The Fix
Addresses the issue of connections being lost from the pool when a check operation is cancelled by handling CancelledError explicitly.
Based on closed psycopg/psycopg issue #1208 · PR/commit linked
Production note: Watch p95/p99 latency and retry volume; timeouts can turn into retry storms and duplicate side-effects.
@@ -7,6 +7,16 @@
==============================
+Future releases
+---------------
+
repro.py
import asyncio
from asyncio import CancelledError
from unittest.mock import create_autospec, AsyncMock
import psycopg
import psycopg_pool
from psycopg_pool import PoolTimeout
DSN = ""
TIMEOUT_POOL = False
async def main():
pool: psycopg_pool.AsyncConnectionPool = psycopg_pool.AsyncConnectionPool(
DSN,
open=False,
min_size=1,
max_size=1,
)
await pool.open()
try:
for i in range(2):
orig_conn = await pool.getconn()
await orig_conn.execute("select 1;")
# fake connection based on the real one
conn = create_autospec(psycopg.AsyncConnection)
for attr, val in vars(orig_conn).items():
setattr(conn, attr, val)
# simulate cancellation during rollback
if TIMEOUT_POOL:
conn.rollback = AsyncMock(side_effect=[CancelledError()])
try:
await pool.putconn(conn)
except BaseException as ex:
print(f"CancelledError caught on iteration {i}: {ex!r}")
except PoolTimeout as error:
print(f"PoolTimeout caught: {error!r}")
else:
print("No PoolTimeout caught")
finally:
# await pool.close()
pass
if __name__ == "__main__":
asyncio.run(main())
verify
Re-run the minimal reproduction on your broken version, then apply the fix and re-run.
fix.md
Option A — Apply the official fix\nAddresses the issue of connections being lost from the pool when a check operation is cancelled by handling CancelledError explicitly.\nWhen NOT to use: This fix should not be used if handling BaseException is undesirable.\n\n
Why This Fix Works in Production
- Trigger: AsyncConnectionPool: connections can leak on CancelledError or (gRPC task cancellation)
- Mechanism: Connections leak from the async pool when CancelledError is raised, not handled properly
Production impact:
- If left unfixed, this can cause silent data inconsistencies that propagate (bad cache entries, incorrect downstream decisions).
Why This Breaks in Prod
- Connections leak from the async pool when CancelledError is raised, not handled properly
- Surfaces as: AsyncConnectionPool: connections can leak on CancelledError or (gRPC task cancellation)
Proof / Evidence
- GitHub issue: #1208
- Fix PR: https://github.com/psycopg/psycopg/pull/1214
- Reproduced locally: No (not executed)
- Last verified: 2026-02-09
- Confidence: 0.70
- Did this fix it?: Yes (upstream fix exists)
- Own content ratio: 0.47
Discussion
High-signal excerpts from the issue thread (symptoms, repros, edge-cases).
“Thank you for the repro, amazing! I'll play with it and see which holes it drills in the pool, and will try to make it…”
“Thank you for the report, it might be a manifestation od other issues that have been reported, such as in #1123. Are you able to…”
“> Thank you for the repro, amazing! I'll play with it and see which holes it drills in the pool, and will try to make…”
“> Thank you for the report, it might be a manifestation od other issues that have been reported, such as in #1123”
Failure Signature (Search String)
- AsyncConnectionPool: connections can leak on CancelledError or (gRPC task cancellation)
Error Message
Stack trace
error.txt
Error Message
-------------
AsyncConnectionPool: connections can leak on CancelledError or (gRPC task cancellation)
Minimal Reproduction
repro.py
import asyncio
from asyncio import CancelledError
from unittest.mock import create_autospec, AsyncMock
import psycopg
import psycopg_pool
from psycopg_pool import PoolTimeout
DSN = ""
TIMEOUT_POOL = False
async def main():
pool: psycopg_pool.AsyncConnectionPool = psycopg_pool.AsyncConnectionPool(
DSN,
open=False,
min_size=1,
max_size=1,
)
await pool.open()
try:
for i in range(2):
orig_conn = await pool.getconn()
await orig_conn.execute("select 1;")
# fake connection based on the real one
conn = create_autospec(psycopg.AsyncConnection)
for attr, val in vars(orig_conn).items():
setattr(conn, attr, val)
# simulate cancellation during rollback
if TIMEOUT_POOL:
conn.rollback = AsyncMock(side_effect=[CancelledError()])
try:
await pool.putconn(conn)
except BaseException as ex:
print(f"CancelledError caught on iteration {i}: {ex!r}")
except PoolTimeout as error:
print(f"PoolTimeout caught: {error!r}")
else:
print("No PoolTimeout caught")
finally:
# await pool.close()
pass
if __name__ == "__main__":
asyncio.run(main())
What Broke
Under high load, the pool stops handing out connections, leading to timeouts.
Why It Broke
Connections leak from the async pool when CancelledError is raised, not handled properly
Fix Options (Details)
Option A — Apply the official fix
Addresses the issue of connections being lost from the pool when a check operation is cancelled by handling CancelledError explicitly.
When NOT to use: This fix should not be used if handling BaseException is undesirable.
Fix reference: https://github.com/psycopg/psycopg/pull/1214
Last verified: 2026-02-09. Validate in your environment.
When NOT to Use This Fix
- This fix should not be used if handling BaseException is undesirable.
Verify Fix
verify
Re-run the minimal reproduction on your broken version, then apply the fix and re-run.
Did This Fix Work in Your Case?
Quick signal helps us prioritize which fixes to verify and improve.
Prevention
- Track RSS + object counts after deployments; alert on monotonic growth and GC pressure.
- Add a long-running test that repeats the failing call path and asserts stable memory.
- Make timeouts explicit and test them (unit + integration) to avoid silent behavior changes.
- Instrument retries (attempt count + reason) and alert on spikes to catch dependency slowdowns.
Related Issues
No related fixes found.
Sources
We don’t republish the full GitHub discussion text. Use the links above for context.