Archives of the TeradataForum

Message Posted: Wed, 29 Jan 2004 @ 04:03:12 GMT

	<Prev	Next>		<<First	<Prev	Next>	Last>>

This is a BAD problem. I opened DR82673 this summer to get it fixed. The DR was made into an RFC though its priority was set to 1.

It's not going to be fixed real quick because it's not easy to fix and the fix - when finally available - will need more testing than the testing we do for fixes on etapes. This is the way bug DRs become RFC DRs.

Until a fix is available I would strongly recommend that you put ALL join indexes in databases strictly reserved for join indexes and NEVER archive any of these databases. This will protect you from the worst problems - the kind you'll need help from the GSC to recover from.

Whenever you restore a table which is a party to a join index all join indexes that reference that table are marked invalid (this is one thing we do right, although the way we do it is not exactly intuitive). You will need to recreate the join indexes before the optimizer will consider using them again. We have to do this. If we didn't, we could return incorrect results due to stale data in the join index.

The problem is that we don't tell you this in a user friendly manner. What we do is mark an index of the table you restored as invalid. We do this in the table header. But ARC doesn't say a word. You have to run check table to have the invalid indexes reported.

Just to make life even more fun, the indexes that were marked invalid are not REALLY indexes. They're - at least conceptually - pointers to join indexes. We chose this less than straightforward implementation because we wanted to get join indexes out between major releases where we're not allowed to change dictionary tables. Because we could not change the dictionary tables, it's not possible to get info from the dictionary the way you should be able to. If you want to get a list of all join indexes referencing a given table, you'll have to resort to terms on CreateText using LIKE. Kinda kludgy, but it's the only way I know.

We WILL fix this. But not right away.

	<Prev	Next>		<<First	<Prev	Next>	Last>>