Some private discussions to be synced

[wey-gu]

I will append my private discussions here to potentially help others in the NebulaGraph Community.

[wey-gu]

The question in raw Chinese, from @abasi20

1.execution plan 类 vs planner的关系（个人理解相同）
所有的 planner 都需要注册一遍
registerDDL（动态库）
registerSequential（自有的nGQL）
registerMatch（cypher match ）
使用的时候直接 clone？

Answer:
@wey-gu
After the plan being generated from the AST, then optimized by the optimizer(to rewrite the planNode DAG, when executor is going to execute it, each node in the plan will be mapped to an executor, and it will call things in Executor.cpp in order: create-open-execute-close:

• https://nebula-graph.com.cn/posts/nebula-graph-source-code-reading-03/ and 04 and 05
• https://nebula-graph.io/posts/nebula-graph-source-code-reading-03/ and 04 and 05

2. object pool的作用，为啥要考虑这样设计，生成对象不可以吗

Answer:
@wey-gu

We indeed implemented ObjectPool in common and used it everywhere, I think it's a practice to help handle object easier? (on mem/object allocation and destruction perspective?), let me ping @dutor for this to correct me ❤.

3. 将 subplan 与 matchClausePlan 连接（SegmentsConnector ）

调用接口类SegmentsConnector 连接 plan （cartesianProduct 笛卡尔积）
操作的意义是啥？

Answer:

@wey-gu

Yes, SegmentsConnector is used to connect sub plans.

CartesianProduct is introduced for multiple matches like:
MATCH <PATTERN_FOO> WHERE <>, <PATTERN_BAR> WHERE <> RETURN

This is by nature a cartesian product to aggregate the RETURN.

4.ExecutionPlan::makePlanNodeDesc 作用是啥？

Answer:
@wey-gu

From my understanding, NodeDescription is at least used for explaining/profiling the plans for human to understand needed context of the execution node.

5.executor是什么时间 、怎样被组装成链，schedule是怎么去执行的？

Answer:

@wey-gu

The plan was initially created by the planner after being parsed with the query into AST by the parser and validated by the validator.

Then it will be optimized by optimizer(a bunch of rules).

Then in the scheduler, the plan node (in DAG) will be BFS traversed from root to generate executors(each planNode to one executor).

refer to blog post 05 as below:

• https://nebula-graph.com.cn/posts/nebula-graph-source-code-reading-03/ and 04 and 05
• https://nebula-graph.io/posts/nebula-graph-source-code-reading-03/ and 04 and 05

6.是否考虑过path的按照拓扑结构执行

@wey-gu
For now, the FIND PATH query without the WITH PROP clause was BFS on edges only(without getting vertex properties), by "based on topology" do you mean to cache or hold topology only info somewhere to accelerate path-seeking?