Normalizing Online-DDL queries

[shlomi-noach]

Backport

NO

Status

REAFY

Description

In this PR we normalize some DDL queries sent as Online DDL:

• CREATE INDEX formats onto ALTER TABLE: this is because gh-ost and pt-osc only know how to handle ALTER TABLE statements. As there is no specific advantage to using CREATE INDEX over the equivalent CREATE TABLE... ADD INDEX we choose to have CreateIndex.Format() function export as ALTER TABLE
• DROP TABLE <multi> - exploded into multiple single-table DROP statements

Related Issue(s)

#6926 (comment)

Todos

• Tests
• Documentation

Impacted Areas in Vitess

List general components of the application that this PR will affect:

• Query Serving
• VReplication
• Cluster Management
• Build

Notes

@GuptaManan100 @systay please notice the following test from parse_test.go:

		input: "create index a on b (col1) using btree key_block_size=12 with parser 'a' comment 'string' algorithm inplace lock none",
		// This is actually an incorrect output: should be the one commented
		output: "alter table b add index a (col1) using btree key_block_size 12 with parser 'a' comment 'string' algorithm inplace lock none",
		// output: "alter table b add index a using btree (col1) key_block_size=12 with parser 'a' comment 'string' algorithm inplace lock none",

The output is incorrect; still to understand why, I think the problem is with the parser, not with the Format. FWIW MySQL itself is unhappy about that complex CREAT EINDEX query.

[systay]

nit:

for query, expect := range tests {
	t.Run(query, func(t *testing.T) {
...
	})
}

sub tests makes it look good in test output

[shlomi-noach]

TIL actually

[shlomi-noach]

fixed

[shlomi-noach]

Still have some tests to fix.

[GuptaManan100]

LGTM 👍

[shlomi-noach]

Documentation: vitessio/website#638