[ https://issues.apache.org/jira/browse/PHOENIX-1556?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16352053#comment-16352053 ]
Maryann Xue edited comment on PHOENIX-1556 at 2/5/18 8:31 AM: -------------------------------------------------------------- Could you please review the patch for me, [~jamestaylor]? The {{CostBasedDecisionIT#testJoinStrategyXXX()}} tests are to verify and demonstrate how join strategies are chosen in different scenarios. The calculation of join costs is designed to follow the guidelines below: # The build side is applied with more weight than the probe side in calculating the cost of a hash-join, so the smaller table is guaranteed to be the build side. # If the build side exceeds the size limit (QueryServices.MAX_SERVER_CACHE_SIZE_ATTRIB), the cost of hash-join is infinitely large. # The cost of sort-merge-join alone is guaranteed to be smaller than hash-join. But when sorting is required to do a sort-merge-join, the cost of the added "order-by" will be counted in the total cost and thus may make the sort-merge-join operation as a whole more expensive. # The {{QueryCompiler#compileJoinQuery()}} method will compare the costs of the plans from all applicable join strategies and chose a local optimal. As a result, with multiple joins (joins between more than two tables), a mix of join strategies may be chosen as the final plan. For example, (A hash-join B) sort-merge-join C. was (Author: maryannxue): Could you please review the patch for me, [~jamestaylor]? > Base hash versus sort merge join decision on cost > ------------------------------------------------- > > Key: PHOENIX-1556 > URL: https://issues.apache.org/jira/browse/PHOENIX-1556 > Project: Phoenix > Issue Type: Sub-task > Reporter: James Taylor > Assignee: Maryann Xue > Priority: Major > Labels: CostBasedOptimization > Attachments: PHOENIX-1556.patch > > > At compile time, we know how many guideposts (i.e. how many bytes) will be > scanned for the RHS table. We should, by default, base the decision of using > the hash-join verus many-to-many join on this information. > Another criteria (as we've seen in PHOENIX-4508) is whether or not the tables > being joined are already ordered by the join key. In that case, it's better > to always use the sort merge join. -- This message was sent by Atlassian JIRA (v7.6.3#76005)