Question about inserting the "Sort" node into the plan

[Taras]

Hi,

I am currently testing the last part of the assignment, and once again I
see some odd behavior sometimes. I insert the Sort node after Sequential
scan based on the pathkeys (index scan, tid scan, unordered sequential
scan are all deleted for this), and when the optimizer chooses to do a
merge join on the two relations, it does not seem to recognize that the
single relation plans already produce tuples in needed order - so I see
a Sort node inserted twice in a row in the plan. Does anyone have any
clue to what might be wrong here?

Thanks,
Taras

[Ken Salem]

Not sure what's going on there. Might you be mistakenly inserting
a sort even for single-relation paths that are not ordered, e.g.,
the original single-relation paths that were generated by postgres?

That could cause the symptom you are describing.

-KMS

Post by Taras
Hi,
I am currently testing the last part of the assignment, and once again I
see some odd behavior sometimes. I insert the Sort node after Sequential
scan based on the pathkeys (index scan, tid scan, unordered sequential
scan are all deleted for this), and when the optimizer chooses to do a
merge join on the two relations, it does not seem to recognize that the
single relation plans already produce tuples in needed order - so I see
a Sort node inserted twice in a row in the plan. Does anyone have any
clue to what might be wrong here?
Thanks,
Taras

[Taras]

I disabled in the code creation of the original single-relation paths.
So for the single relations, the only paths that are created are the
ones with interesting orders.

Taras

Post by Ken Salem
Not sure what's going on there. Might you be mistakenly inserting
a sort even for single-relation paths that are not ordered, e.g.,
the original single-relation paths that were generated by postgres?
That could cause the symptom you are describing.
-KMS

Post by Taras
Hi,
I am currently testing the last part of the assignment, and once again I
see some odd behavior sometimes. I insert the Sort node after Sequential
scan based on the pathkeys (index scan, tid scan, unordered sequential
scan are all deleted for this), and when the optimizer chooses to do a
merge join on the two relations, it does not seem to recognize that the
single relation plans already produce tuples in needed order - so I see
a Sort node inserted twice in a row in the plan. Does anyone have any
clue to what might be wrong here?
Thanks,
Taras