Oracle数据库：访问多个数据库模式

Question

I have several Oracle database schemas: eg MainSchema, and other data schemas like dataSchema1, dataSchema2, dataSchema3

My requirement is to execute sql statements (stored in MainSchema tables) in one of the data schemas and write results to MainSchema result tables.

What I can think of is: Use Java Stored Procedure, which published in MainSchema, connect to data schema (I know which schema to connect for the sql statement) and execute sql, then write results to MainSchema result tables.

I want to know:

• This is a good approach to go? or any alternate is best practice?

• This is multi-threaded environment, meaning the number of connections may grow quickly. How I handle the connection pooling?

EDIT: Assumptions below:

1. sql statements are stored in a clob column in MainSchemas;
2. sql statements are only SELECT queries (read only) and already validated;
3. sql statements do not require any connection information (meaning written locally)
4. "schema" can be different database so prefixing schema name is not an option;

So I think it's hard to do with procedures/functions. Using Java implementation to access different databases may be a good choice?

Experts here please give some advice on how to design this?

Answer 1

I am not quite sure what you mean by sql statements (stored in MainSchema tables) . If you really mean tables with a string column holding SQL text, then this is really an odd thing to do.

A somewhat better approach is to create a bunch of stored procedures, which return a cursor. You would then call one of these procedures and fetch from the cursor and write back to to MainSchema's result tables. The benefit of this approach is that the SQL is checked during compilation, while SQL in Strings (no matter where you store them) is an error prone thing.

Operating on two schemas is actually easy: you need to prefix your tables (and other object) with the schama name. You can do things like

insert into mainSchema.resultTable (
select ... from dataSchema1.dataTable1
where ...
)

If you can write things like this, then .. well .. you just run it. You can also do this from Java if you must.

If however, the schema name is a variable , ie your SQL is not specific to a schema, but you decide which dataSchema to use at runtime, then this doesn't work. Then again having multiple schemas holding similar data is almost always a bad idea. You need to partition by meaning, not by physical things like origin or time-period. A good solution here is to use partitioned tables.

Edit:

If you're just worried about the number of connections, you may want to setup Oracle to use shared connections . This however requires control over the database which you may not have.

Connection pooling is a typical job of an application server. But that would be quite some overhead just for running some selects and it sometimes has unexpected consequences if you're not completely stateless.

If you need to connect to different database, then database links are the standard way to go. Instead of prefixing objects with the schema name, you'll need to add the name of the database link as in

insert into mainSchema.resultTable (
select ... from dataTable1@linkName
where ...
)

If your SQL is written without any particular DB in mind, then you can of course just connect to the DB and leave the SQL as it is. However, you will not be able to use insert-select . You need two DB connections, one where you execute the select and another where you execute the insert.

I believe your approach will work, though I don't see the point why you store SQL in a database and not in the application code. If it was in the application code you can just assemble the SQL adding @linkName where needed, you could use insert-select and you could run such an insert-select from the commandline for testing without having to do any compilations.

Answer 2

For me, Java Stored Procedures is a way to go.

http://docs.oracle.com/cd/B19306_01/java.102/b14187/chfive.htm