schemadiff: support for RangeRotationStrategy diff hint

[shlomi-noach]

Description

Diffing partitions is a complex topic. There can be scenarios where a change of partitioning scheme required an entire rebuild of the table (e.g. change of number of HASH partitions), and cases where optimally one could get by with only modifying some of the partitions (e.g. COALESCE two partitions into one).

Normally, when we see a diff of partitions, we just invoke an ALTER TABLE... PARTITION BY... with the entire new partitioning scheme. The idea? Even a COALESCE for two partitions out of 200 is a blocking schema change. As we push forward Online DDL, we are opinionated that it's better to rebuild the entire table, online, than just 2 partitions out of 200, in a blocking operation.

However, there are exceptions to this.

A common partitioning scheme that does not require blocking: rotation of range partitioning. The classic example is, say, partitioning by day, such that every day we add a new partition; and possibly every day we also purge an old partition.
We could be adding a bunch of partitions at a time, or drop a bunch of partitions at a time.
Adding new partitions is super cheap, and dropping old partitions can be cheap or expensive, depending on the size of dropped partition and version of MySQL.

In this PR schemadiff identifies a rotating range partitioning scheme: two tables that are partitioned by RANGE, share exactly one consecutive non-empty sequence of partitions (i.e., the two tables have some partitions in common), and possibly:

• 1st table (aka "old" table) does not have some newer partitions (these have just been rotated in), and
• 2nd table (aka "new" table) does not have some older partitions (they have been rotated away)

As of this PR schemadiff supports a RangeRotationStrategy, of the following values:

• RangeRotationFullSpec: generate a full partitioning spec, meaning do not optimize for partition rotation; this behavior is the same as before this PR
• RangeRotationIgnore: consider this as a non-diff. This is similar in concept to ignoring an AUTO_INCREMENT increase.
• RangeRotationStatements NOT YET IMPLEMENTED. Will throw an error. Once implemented, will return the proper DROP PARTITION ... and ADD PARTITION ... statements that perform the rotation.

Checklist

• "Backport me!" label has been added if this change should be backported
• Tests were added or are not required
• Documentation was added or is not required

[shlomi-noach]

To support RangeRotationStatements we'd first need an AddPartition and DropPartition statements in AST, that implement AlterOption via iAlterOption().

[GuptaManan100]

It looks good to me! What might be worth noting is that, we are expanding the parsing support for partition definitions (#10127 ) along with sub partitions (soon to follow).

[shlomi-noach]

What might be worth noting is that, we are expanding the parsing support for partition definitions (#10127 ) along with sub partitions

Oh wonderful! Do you think we might be able to then have:

AddPartition and DropPartition statements in AST, that implement AlterOption via iAlterOption().

?

[GuptaManan100]

Oh wonderful! Do you think we might be able to then have:

AddPartition and DropPartition statements in AST, that implement AlterOption via iAlterOption().

?

Yes! Most definitely. We can work on that before subpartitions, since I see you need it for completing this work!

[shlomi-noach]

@GuptaManan100 thank you! There is no urgency for this work.

[GuptaManan100]

@shlomi-noach I checked, and apparently we already have parsing support for ADD PARTITION, DROP PARTITION, etc already. It was added way back in #7199. These don't implement the AlterOption interface and instead are specified in PartitionSpec in the AlterTable field. I don't remember exactly why I implemented it this way but my guess would be, we did this because there can be multiple alter options, but only 1 partition alter in a statement. This is written at the end of the doc here - https://dev.mysql.com/doc/refman/8.0/en/alter-table.html

[GuptaManan100]

With this we augmented the partition definition, which also added that additional parsing to ADD PARTITION so we should be good there too

[shlomi-noach]

@GuptaManan100 ahhh, that's right! Now I see it. Thank you.

[shlomi-noach]

Tracking issue: #10203