Re: Too coarse predicate locks granularity for B+ tree indexes

On Tue, 2023-02-07 at 16:23 +0600, Rinat Shigapov wrote:
> I have a concurrent testsuite that runs 14 test cases. Each test case operates
> on a disjoint set of records, doesn't retry transactions and is run under
> 'serializable' isolation level. The test data is small and likely fits within
> a single tuple page.
>
> When I finished the test suite I was surprised that PostgreSQL 14.5 returns
> serialization failure on every test suite run.

This is no question for the hackers list; redirecting to general.

That behavior sounds perfectly normal to me: if everything is in a single
page, PostgreSQL probably won't use an index scan. With a sequential scan,
the predicate lock will be on the whole table. So you should expect
serialization failures. This is well documented.

Perhaps you should use a more realistic test case with a reasonable
amount of data.

Yours,
Laurenz Albe