Incorrect behaviour of fuzzing parameters with completion suggester

[pigeongcc]

Elasticsearch Version

8.6.2, 8.8.2

Installed Plugins

jmorphy2 (https://github.com/anti-social/jmorphy2), LTR plugin (https://elasticsearch-learning-to-rank.readthedocs.io/en/latest/)

Java Version

openjdk version "19.0.2" 2023-01-17 // OpenJDK Runtime Environment (build 19.0.2+7-44) // OpenJDK 64-Bit Server VM (build 19.0.2+7-44, mixed mode, sharing)

OS Version

Linux es-search-v2-es-data-1 5.4.0-167-generic #184-Ubuntu SMP Tue Oct 31 09:21:49 UTC 2023 x86_64 x86_64 x86_64 GNU/Linux

Problem Description

I use fuzzing in my quries to completion suggester. It seems like min_length and fuzziness parameters of the query can't agree between each other.

The parameters are described in the docs

Example 1: "min_length": 5, "fuzziness": "AUTO:5,8"

query: conte-
expected: query length is 6, we're expecting 1 correction (dash to be removed)
result: contemporary is suggested, everything's ok ✅

query: contempo--
expected: query length is 10, we're expecting 2 corrections
result: nothing is suggested, incorrect behaviour ❌

Example 2: "min_length": 1, "fuzziness": "AUTO:2,5"

query: co-
expected: query length is 3, we're expecting 1 correction
result: contemporary and cheese are suggested, we have 2 corrections in the cheese case ❌

query: contempo--
expected: query length is 10, we're expecting 2 corrections
result: contemporary is suggested, everything's ok ✅

The bug is reproducible with ES 8.6.2 and 8.8.2.

Steps to Reproduce

Complete Python code to reproduce the bug. Copy and run.

The code is split into .ipynb cells.

import elasticsearch
import json
import pandas as pd
import os

from tqdm import tqdm
from typing import List
from elasticsearch import Elasticsearch, helpers


es = Elasticsearch(
    hosts="127.0.0.1:9200",
    http_auth=({login}, {password}),
    timeout=100
)

es.indices.get_alias("*")

INDEX_NAME = 'text-suggestions-bug-check-1'
ALIAS_NAME = 'text-suggestions-bug-check'

# es.indices.delete(index=INDEX_NAME)
# es.indices.delete(index=ALIAS_NAME)

# Analyzers and mappings
SETTINGS = {
    "settings": {
        "index": {
            "number_of_replicas": 1,
            "number_of_shards": 3,
            "refresh_interval": "-1"
        },
        "analysis": {
            "filter": {
                "synonym_filter": {
                    "type": "synonym",
                    "synonyms": [
                    ]
                },
                "autocomplete_filter": {
                    "type": "edge_ngram",
                    "min_gram": 1,
                    "max_gram": 20
                },
            },
            "tokenizer": {
                "cg": {
                    "type": "char_group",
                    "tokenize_on_chars": [
                        "whitespace",
                        ".",
                        ",",
                        ":",
                        ";",
                        "/"
                    ]
                }
            },
            "analyzer": {
                "autocomplete_analyzer": {
                    "type": "custom",
                    "tokenizer": "cg",
                    "filter": [
                        "lowercase",
                        "synonym_filter",
                        "remove_duplicates",
                        "autocomplete_filter"
                    ]
                },
                "autocomplete_search_analyzer": {
                    "type": "custom",
                    "tokenizer": "cg",
                    "filter": [
                        "lowercase",
                        "synonym_filter"
                    ]
                }
            }
        }
    },
    
    "mappings": {
        "properties": {
            "query": {
                "type": "completion",
                "analyzer": "autocomplete_analyzer",
                "search_analyzer": "autocomplete_search_analyzer"
            },
        }
    }
}

# index creation
es.indices.create(index=INDEX_NAME, body=SETTINGS)
elasticsearch.client.indices.IndicesClient(es).put_alias(INDEX_NAME, ALIAS_NAME)

# index filling
def fill_index(es, suggests: dict, index_name: str = INDEX_NAME):
    actions = []
    for text, weight in tqdm(suggests.items()):
        actions.append({
            "_index": index_name,
            "query": {
                "input": text,
                "weight": weight
            }
        })
    helpers.bulk(es, actions, refresh=True)

suggests = {
    'contemporary': 1,
    'cheese': 1,
}

fill_index(es, suggests)

# Elasticsearch query
def suggest_on_prefix(
        es: Elasticsearch,
        prefix: str,
        min_length: int,
        fuzziness: str,
        index_name: str = INDEX_NAME,
        size: int = 4,
        verbose: bool = True
):
    response = es.search(
        index=index_name,
        body={
            "suggest": {
                "text-suggest": {
                "prefix": prefix,
                "completion": {
                        "field": "query",
                        "fuzzy": {
                            "fuzziness": fuzziness,
                            "min_length": min_length,
                            "unicode_aware": True
                        },
                        "size": size
                        }
                    }
                }
            }
    )
    
    text_suggestions = response["suggest"]["text-suggest"][0]["options"]
    
    if verbose:
        print(f"prefix: '{prefix}'\n")
        print(f"Score\tSuggest")
        for option in text_suggestions:
            print(f"{option['_score']}\t{option['text']}")
        print()
    
    return text_suggestions

# example 1
min_length = 5
fuzziness = 'AUTO:5,8'

prefix = 'conte-'
_ = suggest_on_prefix(es, prefix, min_length, fuzziness) # ok

prefix = 'contempo--'
_ = suggest_on_prefix(es, prefix, min_length, fuzziness) # incorrect

# example 2
min_length = 1
fuzziness = 'AUTO:2,5'

prefix = 'co-'
_ = suggest_on_prefix(es, prefix, min_length, fuzziness) # incorrect

prefix = 'contempo--'
_ = suggest_on_prefix(es, prefix, min_length, fuzziness) # ok

Logs (if relevant)

No response

[elasticsearchmachine]

Pinging @elastic/es-search (Team:Search)

[pigeongcc]

Hello there! Anyone can help with the issue? It's highly important for our application. Thanks

[elasticsearchmachine]

Pinging @elastic/es-search-relevance (Team:Search Relevance)

[pigeongcc]

Hello. Anyone can help with the issue? It's highly important for our application. Thank you

[pigeongcc]

Hello there! Anyone can help with the issue? It's highly important for our application 🙏🏻

[jimczi]

"autocomplete_filter": {
"type": "edge_ngram",
"min_gram": 1,
"max_gram": 20
},

Remove the edge_ngram filter as it’s unnecessary and causing the "issue". The completion suggester already operates on prefixes, so indexing ngrams like c, co, cont, etc., isn't required. Since queries on the completion field are prefix-based, c will already match contemporary without needing to index every partial term.
After removing the edge_ngram, a reindex is needed but then all the examples you provided should work fine.

[pigeongcc]

thank you! it solved the problem. I appreciate your support