Parallel impdp threads stall on enq TM
This is an issue with parallel impdp on 10.2.0.4.
I am trying to a) reclaim space in a tablespace and b) replace composite partitioned tables with simple range partitioning.
To do this, I have exported the schema with parallel expdp creating 8 dump files. I then dropped and recreated the tablespace, re-created empty tables for the 25 tables that are being redefined and then began an import of all the tables in the schema using TABLE_EXISTS_ACTION=TRUNCATE. Indexes, constraints and statistics are excluded.
I have tested this procedure several times and sometiimes it works OK but on some occasions (like now) one of the import worker threads holds up all the others. All 8 worker threads report that they are trying to import the same table (one of the ones I explicitly created) but only one is actually importing data. The other threads are showing waits of several minutes at a time on an exclusive 'enq - tm' on the table that they are all trying to import. As each table is imported all 8 threads move onto the next table and the single threaded behaviour continues.
Does anybody know why this happens and how to avoid it?