很少条件过滤Apache Spark

Question

I need to a check a few conditions, so I filtered my RDD this way:

scala> file.filter(r => r(38)=="0").filter(r => r(2)=="0").filter(r => r(3)=="0").count

Is it correct as an alternative of "&&"?

Answer 1

Yes, a series of filters is semantically equivalent to one filter with && in your case.

file.filter(r => r(38) == "0" && r(2) == "0" && r(3) == "0")

However, the variant above is guaranteed to be faster than the earlier version. This can be established via the following:

1. && is a short circuit operator, and the next comparison happens only if the first one evaluates to true . The number of comparisons in both the cases will be the same (yes!).

2. The multiple filter version involves three passes over the RDD vs. one pass for a single filter with && .