MapReduce to replicate self join - mapreduce

Question

In the traditional db way, I can do joins and find say, a list of users who visited 'pageA' but not 'pageB'.
Heres how I'm doing:
Table Schema:
t_user_actions {
user_id,
action,
page
}
Sample Data:
user_id, action, page
111, visit, pageX
222, visit, pageA
222, visit, pageB
333, visit, pageA
I can write this SQL to find list of all users who visited pageA but not pageB:
SELECT distinct u1.user_id user_id
FROM t_user_actions u1 left join t_user_actions u2 USING (user_id)
WHERE u1.page="pageA" and u2.page="pageB" and
u2.user_id is NULL
How do I achieve the same with MapReduce if I'm working on a large data set assuming I can import/insert the raw data into some NOSQL db?
I notice there are ways to do union, intersect but I'm trying to figure out how to do relative complement or difference in tuples.

Answer 1

Depending on which database you are actually using there might be much better ways to do this than MapReduce. But you asked specifically for MapReduce, so...
The Map phase would check all documents for action == "visit" && (page == "pageA" || page == "pageB"). When this is true, it would emit a document with the user_id as key and page as value.
The Reduce phase would iterate all values it receives per user. When there is at least one value with "pageB" it returns "pageB", otherwise it returns "pageA".
When you examine the result-set, ignore all returned values with page == "pageB". Those user visited pageB at least once (but not necessarily pageA as well). Those with page == "pageA" are those who you are searching for: the users who visited only pageA but never pageB.