Partition queries broken

[rustprooflabs]

Details

The dd.partition_parents and dd.partition_children views are broken when partitions exist.

What I did

cargo pgrx run

Create extension & query view (default empty)

(localhost 🐘) ryanlambert@pgdd=# create extension pgdd;
CREATE EXTENSION
Time: 63.543 ms
(localhost 🐘) ryanlambert@pgdd=# select * from dd.partition_parents;
(0 rows)

Load test partition data

(localhost 🐘) ryanlambert@pgdd=# \i ~/git/pgdd/tests/test-partition-data.sql 
psql:/home/ryanlambert/git/pgdd/tests/test-partition-data.sql:5: NOTICE:  schema "pgdd_test" does not exist, skipping
DROP SCHEMA
Time: 1.053 ms
CREATE SCHEMA
Time: 2.004 ms
CREATE TABLE
Time: 2.069 ms
CREATE TABLE
Time: 5.403 ms
CREATE TABLE
Time: 5.205 ms
CREATE TABLE
Time: 5.409 ms
INSERT 0 15
Time: 2.895 ms

Then tried query against partition view again. It fails with a type error.

select * from dd.partition_parents;
ERROR:  Datum error: Postgres type bigint oid={#20, builtin: INT8OID} is not compatible with the Rust type pgrx_pg_sys::submodules::oids::Oid oid={#26, builtin: OIDOID}

Findings so far...

But... it is oid

The code behind the dd.partition_parent() function is pretty straight foward and should be returning an OID. After all, it's pulling directly from pg_catalog.pg_class.oid. While a CTE isn't helpful at all there, that also shouldn't impact the data type. I double checked the type returned via pg_typeof() to be paranoid.

SELECT pg_typeof(c.oid), c.oid,
        n.nspname::TEXT AS s_name,
        c.relname::TEXT AS t_name,
        CASE WHEN pt.partrelid IS NOT NULL THEN 'declarative'
            WHEN c.relkind = 'r' THEN 'inheritance'
            ELSE 'unknown' END AS partition_type,
        COUNT(i.inhrelid) AS partitions
    FROM pg_catalog.pg_class c
    INNER JOIN pg_catalog.pg_inherits i ON c.oid = i.inhparent
    LEFT JOIN pg_catalog.pg_partitioned_table pt
        ON c.oid = pt.partrelid 
    LEFT JOIN pg_namespace n ON n.oid = c.relnamespace
    WHERE c.relkind != 'I' -- Exclude partitioned indexes
    GROUP BY c.relkind, c.oid, pt.partrelid, n.nspname::TEXT, c.relname::TEXT
;

┌───────────┬───────┬───────────┬────────┬────────────────┬────────────┐
│ pg_typeof │  oid  │  s_name   │ t_name │ partition_type │ partitions │
╞═══════════╪═══════╪═══════════╪════════╪════════════════╪════════════╡
│ oid       │ 42225 │ pgdd_test │ parent │ declarative    │          3 │
└───────────┴───────┴───────────┴────────┴────────────────┴────────────┘

Rust code

I don't see any obvious differences in the Rust code used regarding how we are handling the OID. Example, no issue here.

SELECT oid, pg_typeof(oid), * FROM dd.tables;

Comparing what is done for the tables function vs the partition functions, I don't see any obvious differences.
Functional tables code does this (here and here):

name!(oid, Result<Option<pg_sys::Oid>, pgrx::spi::Error>)
...
let oid = row["oid"].value::<pg_sys::Oid>();

The broken partition_parents version does the same things (here and here)

name!(oid, Result<Option<pg_sys::Oid>, pgrx::spi::Error>),
...
let oid = row["oid"].value::<pg_sys::Oid>();

[rustprooflabs]

Ah, the casting was happening in the partition_children query... See #48.

[rustprooflabs]

Verified this fixes the problem on Pg15 where it was detected. Will tag and release new version 0.5.2 including new installers.