Review Checklist

Hello reviewers! 👋 Please follow this checklist when reviewing this Pull Request.

General

• Ensure that the Pull Request has a descriptive title.
• Ensure there is a link to an issue (except for internal cleanup and flaky test fixes), new features should have an RFC that documents use cases and test cases.

Tests

• Bug fixes should have at least one unit or end-to-end test, enhancement and new features should have a sufficient number of tests.

Documentation

• Apply the release notes (needs details) label if users need to know about this change.
• New features should be documented.
• There should be some code comments as to why things are implemented the way they are.
• There should be a comment at the top of each new or modified test to explain what the test does.

New flags

• Is this flag really necessary?
• Flag names must be clear and intuitive, use dashes (-), and have a clear help text.

If a workflow is added or modified:

• Each item in Jobs should be named in order to mark it as required.
• If the workflow needs to be marked as required, the maintainer team must be notified.

Backward compatibility

• Protobuf changes should be wire-compatible.
• Changes to _vt tables and RPCs need to be backward compatible.
• RPC changes should be compatible with vitess-operator
• If a flag is removed, then it should also be removed from vitess-operator and arewefastyet, if used there.
• vtctl command output order should be stable and awk-able.

Up to 7cee6d3, we deal with two situations that are actually manifestations of the same concept: dropping a foreign key.

When you drop a foreign key, you drop a constraint. When a constraint is dropped, your data becomes less constrained, and it is then possible for the data to take forms it could not, before. This is why reverting the drop of a foreign key is a problem: you reintroduce a constraint, and the data may already disagree with the constraint.

Moving on:

• Adding a FK constraint should not be an issue for the immediate revert. Of course, if you revert-the-revert then you're back in the "re-introduce a FK" problem. So in general, we don't need to worry about a CREATE TABLE.
• In MySQL, there's no ALTERing a constraint. A constraint can be created or dropped. So we don't need to sorry about modifying a constraint (ALTER CONSTRAINT ... ENFORCE is outside the scope of this discussion).
  Any change to a foreign key definition, e.g. expanding from one column covered to two columns covered or vice versa, is interpreted as a DROP CONSTRAINT + ADD CONSTRAINT. Which means we take care of its revertibility by sheer use of DROP CONSTRAINT as per above.
• In MySQL, it is impossible to change the data type of a column covered by a foreign key (!). Vitess might actually want to offer this functionality in the long term, and in this case the revert can be impacted by the expansion of the data type scope (e.g. int to bigint). The impact for foreign keys exists, but the scenario is already covered outside the scope of foreign keys.
• This leaves us with "normal" ALTER TABLE statements. Can there be a scenario where an ALTER TABLE does not add/drop foreign keys, does not change data type, and still has revertibility issues? What could be such scenarios?
  • An immediate suspect is a change to GENERATED column, but thankfully MySQL's foreign keys do not allow use of GENERATED columns neither in parent nor child.
  • There can be an issue if a FK covered column is impacted by a change to a CHECK constraint. This is actually outside the scope of FK. This is really complex because the CHECK constraint expression may be complex and we cannot anticipate exactly how it might limit/expand the possible values.

More thinking to come.

Noteworthy: at this time Vitess does not prevent the scenario where a FK parent table is DROPped. That is, it allows it. As reminder, in Online DDL tables aren't actually immediately dropped.

First, they are RENAMEd. Children's FK will migrate with the renamed table. What happens from that point on depends on whether this MySQL patch is used or not. If used, then then effective immediately these FK constraints are ignored by InnoDB.
Then, at some point later on, the table is actually dropped. Again, with said patch, children tables still ignore that fact and continue to live happily. But the schema becomes inconsistent, and a logical backup+restore will fail.

So at this time this is not being handled, and we assume that a child table gets dropped (or its foreign key constraints are dropped), before the parent table is dropped.

This is one of the reasons why the DDL strategy flag that allows foreign keys starts with "--unsafe-".

At this time I'm only able to see risks with all things removing-a-foreign-key. To that effect, I'm making this PR ready to review. We can iterate in the future in case we meet new scenarios.