Class CrossJoinHelper
TableOperations.join(TABLE)
(referred to as simply join or "cross join") and a left outer join. The left outer join does not currently have any
user visible API.
When there are zero keys, the result table uses BitShiftingColumnSources for the columns derived from the
left table and a BitMaskingColumnSource for the columns on the right. The rowkey space has some number of
bits for shifting, the low order bits are directly translated to the right table's rowset; the high order bits are
shifted to the right and indexed into the left table. For example if the right table has a rowset of {0, 1, 7}; then
3 bits are required to address the right table and the remainder of the bits are used for the left table.
For a bucketed cross join, the number of RHS bits is determined by the size of the largest group. The LHS sources are
similarly shifted using a BitShiftingColumnSource, but instead of using the rowkey space of the right hand
side directly each group is flattened. The RHS ues a CrossJoinRightColumnSource to handle these flat indices.
So in the case where we had a group containing an rowset of {0, 1, 7} it only requires 2 bits (for 3 values) not 3
bits (to represent the key 7). When values are added or removed from the RHS, the remaining values must appropriately
shift to maintain the flat space. If the largest right hand side group increases, then we must increase the number of
bits dedicated to the RHS and all of the groups require a shift.
The difference between a cross join and a left outer join is that in the case there are zero matching RHS rows, the cross join does not produce any output for that state. For the left outer join, a single row with null RHS values is produced. When a tick causes a transition from empty to non-empty or vice-versa the matched row is added or removed and the corresponding null row is removed or added in the downstream update (as opposed to being represented as a modification).
From a user-perspective, when the operation can be suitably performed using a TableOperations.naturalJoin(TABLE, java.lang.String), that
operation should be preferred. The LHS columns and RowSet are passed through unchanged in a naturalJoin and the right
columns have a simpler redirection. The simpler naturalJoin is likely to provide better performance, though can not
handle results that require multiple RHS rows.
-
Field Summary
Fields -
Method Summary
Modifier and TypeMethodDescriptionstatic TableleftOuterJoin(QueryTable leftTable, QueryTable rightTable, MatchPair[] columnsToMatch, MatchPair[] columnsToAdd, int numReserveRightBits)
-
Field Details
-
DEFAULT_NUM_RIGHT_BITS_TO_RESERVE
public static final int DEFAULT_NUM_RIGHT_BITS_TO_RESERVE
-
-
Method Details
-
leftOuterJoin
public static Table leftOuterJoin(QueryTable leftTable, QueryTable rightTable, MatchPair[] columnsToMatch, MatchPair[] columnsToAdd, int numReserveRightBits)
-