Re: [jira] Commented: (DERBY-106) HashJoinStrategy leads to java.lang.OutOfMemoryError

[Jeffrey Lichtman <sw...@rcn.com>]

>This would appear to indicate that the optimizer is rejecting a
>NestedLoopJoinStrategy because the table is too big (i.e. larger than the
>1MB default).

The logic to reject a join strategy that uses too much memory is in 
OptimizerImpl.costBasedCostOptimizable(). It gives an estimated number of 
rows per scan to optimizable.memoryUsage(), which gives an estimated memory 
usage per row and a row count to joinStrategy.memoryUsage(). The 
implementation of memoryUsage() in HashJoinStrategy returns the product of 
the two numbers, while the implementation in NestedLoopJoinStrategy returns 
a constant 0.0. So, the estimated memory usage for a nested loop is always 
zero, while for a hash join it's based on the width of the rows and the 
estimated number of rows.

Back in OptimizerImpl.costBasedCostOptimizable(), it rejects any join 
strategy that would use more memory than maxMemoryPerTable. Since this is 
always zero for nested loops, it doesn't ever reject a nested loop for this 
reason.


                        -        Jeff Lichtman
                                 swazoo@rcn.com
                                 Check out Swazoo Koolak's Web Jukebox at
                                 http://swazoo.com/