"Terris Linenbach" <noreply@nowhere.nwh> wrote in message
news:ucY1ipfjDHA.644@TK2MSFTNGP11.phx.gbl...
> Let's say all of my report queries use READ_UNCOMMITTED in order to avoid
> deadlocking with load processes. The queries select from views which have
> indexes.
>
> Now suppose a query is running while a nightly load operation is running.
> The nightly load operation inserts additional records into the database
> which causes the aggregates in the materialized views to be recalculated.
>
> The following statements are also true:
>
> (1) The query's where clause contains a statement sort of like "select
> sum(purchase_amount) where date <= 10/10/2002"
> (2) The view aggregates purchase_amount for each day
> (3) When the query is running, data is being inserted into the physical
> table upon which the view is built
>
> Case 1:
>
> The new data loaded applies to 10/20/2002
>
> Is it possible to read inconsistent aggregate data in this case? I
wouldn't
> expect it to, since the where clause filters out the data that is being
> loaded into the star. However, I have no idea how materialized views are
> actually implemented or how they have been specified to behave when
> READ_UNCOMMITTED has been specified.
>
> Case 2:
>
> The new data loaded applies to 10/5/2002
>
> In this case I would expect a dirty read to return inconsistent data. How
> "inconsistent" can the data be? Could it be totally random or would it be
> constrained between, say two values: the value before the insert
operation
> and the value consistent with the data in the underlying table?
>
> Does any of this make sense?
>
> Thanks,
> Terris
>
>