 |
 |
Archives of the TeradataForum
Message Posted: Sat, 02 Jan 2010 @ 20:02:36 GMT
| Subj: | | Re: Row Hash Deadlock in Active D/W due to concurrent MSR |
| |
| From: | | Fred W Pluebell |
A true deadlock always involves at least two different locks - different object, different RowHash, and/or different type (e.g. Read vs.
Write).
Teradata acquires any table-level locks in special steps at the start of the request (most restrictive first), but RowHash locks are acquired
locally on the AMP "on demand" when a step requires it.
Do your MSRs contain single-AMP SELECT requests as well as UPDATE, INSERT, DELETE to the same tables? If so, is it feasible for your
application to upgrade RowHash locks to WRITE (LOCKING ROW FOR WRITE) on the SELECT?
Do your MSRs update multiple tables? If so, are they always modified in the same order (TableA first, TableB next, etc.)?
| |
