fix db issues

applications/modelling_e2e/handler.py

@@ -23,6 +23,7 @@ invocations (ADR-0012).
 
 from __future__ import annotations
 
+import dataclasses
 import io
 import os
 from collections.abc import Callable
@@ -92,7 +93,11 @@ logger = setup_logger()
 def _get_engine() -> Engine:
     global _engine
     if _engine is None:
-        _engine = make_engine(PostgresConfig.from_env(dict(os.environ)))
+        config = PostgresConfig.from_env(dict(os.environ))
+        # Reduced pool for Lambda: 32 concurrent containers × 3 connections = 96 max,
+        # vs the default 3+5=8 which would reach 256+ and exhaust RDS max_connections.
+        # pool_size=2 covers the simultaneous read_session + UoW session per invocation.
+        _engine = make_engine(dataclasses.replace(config, pool_size=2, max_overflow=1))
     return _engine
 
 

infrastructure/postgres/engine.py

@@ -1,20 +1,25 @@
 from collections.abc import Iterator
 from contextlib import contextmanager
+from typing import Any, Optional, Type
 
 from sqlalchemy.engine import Engine
+from sqlalchemy.pool import Pool
 from sqlmodel import Session, create_engine
 
 from infrastructure.postgres.config import PostgresConfig
 
 
-def make_engine(config: PostgresConfig) -> Engine:
-    return create_engine(
-        config.url(),
-        pool_size=config.pool_size,
-        max_overflow=config.max_overflow,
-        pool_pre_ping=config.pool_pre_ping,
-        pool_recycle=config.pool_recycle,
-    )
+def make_engine(
+    config: PostgresConfig, poolclass: Optional[Type[Pool]] = None
+) -> Engine:
+    kwargs: dict[str, Any] = {"pool_pre_ping": config.pool_pre_ping}
+    if poolclass is None:
+        kwargs["pool_size"] = config.pool_size
+        kwargs["max_overflow"] = config.max_overflow
+        kwargs["pool_recycle"] = config.pool_recycle
+    else:
+        kwargs["poolclass"] = poolclass
+    return create_engine(config.url(), **kwargs)
 
 
 def make_session(engine: Engine) -> Session:

scripts/trigger_modelling_e2e_sqs.py

@@ -22,7 +22,7 @@ SQS_QUEUE_NAME: str = "modelling_e2e-queue-dev"
 
 # Number of postcodes to process this run (postcodes where all properties are
 # already completed are skipped and do not count toward this limit).
-POSTCODES_LIMIT: int = 100
+POSTCODES_LIMIT: int = 1000
 
 # True → Lambda runs the full pipeline but skips all DB writes (safe for testing).
 DRY_RUN: bool = False
@@ -73,16 +73,21 @@ def _completed_property_ids() -> set[int]:
     with engine.connect() as conn:
         rows = conn.execute(
             text("""
-                SELECT DISTINCT (elem.value::int) AS property_id
+                SELECT DISTINCT elem.value::int AS property_id
                 FROM sub_task st
                 JOIN tasks t ON t.id = st.task_id
-                CROSS JOIN jsonb_array_elements_text(st.inputs->'property_ids') AS elem(value)
+                CROSS JOIN jsonb_array_elements_text(
+                    (st.inputs::jsonb)->'property_ids'
+                ) AS elem(value)
                 WHERE t.task_source = 'modelling_e2e'
-                  AND st.status = 'completed'
-                  AND (st.inputs->>'portfolio_id')::int = :portfolio_id
-                  AND (st.inputs->>'scenario_id')::int = :scenario_id
+                AND st.status = 'complete'
+                AND ((st.inputs::jsonb)->>'portfolio_id')::int = :portfolio_id
+                AND ((st.inputs::jsonb)->>'scenario_id')::int = :scenario_id
             """),
-            {"portfolio_id": PORTFOLIO_ID, "scenario_id": SCENARIO_ID},
+            {
+                "portfolio_id": PORTFOLIO_ID,
+                "scenario_id": SCENARIO_ID,
+            },
         ).fetchall()
     return {int(r[0]) for r in rows}
 

utilities/aws_lambda/default_orchestrator.py

@@ -2,6 +2,7 @@ import os
 from collections.abc import Generator
 from contextlib import contextmanager
 
+from sqlalchemy.pool import NullPool
 from sqlmodel import Session
 
 from infrastructure.postgres.config import PostgresConfig
@@ -17,8 +18,11 @@ def default_orchestrator() -> Generator[TaskOrchestrator, None, None]:
 
     Connection params come from os.environ via PostgresConfig.from_env. Each
     handler invocation gets its own session, cleaned up on context exit.
+
+    NullPool is intentional: a new engine is created on every invocation, so
+    pooling would accumulate idle connections across warm Lambda containers.
     """
-    engine = make_engine(PostgresConfig.from_env(dict(os.environ)))
+    engine = make_engine(PostgresConfig.from_env(dict(os.environ)), poolclass=NullPool)
     with Session(engine) as session:
         yield TaskOrchestrator(
             task_repo=TaskPostgresRepository(session=session),