Explore using ULIDs as identifiers about go-kallax

[erizocosmico]

Using ULID would require us to use CHAR(26) or TEXT instead of UUID, which is almost twice the size as an uuid column.

So, even if we used an uuid as primary key and a created_at (which is a column that we will have anyway and most of the tables will already have an index there to search by that) as sorting index, the size of both columns would be smaller than a single ULID column (16+8 < 30)

select pg_column_size('01AN4Z07BY79KA1307SR9X4MV3'::text) as text_size,  
pg_column_size('110ec58a-a0f2-4ac4-8393-c866d813b8d1'::uuid) as uuid_size, 
pg_column_size('01AN4Z07BY79KA1307SR9X4MV3'::char(26)) as char_size, 
pg_column_size(current_timestamp::timestamptz) as timestamp_size;

 text_size | uuid_size | char_size | timestamp_size 
-----------+-----------+-----------+----------------
        30 |        16 |        30 |              8

I can do some research to check which option would have bigger indexes, but my guess is that it will probably still be smaller.

[joelvh]

@erizocosmico curious what you did here. We are starting to use PostgreSQL UUID and are looking to store ULID in there. What we do is decode the ULID to integer, then integer to hex, and then we pad it with zeros (usually one zero) to get to the proper UUID length.

What are your thoughts or what did you do?