sql: fix partial index and foreign key bugs

[mgartner]

opt: build partial index predicates only when all table columns are in scope

This commit fixes a panic induced by trying to build partial index
predicate expressions without all of a table's columns in-scope. Some
scans, like scans built for foreign key checks, do not include all of a
table's column in their scope. For such scans, optbuilder no longer
attempts to build a partial index predicate because the predicate may
reference a column that is not in scope.

As a result of this change, opt.TableMeta may not have a predicate
expression for all partial indexes. The memo.PartialIndexPredicate
function which retrieves the predicate expressions has been updated to
account for this case.

Fixes #55672

Release justification: This is a critical bug fix for a new feature,
partial indexes.

Release note (bug fix): An INSERT into a table with a foreign key
reference to a table with a partial index no longer causes an error.

sql: disqualify partial unique indexes as foreign key reference indexes

Release justification: This is a critical bug fix for a new feature,
partial indexes.

Release note (bug fix): Foreign keys can no longer reference columns
that are only indexed by a partial unique index. A partial unique index
does not guarantee uniqueness in the entire table, therefore the column
indexed is not guaranteed to be a unique key.

[cockroach-teamcity]

This change is 

[RaduBerinde]

Thanks for the quick investigation and fix!

Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @mgartner and @RaduBerinde)

---

pkg/sql/opt/optbuilder/select.go, line 531 at r2 (raw file):

		// index predicates cannot be built because they may reference columns
		// not in outScope.
		containsAllOrdinaryTableColumns := true

[nit] We could skip the loop if we have tab.ColmnCount() columns (which I believe is the common case)

[andy-kimball]

If the customer that ran into this issue creates a full index on the column (in addition to existing partial index), would it "fix" their issue? Can we tell them to workaround the problem by doing this?

[rytaft]

Reviewed 6 of 6 files at r1, 3 of 3 files at r2.
Reviewable status: complete! 2 of 0 LGTMs obtained (waiting on @mgartner)

---

pkg/sql/opt/memo/expr.go, line 909 at r1 (raw file):

// Note that TableMeta.PartialIndexPredicates is not initialized for mutation
// queries. This function will never return ok=true in the context of a mutation.
func PartialIndexPredicate(

[nit] Do you need this function anymore? Doesn't seem like it's adding much on top of TableMeta.PartialIndexPredicates...

[RaduBerinde]

No, they already had one (the PK).

[mgartner]

Reviewable status: complete! 0 of 0 LGTMs obtained (and 2 stale) (waiting on @rytaft)

---

pkg/sql/opt/memo/expr.go, line 909 at r1 (raw file):

Previously, rytaft (Rebecca Taft) wrote…

[nit] Do you need this function anymore? Doesn't seem like it's adding much on top of TableMeta.PartialIndexPredicates...

Good point. It's also used far less with the recent changes to scanIndexIter. I've removed it here, but will not remove it in the backport to minimize changes.

---

pkg/sql/opt/optbuilder/select.go, line 531 at r2 (raw file):

Previously, RaduBerinde wrote…

[nit] We could skip the loop if we have tab.ColmnCount() columns (which I believe is the common case)

Good idea. Done.

[rytaft]

Reviewed 3 of 6 files at r3, 3 of 3 files at r4.
Reviewable status: complete! 1 of 0 LGTMs obtained (and 1 stale) (waiting on @mgartner)

---

pkg/sql/opt/optbuilder/select.go, line 531 at r4 (raw file):

		// index predicates cannot be built because they may reference columns
		// not in outScope. In the most common case, the outScope has the same
		// number of columns as the table and we can skip checking the each

[nit] the each -> that each

[mgartner]

Reviewable status: complete! 0 of 0 LGTMs obtained (and 2 stale) (waiting on @rytaft)

---

pkg/sql/opt/optbuilder/select.go, line 531 at r4 (raw file):

Previously, rytaft (Rebecca Taft) wrote…

[nit] the each -> that each

Done.

[mgartner]

bors r+

[mgartner]

bors r-

[craig]

Canceled.

[mgartner]

bors r+

[craig]

Build succeeded:

• GitHub CI (Cockroach)