MicrosoftThere is a type of query that happens in Relativity eDiscovery. There is a table of say10M records. The query is to find matches in a subset (A) of 100K records but not in a subset (B) of 1M records. What is the probability that the 1% (100K out of 10M) of A is not in 10% (1M of 10M) in B. There is no general rule that will help estimate the net of in A but not in B. You could say B represent 10% of All so exclude 10% of A, or you could argue only a small percentage of A meets the criteria of not in B. Either could happen. Says this the part of the query that contributes to filtering, followed by additional joins.
I would hope that the resolution is to detect this, evaluate this part of the plan, then re-optimize for the remainder of the query, even if means rexecuting the first part of the query.
