match p=(:foo)-->() return p returns empty instead of rejecting error on not being able to pushdown

[wey-gu]

Could we pushdown limit in this pattern now? If not, it should be rejected, right?

In my own test on v3.0.0, it's accepted and resulting in BAD_TYPE.

(root@nebula) [amundsen]> match p=(:`User`)-->() return p limit 3
+----------+
| p        |
+----------+
| BAD_TYPE |
| BAD_TYPE |
| BAD_TYPE |
+----------+
Got 3 rows (time spent 7657/54521 us)

Thu, 10 Mar 2022 11:08:12 CST

(root@nebula) [amundsen]> match p=(n:`User`)-->() return p limit 3
[ERROR (-1005)]: Scan vertices or edges need to specify a limit number, or limit number can not push down.

Thu, 10 Mar 2022 11:08:28 CST

(root@nebula) [amundsen]> show hosts graph
+-------------+------+----------+---------+--------------+---------+
| Host        | Port | Status   | Role    | Git Info Sha | Version |
+-------------+------+----------+---------+--------------+---------+
| "127.0.0.1" | 9669 | "ONLINE" | "GRAPH" | "02b2091"    | "3.0.0" |
+-------------+------+----------+---------+--------------+---------+
Got 1 rows (time spent 1647/48999 us)

It's reported from below thread as accepted & with empty result:

ref: https://discuss.nebula-graph.com.cn/t/topic/7842/10

[wey-gu]

labeled as bug for now, please kindly remove it if it's not the case ;)
Thanks!

[wey-gu]

I am working on this.

[CPWstatic]

I am working on this.

Looking forward to your contribution.

[wey-gu]

It turned out:

a. p=( :foo)-->() return p resulted in Limit->AppendVertices->Project->ScanEdges ==> Limit->AppendVertices->Project->ScanEdges(Limit)

b. p=(n:foo)-->() return p resulted in Limit->AppendVertices->Project->Traverse->ScanVertices ==> cannot be pushed down

I think:

- a. should be corrected as cannot be pushed down, right? on which phase should I fix please, then? I guess the existing plan is good(to start from scanEdges rather than scanVertices), so should I fix it in ScanEdgesExecutor::scanEdges to check condition on qctx to return Error or?

update: after further look, it's found the a. was due to the opt rule: GetEdgesTransformRule, now the issues are:

1. fixing the bad type, with minimal reproduce:

(root@nebula) [basketballplayer]> match p=()-[e]->() return p limit 1
+----------+
| p        |
+----------+
| BAD_TYPE |
+----------+
Got 1 rows (time spent 6635034/6633859 us)

Wed, 16 Mar 2022 06:53:49 UTC

(root@nebula) [basketballplayer]> match ()-[e]->() return e limit 1
+-----------------------------------------------------------------------+
| e                                                                     |
+-----------------------------------------------------------------------+
| [:serve "player102"->"team203" @0 {end_year: 2015, start_year: 2006}] |
+-----------------------------------------------------------------------+
Got 1 rows (time spent 3404514/3404425 us)

Wed, 16 Mar 2022 06:54:06 UTC

2. fix for p=(:foo)-->() return p limit 1, which should be applied to GetEdgesTransformRule? as foo edge-src tag type cannot be pushdown.

update: after talking with @Shylock-Hg (thanks!), I came to know a. should be excluded from the opt-rule: GetEdgesTransformRule, working on it now.

@CPWstatic @Shylock-Hg

Thanks!

[wey-gu]

#4053