sql: fix "no user specified" errors in statement bundles

[mgartner]

A regression was introduced in #71246 that caused errors when running
internal queries to populate files in statement bundles. As a result,
critical information was missing from the env.sql, schema.sql, and
stats*.sql files. This commit fixes the issue by using the internal
executor factory to create an internal executor with the current
session's session data.

Fixes #80396

Release note: None

[cockroach-teamcity]

This change is 

[mgartner]

Is it safe to use a root user override here? If not, what is a better alternative?

[knz]

Reviewed all commit messages.
Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @mgartner, @rafiss, and @RichardJCai)

---

pkg/sql/explain_bundle.go, line 438 at r1 (raw file):

Previously, mgartner (Marcus Gartner) wrote…

Is it safe to use a root user override here? If not, what is a better alternative?

This looks suspicious. Why can't this use the identity of the user requesting the bundle?

[knz]

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

---

pkg/sql/explain_bundle.go, line 438 at r1 (raw file):

Previously, knz (kena) wrote…

This looks suspicious. Why can't this use the identity of the user requesting the bundle?

As in, make stmtEnvCollector take the requester's username as argument or something like that.

[rafiss]

a node user is insufficient - it does not have
privileges on user-defined relations which is needed for the
SHOW CREATE queries that populate schema.sql.

was this fixed by #79098 ?

[mgartner]

How about this second attempt that doesn't use an override at all?

[knz]

is it already common practice to instantiate InternalExecutors on-the-fly? Is this an expensive operation?

(Asking for a friend - if we don't need to keep a predefined instance around, I would cut down on the server code)

[mgartner]

is it already common practice to instantiate InternalExecutors on-the-fly? Is this an expensive operation?

I copied this pattern from:

cockroach/pkg/sql/opt_exec_factory.go

Lines 1160 to 1163 in 2115a44

	ie := ef.planner.extendedEvalCtx.ExecCfg.InternalExecutorFactory(
	ef.planner.EvalContext().Context,
	ef.planner.SessionData(),
	)

It seems fairly prolific given that planner.QueryRowEx, which is used in numerous places, creates a new internal executor:

cockroach/pkg/sql/planner.go

Line 879 in d4daa99

ie := p.ExecCfg().InternalExecutorFactory(ctx, p.SessionData())

[knz]

Amazing. TIL. Thanks for explaining.

[mgartner]

TFTR!

bors r+

[craig]

Build succeeded:

• GitHub CI (Cockroach)