AssertionError on queueing_lock conflict when PostgreSQL lc_messages is not English

[pararti]

Hi,

I ran into an issue with queueing_lock handling when PostgreSQL uses a non-English locale for server messages.

What happened

When a second job is enqueued with the same queueing_lock, Procrastinate is expected to raise AlreadyEnqueued.

Instead, it crashes with an AssertionError inside sync_psycopg_connector.wrap_exceptions().

The failure happens here:

match = re.search(r"Key \((.*?)\)=\((.*?)\)", exc.diag.message_detail)
assert match

This seems to assume that exc.diag.message_detail is always in English.

Environment

• Procrastinate: 3.7.2
• psycopg: 3.3.3
• PostgreSQL: 17.8
• lc_messages = ru_RU.UTF-8

Expected behavior

A queueing lock conflict should raise procrastinate.exceptions.AlreadyEnqueued, regardless of PostgreSQL message locale.

Actual behavior

Procrastinate raises AssertionError while trying to parse the localized message_detail.

Stack trace

  File ".../procrastinate/psycopg_connector.py", line 52, in wrap_exceptions
    with sync_psycopg_connector.wrap_exceptions():
  File ".../procrastinate/sync_psycopg_connector.py", line 36, in wrap_exceptions
    assert match
  AssertionError

Why this is a problem

On installations where PostgreSQL uses a localized lc_messages, duplicate enqueue with queueing_lock becomes an internal library error instead of a normal AlreadyEnqueued flow.

Reproduction idea

1. Use PostgreSQL with lc_messages = ru_RU.UTF-8
2. Define a task and enqueue a job with a specific queueing_lock
3. Try to enqueue another job with the same queueing_lock
4. Observe AssertionError instead of AlreadyEnqueued

Note

In our case this happens on a shared PostgreSQL instance, so changing lc_messages is not a practical workaround.

It looks like parsing localized message_detail is too fragile here.