autoid: support system variables auto_increment_increment and auto_increment_offset

[bb7133]

Feature Request

Is your feature request related to a problem? Please describe:
The system variable auto_increment_increment and auto_increment_offset are useful in some scenarios. For example, when the users want to do a Master-Master(bidirectional) replication with auto_increment columns, duplicate-key conflict and be avoided by setting different variable values in the clusters, for example:

Cluster 1:

mysql> create table t(a int key auto_increment);
mysql> set @@auto_increment_offset = 1
mysql> set @@auto_increment_increment = 2
mysql> insert into t values (...)
# the sequence of a will be 1, 3, 5, 7...

Cluster 2:

mysql> create table t(a int key auto_increment);
mysql> set @@auto_increment_offset = 2
mysql> set @@auto_increment_increment = 2
mysql> insert into t values (...)
# the sequence of a will be 2, 4, 6, 8...

For now, TiDB defines those 2 variables: users can set them but they don't actually take effect. We should make them work for those users who need them.

Describe alternatives you've considered:
The main implementation in MySQL Server can be found here:

https://github.com/mysql/mysql-server/blob/91a17cedb1ee880fe7915fb14cfd74c04e8d6588/sql/handler.cc#L3431

Basically, it made a rounding and then calculate next ID from the following formula:

nr = nr * auto_increment_increment + auto_increment_offset

The result can be guaranteed to be a multiple of auto_increment_increment offset with auto_increment_offset.

We can port this formula to TiDB's ID allocator:

tidb/meta/autoid/autoid.go

Line 312 in 0d7edc7

func (alloc *allocator) Alloc(tableID int64, n uint64) (int64, int64, error) {

Somethings we may need to consider:

• The formula applies to the cached values only. The allocation of the next step from KV is not affected.
• The allocation of rowid should not be affected.
• The rebase behavior from explicit value insertion or ALTER TABLE ... AUTO_INCREMENT should not break the constraint of these 2 variables: looks it should be fine, the next value is calculated instantly by the formula above.

Some manual tests have been done from a raw prototype: bb7133@adb83e3.

Teachability, Documentation, Adoption, Migration Strategy:

This is simply missing compatibility feature, seems we don't need to care them.