overflow_error: robin_hood::map overflow

[nebula-bots]

Your Environments (required)
nebula-graphd version 3.3.0-ent, Git: bc87042, Build Time: Oct 8 2022 20:40:03
This source code is licensed under Apache 2.0 License.

How To Reproduce(required)

Steps to reproduce the behavior:

(root@nebula) [sf1]> use sf1
Execution succeeded (time spent 874/444017 us)

Mon, 10 Oct 2022 15:19:10 CST

(root@nebula) [sf1]> profile {GO FROM 32985348833536,24189255813121 OVER KNOWS  REVERSELY  YIELD DISTINCT  properties(edge)}
[ERROR (-1005)]: std::overflow_error: robin_hood::map overflow

Mon, 10 Oct 2022 15:19:12 CST

Expected behavior

ngql execution success

[jievince]

The dedup executor uses robin_hood to deduplicate data.
The readme of robin_hood says:

Depends on good Hashing. For a really bad hash the performance will not only degrade like in std::unordered_map, the map will simply fail with an std::overflow_error

In the case YIELD DISTINCT properties(edge), properties(edge) returns a MAP, which has a bad hash method for this case:

namespace std {
std::size_t hash<nebula::Map>::operator()(const nebula::Map& m) const noexcept {
  size_t seed = 0;
  for (auto& v : m.kvs) {
    seed ^= hash<std::string>()(v.first) + 0x9e3779b9 + (seed << 6) + (seed >> 2);
  }
  return seed;
}

Only the key of MAP particates in hashing.
While the MAPs returned by properties(edge) have the same keys, which leads to the same hashing result.