retry on locked errors when getting lost followers (#15187)

src/prefect/server/events/triggers.py

@@ -23,6 +23,7 @@
 from sqlalchemy.ext.asyncio import AsyncSession
 from typing_extensions import Literal, TypeAlias
 
+from prefect._internal.retries import retry_async_fn
 from prefect.logging import get_logger
 from prefect.server.database.dependencies import db_injector
 from prefect.server.database.interface import PrefectDBInterface
@@ -513,6 +514,13 @@ async def reactive_evaluation(event: ReceivedEvent, depth: int = 0):
                     await session.commit()
 
 
+# retry on operational errors to account for db flakiness with sqlite
+@retry_async_fn(max_attempts=3, retry_on_exceptions=(sa.exc.OperationalError,))
+async def get_lost_followers():
+    """Get followers that have been sitting around longer than our lookback"""
+    return await causal_ordering().get_lost_followers()
+
+
 async def periodic_evaluation(now: DateTime):
     """Periodic tasks that should be run regularly, but not as often as every event"""
     offset = await get_events_clock_offset()
@@ -523,7 +531,7 @@ async def periodic_evaluation(now: DateTime):
     # Any followers that have been sitting around longer than our lookback are never
     # going to see their leader event (maybe it was lost or took too long to arrive).
     # These events can just be evaluated now in the order they occurred.
-    for event in await causal_ordering().get_lost_followers():
+    for event in await get_lost_followers():
         await reactive_evaluation(event)
 
     async with automations_session() as session: