Pseudo Perfect Hashing of INT64 -> INT64 to distribute autoincrement values

[ddorian43]

Hi,

I'm not really sure how possible it would be, but a kinda perfect hash function I think should exist that translated autoincrement values to pseudo random so they get distributed into different ranges when primary key is `(id autoincrement)` ? I mostly care the `id` to be small (64bit integer) + unique + random distributed.

Another way would be to prefix the primary-key with (hash(id), id) but this will make primary keys larger.

Any ideas ?

Regards,
ddorian43

[Ewan Chou]

Hi, 

We support a `shard_rowid_bits` syntax that shards an implicit row id.

See https://github.com/pingcap/docs/blob/master/sql/tidb-specific.md#shard_row_id_bits 

It's possible to make the explicit auto-id sharded, but we don't support it yet.