(
                 session=session, pending_message=pending_message, exception=exception
             )
         else:
-            return self._handle_retry(
+            return await self._handle_retry(
                 session=session, pending_message=pending_message, exception=exception
             )