Red Brick may produce erroneous results, data corruptions, assertions and internal errors when running Red Brick 6.30.FC1 through 6.30FC11 versions on systems with multiple POWER5 and subsequent CPUs in the POWER series.
When running Red Brick on POWER5, POWER6 or POWER7 CPUs wrong results, data corruption, assertions, internal errors and hung loads may result. These problems may occur sporadically and with differing probability of occurrence as described below:
|Parallel queries produce wrong results, assertions or internal errors|
|Parallel aggregation maintenance produce corrupt data, assertions or internal errors|
|DML (insert, update, delete), DDL (alter, clear) in parallel queries or aggregate maintenance produce corrupt data, assertions or internal errors|
|Load or Reorg with aggregate maintenance produce corrupt data, assertions or internal errors|
|Parallel Reorg produces corrupt tables and indexes, incorrect number of duplicates, or wrong discards|
|Parallel Reorg hangs|
|Load or Reorg with aggregate maintenance generate invalid aggregates|
|Errors on non-versioned loads produce corrupt tables and indices|
|Load or restore from standard label tape produces corrupt tables and indexes|
Customer systems with ALL of the following may experience this problem:
- POWER5, POWER6 or POWER7 CPUs
- Multiple CPUs
- Red Brick version 6.30.FC1, 6.30.FC2, 6.30.FC3, 6.30.FC4, 6.30.FC5, 6.30.FC6, 6.30.FC7, 6.30.FC8, 6.30.FC9, 6.30.FC10 or 6.30.FC11
- Enabled parallel processing in query and reorg
If you have:
6.30.FC1, 6.30.FC2, 6.30.FC3, 6.30.FC4, 6.30.FC5, 6.30.FC6, 6.30.FC7, 6.30.FC8, 6.30.FC9, 6.30.FC10 or 6.30.FC11
6.30.FC11.R1 and later
Turn parallelism OFF; do this by setting QUERYPROCS as described below:
- Open the rbw.config file in a text editor
- Uncomment and set the TUNE QUERYPROCS value to 0
- Save the rbw.config file
- Stop and start Red Brick.
All subsequent connections will now use the updated value.
For problems with the loader
Use serial loader ( rb_tmu) to perform reorg operations.