Parallel Rollback Problem

AIX 4.3
Oracle 8.1.7.4

Twice in the last two days. I have encountered a problem with parallel rollback. We do not know what is triggering the problem but I suspect someone is killing a session either with ALTER SESSION or by killing a session's shadow process in unix. The end result is 4 parallel query slaves start doing parallel rollback as evidenced by v$fast_start_servers showing 4 rows with state = "RECOVERING". When I look at the sessions that are doing all the block changes in the database, they are the 4 PQ slaves and they tie back to SMON as the parent process. The problem is that they generate far more redo during this rollback than any transaction that takes place on the database. It almost appears as if they are looping and spitting out redo. Eventually the archived log filesystem fills and the database hangs. I have seen this scanario generate 9g of logs in 20 minutes.

After the 2nd occurence I tried to do an "Alter system set fast_start_parallel_rollback = false" per note 144332.1 but the command hung as did all subsequent attempts to connect to the database until I killed the session doing the "alter system".

Does anyone know of any bugs with parallel rollback that would be reponsible for this? Will they be fixed if I upgrade to 9.2?

-- 
Chuck
Remove "_nospam" to reply by email