Re: UUID or auto-increment

Stephen Frost <sfrost@snowman.net>

From: Stephen Frost <sfrost@snowman.net>
To: Israel Brewster <ijbrewster@alaska.edu>
Cc: Ron <ronljohnsonjr@gmail.com>, pgsql-general@lists.postgresql.org
Date: 2020-08-10T17:16:29Z
Lists: pgsql-general
Greetings,

* Israel Brewster (ijbrewster@alaska.edu) wrote:
> > On Aug 10, 2020, at 8:53 AM, Stephen Frost <sfrost@snowman.net> wrote:
> > * Ron (ronljohnsonjr@gmail.com) wrote:
> >> On 8/10/20 11:38 AM, Ravi Krishna wrote:
> >>> Finally UUID results in write amplication in wal logs.  Keep that in mind
> >>> if your app does lot of writes.
> >> 
> >> Because UUID is 32 bytes, while SERIAL is 4 bytes?
> > 
> > and because it's random and so will touch a lot more pages when you're
> > using it...
> 
> I would point out, however, that using a V1 UUID rather than a V4 can help with this as it is sequential, not random (based on MAC address and timestamp + random). There is a trade off, of course, as with V1 if two writes occur on the same computer at the exact same millisecond, there is a very very small chance of generating conflicting UUID’s (see https://www.sohamkamani.com/blog/2016/10/05/uuid1-vs-uuid4/ <https://www.sohamkamani.com/blog/2016/10/05/uuid1-vs-uuid4/>). As there is still a random component, however, this seems quite unlikely.

Sure, that helps, but it's still not great, and they're still much, much
larger than you'd ever need for an identifier inside of a given system,
so best to map it to something reasonable and avoid them as much as
possible.

Thanks,

Stephen