sql: implement pg_my_temp_schema and pg_is_other_temp_schema

[nvanbenschoten]

Needed for #69010.

This PR adds a real implementation for the pg_my_temp_schema builtin
function. pg_my_temp_schema returns the OID of session's temporary schema, or 0
if the session has not yet created a temporary schema.

The PR then implements the pg_is_other_temp_schema builtin function.
pg_is_other_temp_schema returns true if the given OID is the OID of another
session's temporary schema. This can be useful, for example, to exclude other
sessions' temporary tables from a catalog display.

Release note (sql change): The pg_my_temp_schema builtin function now properly
returns the OID of the active session's temporary schema, if one exists.

Release note (sql change): The pg_is_other_temp_schema builtin function is now
supported, which returns whether the given OID is the OID of another session's
temporary schema.

Release justification: None, waiting for v22.1.

[cockroach-teamcity]

This change is 

[rafiss]

thanks for implementing these!

Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @nvanbenschoten)

---

pkg/sql/sem/builtins/pg_builtins.go, line 847 at r3 (raw file):

					return tree.NewDOid(0), nil
				}
				oid, err := ctx.Planner.ResolveOIDFromString(

would be nice if we could use planner.Descriptors().GetSchemaByName() (and avoid going through the internal executor). but i believe it's not possible to add that to the EvalPlanner interface because of the dependencies it has.

how would you feel about adding a test that uses this builtin to rttanalysis.BenchmarkORMQueries? then if we resolve #69495 or improve this in some other way, we'd be able to measure it

---

pkg/sql/sem/builtins/pg_builtins.go, line 875 at r4 (raw file):

			Fn: func(ctx *tree.EvalContext, args tree.Datums) (tree.Datum, error) {
				schemaArg := tree.UnwrapDatum(ctx, args[0])
				schema, err := getNameForArg(ctx, schemaArg, "pg_namespace", "nspname")

similarly, could you add a benchmark for this?

[nvanbenschoten]

TFTR!

Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @rafiss)

---

pkg/sql/sem/builtins/pg_builtins.go, line 847 at r3 (raw file):

Previously, rafiss (Rafi Shamim) wrote…

would be nice if we could use planner.Descriptors().GetSchemaByName() (and avoid going through the internal executor). but i believe it's not possible to add that to the EvalPlanner interface because of the dependencies it has.

how would you feel about adding a test that uses this builtin to rttanalysis.BenchmarkORMQueries? then if we resolve #69495 or improve this in some other way, we'd be able to measure it

Done.

BenchmarkORMQueries/pg_my_temp_schema-16         	     177	   6485745 ns/op	         2.000 roundtrips

---

pkg/sql/sem/builtins/pg_builtins.go, line 875 at r4 (raw file):

Previously, rafiss (Rafi Shamim) wrote…

similarly, could you add a benchmark for this?

Done.

BenchmarkORMQueries/pg_is_other_temp_schema-16         	     100	  12407798 ns/op	         3.000 roundtrips

[rafiss]

lgtm! would be good to get a look from sql-schema, specifically for awareness about the ResolveOIDFromString and getNameForArg calls, just in case there's an immediately obvious improvement to be had.

[ajwerner]

Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @nvanbenschoten and @rafiss)

---

pkg/bench/rttanalysis/orm_queries_bench_test.go, line 187 at r8 (raw file):

Quoted 12 lines of code…

		{
			Name: "pg_my_temp_schema",
			Setup: `SET experimental_enable_temp_tables = true;
              CREATE TEMP TABLE t(a int primary key, b int)`,
			Stmt: `SELECT pg_my_temp_schema()`,
		},
		{
			Name: "pg_is_other_temp_schema",
			Setup: `SET experimental_enable_temp_tables = true;
              CREATE TEMP TABLE t(a int primary key, b int)`,
			Stmt: `SELECT pg_is_other_temp_schema('t'::regclass)`,
		},

thoughts on adding another one that does it twice in the same statement and shows that that is currently worse than doing it once?

[nvanbenschoten]

Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @rafiss)

---

pkg/bench/rttanalysis/orm_queries_bench_test.go, line 187 at r8 (raw file):

Previously, ajwerner wrote…

		{
			Name: "pg_my_temp_schema",
			Setup: `SET experimental_enable_temp_tables = true;
              CREATE TEMP TABLE t(a int primary key, b int)`,
			Stmt: `SELECT pg_my_temp_schema()`,
		},
		{
			Name: "pg_is_other_temp_schema",
			Setup: `SET experimental_enable_temp_tables = true;
              CREATE TEMP TABLE t(a int primary key, b int)`,
			Stmt: `SELECT pg_is_other_temp_schema('t'::regclass)`,
		},

thoughts on adding another one that does it twice in the same statement and shows that that is currently worse than doing it once?

Done for both.

name                                                  time/op
ORMQueries/pg_my_temp_schema-16                       6.77ms ± 0%
ORMQueries/pg_my_temp_schema_multiple_times-16        6.54ms ± 0%
ORMQueries/pg_is_other_temp_schema-16                 7.41ms ± 0%
ORMQueries/pg_is_other_temp_schema_multiple_times-16  11.3ms ± 0%

name                                                  roundtrips
ORMQueries/pg_my_temp_schema-16                         2.00 ± 0%
ORMQueries/pg_my_temp_schema_multiple_times-16          2.00 ± 0%
ORMQueries/pg_is_other_temp_schema-16                   6.99 ± 0%
ORMQueries/pg_is_other_temp_schema_multiple_times-16    11.0 ± 0%

[nvanbenschoten]

bors r+

[craig]

Build succeeded:

• GitHub CI (Cockroach)