Exists with mutations execute on initator

[Fgrtue]

Changelog category (leave one):

• Bug Fix (user-visible misbehavior in an official stable release)

Changelog entry (a user-readable short description of the changes that goes into CHANGELOG.md):

The problem appears when we set a setting:

SET mutations_execute_subqueries_on_initiator = 1;

And run an ALTER query with a function exists with a scalar subquery inside. This causes the scalar subquery in exsits being evaluated to a number. Then, after the mutation is applied, we get either an error, or we get OK with mutation stored with a corrupt command, which makes table unloadable on the next server restart. The response of the server depends of the query. Test in the pull request shows two queries, that cause different errors, but a rooted in the same problem.

AST fuzzer report: https://s3.amazonaws.com/clickhouse-test-reports/json.html?PR=96790&sha=cbdee8ffcd02bf966e588b7d06f2e39718da5486&name_0=PR&name_1=AST%20fuzzer%20%28amd_debug%29

Continuation of #96985

[clickhouse-gh]

Workflow [PR], commit [e168e9e]

Summary: ❌

job_name	test_name	status	info	comment
BuzzHouse (arm_asan)		failure
	Logical error: Digest does not match (STID: 5007-4bc5)	FAIL	cidb, issue	ISSUE EXISTS

[Fgrtue]

This bug can also crash the server in some cases such as the following:

https://fiddle.clickhouse.com/99e7405f-a80c-48b3-919e-b514f2a86ec9

[yakov-olkhovskiy]

strange thing is that I'm getting "Unknown function exists" in the fiddle:
https://fiddle.clickhouse.com/1178e949-a3ae-43e8-ae0c-181e57014589
maybe it's a manifestation of this bug in release build?

[Fgrtue]

Good question, I suppose something changed which made us skip the bug. I don't know what exactly though. But with the fix it should work normally, since it is a perfectly valid query, right?

[yakov-olkhovskiy]

well, test is passing in CI and I downloaded both debug and release and it definitely works... but still strange...
this is a stacktrace from current head:
https://pastila.clickhouse.com/?0079b799/083dd9d85a2482d60a23bbbba7b2958c#hkwvO75/IPLUivGwi2T+5Q==GCM

[yakov-olkhovskiy]

This bug can also crash the server in some cases such as the following:

I see you test it on 25.6 and it doesn't reproduce since 25.7 - was there some fix addressing this issue?

[Fgrtue]

@yakov-olkhovskiy I am not sure what made the server not to crash in the later versions, but in the latest version, we can still notice an erroneous behavior with these types of queries:

ALTER TABLE t_mutations_nondeterministic (UPDATE v = now((exists((SELECT 1023)))) WHERE NULL);

To see this behavior we have to run the test first, and then restart the server, connect with the client again, and the error should be shown automatically. It should also be shown if you try to DROP the table, and, probably, do anything else with it. This happens because the server tries to apply the latest mutation, which contains _CAST(1023, 'Nullable(UInt16)') (or something similar).

[Fgrtue]

Also, in debug build the server should still crash AFAIK