Thanks guys for the help :)
Another option is to embed the domain id (assuming there is one) into each table's primary key (i.e compound primary key) and thus use 1 schema and 1 database, without the need to dynamically change table prefixes. regards Mark On 10/09/18 14:13, Evan Bauer wrote: Joao,
I strongly agree with Tim’s recommendation to create a schema for each “domain.” In addition to the reasons below, the prefix model would seem to condemn your programs and programmers to using dynamic SQL to construct your business logic in a manner that would work for any domain, as the table names for each SQL statement would vary based on the domain.
Schemas are a solid approach to multi-domain design, they should work well for you.
Cheers,
- Evan
Evan Bauer
eb@xxxxxxxxxxxxx <mailto:eb@xxxxxxxxxxxxx>
+1 646 641 2973
Skype: evanbauer
On Sep 9, 2018, at 18:02, Tim Cross <theophilusx@xxxxxxxxx <mailto:theophilusx@xxxxxxxxx>> wrote:
Joao Ribeiro <joao.ribeiro@xxxxxxxxxxxxx <mailto:joao.ribeiro@xxxxxxxxxxxxx>> writes:
Hello,
We are using Postgres and we are with a issue. We have splited our tables by domain and
each domain has a separated database, but we are trying to change it to a single database
model. We have two options, create the same database in a single schema and add a prefix
on each database domain or create a schema for each database (we have about 15
different models). In this approach we still want to do some joins and other queries cross
schema, but we don’t know what would be the best approach :) Could you help us to know
what would be the best approach ?
* having just one database with one schema and all domain databases with a prefix
* having a schema for each domain (15 domains) with the domain databases (+/ 20 tables)
(knowing that we do cross schema queries)
_____
It is probably just a matter of taste to some extent. I personally would
favour separate schemas over tables in same schema with prefixes because
I think that gives you more flexibility i.e. easier to select all the
data associated with a domain as it is all in one schema. I also think
it is easier to define security roles on a per schema basis rather than
complex roles in the same schema, especially if you add new objects
etc. Auditing is also less complex.
There are no issues with cross-schema queries/joins etc apart from
having to include the schema name in the query. Some people don't like
this because you have to type more and have a longer search_path, but if
the alternative is table prefixes, the amount of typing is similar
anyway.
Essentially, the schema gives you a predefined unit which many
tools/commands understand. If everything is in the same schema, then you
will often need to replicate some level of this functionality yourself
and then ensure it is maintained. Extent to which this has an impact
really depends on your use case.
Tim
--
Tim Cross
|
