The expert answer is (always) that it depends.
Whence the UUID? (Mladen, as was reasonable on oracle-l presumed sys_guid()),
but you may be using something local to the client.
Is there a constraint on the database ensuring the UUID is in fact unique?
If you have a UUID generator you trust on the client that might be a slight
advantage over a good sequence+ range for your planned cache of sequence values
with one fetch for the client preparing a slug of rows to insert.
Sequence is very likely to be more compact and if you are tracking integrity of
insert slugs or auditing after the fact, contiguous sequence values can be
useful in puking out the transaction details.
If you’re reserving a range of sequence value per inserting client, you tend to
minimize the likelihood of a hot spot on any indexes including the sequence
since the likely numbers of inserters into the same index leaf are 1 and 2 (if
clients with adjacent reserved ranges insert at the same time they might easily
hit the same leaf, and it could be more with small reserved ranges and small
batches).
Time plus sequence can often be useful, but that is not generic and only
applies when ordered “born on” date for rows inserted is useful (very often
that is true for business data, less often so for spectrographic analysis and
maps).
Others can probably wax on for additional case analysis parameters of which
might be a better prediction of better behavior without an actual test.
But it depends. I tend to use sequence (or time plus sequence) unless something
about an actual case in hand screams “a sequence will be a problem.”
A test of meaningful scale is the only way to be sure for a given case. When
computers were a lot slower actual tests of doing things two or more different
ways took place. Now we mostly look for a forecast and re-tool in the event the
horse we have chosen stumbles.
Good luck and let me know if you find and expert (especially one who claims the
answer is invariant and not “it depends.”)
mwf
From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx] On ;
Behalf Of ahmed.fikri@xxxxxxxxxxx
Sent: Thursday, April 11, 2024 10:35 AM
To: peter.m.gram@xxxxxxxxx
Cc: list, oracle
Subject: AW: Re: UUID vs. Sequential ID as Primary
Thanks for the hint, you’re right, one can use that. The problem is that when
working with hibernate (or any other client lib) one doesn’t want to go to the
database to fetch the IDs when creating an entry on the client side. While one
does want to create the ID uniquely, it should be done with minimum visiting
the database (using sequences has now been optimally solved e.g. hilo
algorithm, not just for Oracle, but almost all other databases).
I’m interested in our experts have to say about this: is the trend of using
UUID legitimate, or is using numbers better?
Gesendet mit der Telekom Mail App
<http://www.t-online.de/service/redir/emailmobilapp_ios_smartphone_footerlink.htm>
-----Original-Nachricht-----
Von: Peter Gram <peter.m.gram@xxxxxxxxx>
Betreff: Re: UUID vs. Sequential ID as Primary
Datum: 11.04.2024, 16:10 Uhr
An: <ahmed.fikri@xxxxxxxxxxx>
CC: list, oracle <oracle-l@xxxxxxxxxxxxx>
Hi
If you use the returning clause on the first statement you don’t get a extra
round trip to the database to get the sequence.
Med venlig hilsen
Peter Gram
Sæbyholmsvej 18
2500 Valby
Mobile: (+45) 5374 7107
Email: peter.m.gram@xxxxxxxxx
On Wed, 10 Apr 2024 at 23.38, ahmed.fikri@xxxxxxxxxxx <
ahmed.fikri@xxxxxxxxxxx> wrote:
Hi there,
Once more, a question about fundamentals:
In several Hibernate projects, I've observed developers leaning towards using
UUIDs as primary keys instead of numerical values. This preference likely stems
from the avoidance of sequences for numerical primary keys, which necessitate
round trips to the database after each insertion. Additionally, there's a
concern about potential contention with sequences, and some developers may
prefer to avoid predictability in the next generated value. Personally, I
remain skeptical about the widespread use of UUIDs due to their larger storage
footprint (both in tables and indexes) compared to numerical IDs. Numeric IDs
also offer benefits in issue analysis. However, it's worth noting that
Hibernate employs algorithms to minimize round trips to the database,
effectively reducing their impact. Could you please share your experience and
preference? What choice would you make?
Regards
Ahmed
