[master < T1085] Remove double scan with expand from the planner

[Josipmrden]

Possibly it is not needed to do double ScanAll in the operator tree but expand from already existing node.

@vpavicic An optimization to the query engine was added in order to prefer at all times scanning + expansion of nodes instead of scanning both source and destination vertices and then expanding to the edge between them.
No need for documentation updates.

[master < Task] PR

• Check, and update documentation if necessary
• Provide the full content or a guide for the final git message

To keep docs changelog up to date, one more thing to do:

• Write a release note here, including added/changed clauses
• Tag someone from docs team in the comments

[Ignition]

@Josipmrden Looking at Grafana dashboard it does drop throughput on some of the mgbench workloads. I think we need to understand this some more.

The expand.common_.existing_node looks like it should prevents the change in the plan to the double scan. Maybe we should check why that isn't there in our case.

CREATE INDEX ON :Post;

// PLAN before data inserted (shouldn't be double scan because of post)
EXPLAIN MATCH (post:Post)
WHERE NOT EXISTS( (:Post)<-[:ROOT]-(post))
RETURN post;

UNWIND RANGE(0,1000) AS i
CREATE (:Post),(:Post),(:Post),(r:Post)
CREATE (:Post)-[:ROOT]->(r);

// PLAN remains unchanged
EXPLAIN MATCH (post:Post)
WHERE NOT EXISTS( (:Post)<-[:ROOT]-(post))
RETURN post;

[gitbuda]

Seems like a good update, but not good results on the mgbench -> investigation needed

[Josipmrden]

@gitbuda After some talk with @Ignition we see there are no performance degradations on the benchmarks. We will proceed with the removal of the feature. Also, if the need occurs again, we will re-add the optimization with the appropriate benchmark tests that will point whether we have degraded any queries.