Great work, just one suggestion for a test.

Reviewed 15 of 15 files at r1, all commit messages.
Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @spilchen)

pkg/sql/logictest/testdata/logic_test/materialized_view line 329 at r1 (raw file):

statement ok
REFRESH MATERIALIZED VIEW mat_view_as_of AS OF SYSTEM TIME '$ts_after_drop_c'

Should we validate if before object creation?

Looks good. I just had a few questions.

Reviewed all commit messages.
Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @rafiss)

pkg/sql/create_view.go line 230 at r1 (raw file):

					}

					if creationTime.Less(mostRecentModTime) {

Is there any concurrency issue here if one of the dependent tables is changed immediately after this check? I assume once this create view commits, then other schema changes on the dependent tables will pick up the dependency. But what if the schema change and create view are happening at the same time?

pkg/sql/opt/optbuilder/create_view.go line 117 at r1 (raw file):

			Syntax: cv,
			Schema: schID,
			// We need the view query to include user-defined types as a 3-part name to

nit: should this comment be removed? I think it was copied above, which seems like a better place for it.

pkg/sql/logictest/testdata/logic_test/materialized_view line 275 at r1 (raw file):

statement ok
CREATE MATERIALIZED VIEW mat_view_as_of AS SELECT a FROM tab_as_of AS OF SYSTEM TIME '$ts_before_insert_1'

If I understand it correctly, the ASOF time applies only during the refresh of the MQT. Is it valid to specify an ASOF time and request NO DATA? For instance, is this something that we allow:

CREATE MATERIALIZED VIEW v1 AS SELECT * FROM tab_as_of AS OF SYSTEM TIME '-5s' WITH NO DATA

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

pkg/sql/create_view.go line 230 at r1 (raw file):

i believe we should be safe here. since this operation modifies the backrefs on the dependent tables, and a competing schema change would need to make its own modifications to the table descriptor, that means at commit time only one of them will be able to succeed.

pkg/sql/opt/optbuilder/create_view.go line 117 at r1 (raw file):

done

pkg/sql/logictest/testdata/logic_test/materialized_view line 275 at r1 (raw file):

good question. i added a test case for this, and although i don't see why anyone would want to use it, i don't see a reason to disallow it, as it still makes sense semantically.

pkg/sql/logictest/testdata/logic_test/materialized_view line 329 at r1 (raw file):

i added a test.

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

pkg/sql/create_view.go line 230 at r1 (raw file):

i tested this manually and the correct thing happened. i started the CREATE MATERIALIZED VIEW, then made it sleep after this check. then i ran an ALTER TABLE on the dependent table in a different session. this ALTER TABLE succeeded.

when the CREATE MATERIALIZED VIEW unpaused, it resumed execution, but encountered a transaction retry error. the CREATE MATERIALIZED VIEW was automatically retried, and on this second attempt it failed this check so returned the correct error.