Oracle 11g imp erroneously tries to recreate existing tables with CLOBs?

Question

I have a shell script for loading database dumps from both Datapump and the older exp/imp.
Often when loading dumps, I need to rename the schema owner and tablespace names (which is handled by REMAP_SCHEMA and REMAP_TABLESPACE in Datapump).
However I have a whole bunch of dumps created with exp at this point and not that many Datapump dumps yet. As such the old style dumps are handled by the shell script in this way:
1) A first pass imp is run using INDEXFILE to generate a file with the SQL to create tables and indexes. Options also include FROMUSER and TOUSER.
2) A series of sed command edit the SQL file to change the tablespace names (which are schema owner specific in our case).
3) The editted SQL file is run with sqlplus to create the tables and indexes.
4) A second pass imp is run to load the table rows as well as triggers, stored procedures, views, etc. Options include FROMUSER, TOUSER, COMMIT=Y, IGNORE=Y, BUFFER, STATISTICS=NONE, CONSTRAINTS=N
This shell script has been working great for loading exp dump files into Oracle 9 and Oracle 10 databases, but now that I'm trying to load these dumps into Oracle 11, it fails.
The problem is in step 4, the imp program is trying to create some of the tables that already were created with sqlplus in step 2. The problematic tables all seem to have CLOB columns in them. The table creation fails because it tries to use the tablespace names from the dump file, which do not exist in the destination database. And when the table creation fails, imp then decides not to load the rows for those table.
This seems like a bug in the Oracle 11 imp program. I don't understand why it thinks it needs to recreate tables that already exist when those tables have CLOB columns. Is there something different about CLOB columns in Oracle 11 that I should know about that might be confusing imp into thinking that it needs to create tables when they already exist? Maybe I need to do something to those tables in SQL so that imp does not think it needs to recreate them?
I know that the tables with the CLOBs were created correctly because I was trying to find some way to workaround this. For step 4, I tried using DATA_ONLY=Y, in which case imp does not try to create the tables and just loads the table rows. Of course using DATA_ONLY, I don't get a lot of other things like triggers, view and stored procedures. I started to try to get around that by doing 3 passes with imp, so that I could pick up the missing pieces by using an imp pass with ROWS=N, but strangely that has the same problem of trying to recreate the existing tables.

Answer 1

The only solution I've found so far as a workaround is rather convoluted.
1. I took an export using datapump's expdp of SCHEMA1 (in 10g it will skip the table with the xmltype).
2. I imported the data to my empty schema (SCHEMA2) using impdp. To avoid the error that the type already exists with another OID, I used the TRANSFORM=oid:n parameter e.g.
impdp user/pwd dumpfile=noxmltable.dmp logfile=importallbutxmltable.log remap_schema=SCHEMA1:SCHEMA2 TRANSFORM=oid:n directory=MYDUMPDIR
3. I then manually created my xmltype table in the SCHEMA2 and did a select into to load it (make sure you have the select privileges to do so):
INSERT INTO SCHEMA2.XMLTABLE2 SELECT * FROM SCHEMA1.XMLTABLE1;
4. I am still taking an export with exp of the xmltable as well even though I'm not sure I can do anything with it.
Thanks!
Edited by: stacyz on Jul 28, 2009 9:49 AM