Tim Uckun <timuckun@gmail.com> writes:
>> Well, I'm not asking for perfect reproducibility --- a test case that
>> fails even 1% of the time would be great.
> What exactly do you need?
A self-contained test case (code and data) that triggers the error.
If it only does so probabilistically, once in every-so-many runs,
that's fine.
> The database is not too large but the data is proprietary. Despite
> this I am willing to provide a sampling of the data in the two tables
> involved.
Perhaps you could sanitize or anonymize the data? It's unlikely that a
bug of this sort has all that much to do with the exact data content.
Whether the bug would still show for a "sample" is a different question,
and one you'd have to resolve by experiment.
regards, tom lane