By its definition, GUID ( http://en.wikipedia.org/wiki/GUID ) is not
guaranteed to be unique. As a result, it could cause problems in
tables that have a large number of rows - actually this could result
in hard to reproduce, seemingly random errors in your systems that use
the database. How does a GUID help you over a BIGSERIAL?
It would help me in restoring archived projects. There is a quite
frequent scenario where a once archived (e.g. in a file) subset of the
db-contents will be restored. In that occasion I cannot be sure wether
the bigserial-values the entries has before are free. One can easily
imagine a system holding one project v(aka. subset of the content). Then
one archives that projects, deletes the db, resets it's schema and then
enteres two new projects. Afetr that, the once archived project is
restored from the archive. That would most likely fail as the sequences
will have given the same IDs (starting with 1) to elements of the other
two projects. I would need a handmade merge which would really be painfull.
Using GUIDs that use the MAC-adresses as well as the creation-time (as
in version 4 of uuids e.g.) would most likely (I see, there IS the
probability but I doubt it happing inside a common LAN, at least in the
life-time of the product) create distinct GUIDS that would reside
side-by-side without any trouble.
Unless you are generating the PK in multiple databases, a BIGSERIAL
should act as a UUID.
Maybe if you provide a little more detail on why you want to use GUID
or UUID fields we can help you come up with more options.
We're talking about a Client-Server-Application that would normally be
run on about 10 to 50 clients simultanously using the same backend.
Project-archivment will be frequent as single projects are open for
approximatly one to three month.
I'm well aware that as GUIDS are not really guaranteed to be unique
there could be scenarios where double ID's would occur. But I could
easily sense that before entering the project and then do this
praticular one by hand. I don't want to do this on every project that is
restored.
As a second subject I will need to extract subsets of the content for
external use in form of lists. Using the GUIDs of the entries would make
them be more distinct and thus no addidtional numbering would be
necessary for those lists. Currently I'm storing entries of different
type in different tables. To have them distinct on a combined list I
would need both the type and the id of a single row which would make up
16 Bytes if the type is bigserial as well (which it is in my scenario).
Thanks,
Aaron
Thanks for your quick answer.
Best regards
Dennis
