CODEWITHSUNDEEP
postgresql
Justin Leo
Justin Leo

Reputation: 2245

UUID or SEQUENCE for primary key?

I am coming from MySQL, and in MySQL you can use AUTOINCREMENT for a row's unique id as the primary key.

I find that there is no AUTOINCREMENT in Postgresql, only SEQUENCE or UUID.I have read somewhere that we can use UUID as the primary key of a table. This has the added advantage of masking other user's id (as I want to build APIs that take the ID in as a parameter). Which should I use for Postgresql?

Upvotes: 125

Views: 176451

Answers (6)

Rahul Tripathi
Rahul Tripathi

Reputation: 172628

You can use UUID as primary key in your table as it will be unique. However do keep in mind that UUID will occupy a bit more space as compared to SEQUENCE. And also they are not very fast. But yes they are for sure unique and hence you are guaranteed to get a consistent data.

You can also refer:

Upvotes: 30

bbsimonbb
bbsimonbb

Reputation: 29022

Other answers discuss the potential impact of UUIDs on index size and join performance. There's lots more discussion here. If you go for UUIDs you really need to understand this stuff to rule out potential problems.

A very good option is to stick with sequential identifiers, then use squids to convert them to and from short character strings. Your user count and database volumes are hidden from the casual observer, and you still enjoy the benefits and simplicity of sequential numeric identifiers.

Upvotes: 1

neel4soft
neel4soft

Reputation: 537

If case you are storing UUID in String type use this query to convert column type.

ALTER TABLE TABLE_NAME
    ALTER COLUMN ID TYPE uuid USING ID::uuid;

Next use the below query to make UUID column as auto-increment.

ALTER TABLE TABLE_NAME ALTER COLUMN ID SET DEFAULT gen_random_uuid();

If this still doesn't work, you may need to do the following as well:

CREATE EXTENSION pgcrypto;

Upvotes: 2

Alleo
Alleo

Reputation: 8548

Other answers mention performance issues of UUIDs without much details or references (quote: 'they are not very fast').

I've found one synthetic test which measured UUID joins to be ~5% slower that int8, and index size ~40% higher (not twice, as one can expect from other answers, as index size is not the same as size of all elements).

Upvotes: 11

Patrick
Patrick

Reputation: 32384

A sequence in PostgreSQL does exactly the same as AUTOINCREMENT in MySQL. A sequence is more efficient than a uuid because it is 8 bytes instead of 16 for the uuid. You can use a uuid as a primary key, just like most any other data type.

However, I don't see how this relates to the masking of a user ID. If you want to mask the ID of a certain user from other users, you should carefully manage the table privileges and/or hash the ID using - for instance - md5().

If you want to protect a table with user data from snooping hackers that are trying to guess other IDs, then the uuid type is an excellent choice. Package uuid-ossp has several flavours. The version 4 is then the best choice as it has 122 random bits (the other 6 are used for identification of the version). You can create a primary key like this:

id uuid PRIMARY KEY DEFAULT uuid_generate_v4()

and then you will never have to worry about it anymore.

PostgreSQL 13+

You can now use the built-in function gen_random_uuid() to get a version 4 random UUID.

Upvotes: 182

Clayton Bonelli
Clayton Bonelli

Reputation: 471

For many years I developed applications for databases using PKs and FKs as numerical sequential values. This has worked perfectly, but in recent years when creating cloud applications where information will be exchanged between applications and we will have integrations between various applications developed by us, we realized that the use of sequential IDs in our APIs ended up creating an effort.

In some applications we have to find the ID (of the target application) to be sent via the API call, on the other hand our database tables, in all our applications have, in addition to the sequential PK / FK column, a UUID column, which was not used in API calls. In this scenario we decided to rewrite the APIs so that the UUID column was used.

This solved some of the problems because one of our desktop applications would have their data migrated to another cloud application, this cloud application also used PK / FK columns. When migrating this data we had to change the values ​​of the PKs / FKs for new sequences as the sequences could clash between the values ​​of the desktop application and the values ​​of the cloud application. With this in mind we chose to switch cloud application PKs / FKs to UUID, since data coming from the desktop application had a UUID column.

The problem then was to convert the cloud application tables by turning the INT columns (PKs and FKs) into UUID columns without losing the table information. That was a big task, but it was made easier because I ended up building an application that makes this change easer. The application changes every PK / FK integer column to UUID, keeping the data and relationships. Anyone interested follows the link:

https://claytonbonelli.github.io/int_pk2uuid_pk/

Upvotes: 37

Related Questions