Atlas Search query performance severely degrading (~20x) if score is boosted by path

Oliver_Haas1 · 2025-03-04T09:05:37.590Z

I have a query where I boost the total score of each document by a path

[
  {
    "$search": {
      "index": "exampleIndex",
      "compound": {
        "should": [
          {
            "text": {
              "query": ["exampleQueryTerm"],
              "path": ["examplePath"]

            }
          }
        ],
        "minimumShouldMatch": 1,
        "score": {
          "boost": {
            "path": "customBoostScore",
            "default": 0.1
          }
        }
      }
    }
  }
]

But boosting the score by a path seems to degrade performance by up to 20x, meaning a ~700ms second query now takes more than 10s, which is not acceptable for my use case.

Is this a known caveat of boosting the score by a path?

Since the I indexed customBoostScore (as number as required) and have store: true I definitely did not expect that.

Oliver_Haas1 · 2025-03-05T07:07:57.976Z

I should have mentioned that my collection is about 200M documents, and the search index is about 100GB.

I’m not totally surprised that it isn’t super fast overall, but the difference just by using the boost by path is so big for me… Either I’m missing something, or it’s quite a big caveat.