DocumentDB find by dynamic subdocument's value (Spring Data Mongo) - spring

Question

I have a collection with multi-million documents. Each contains a dynamic subdocument (let's call it as context).
I need to query for all of those which contains a given value.
Example documents:
{
"name": "test"
[...]
"context": {
"key": "some value",
"another key": "some other value",
[...]
}
},
{
"name": "joe doe"
[...]
"context": {
"just another key": "a value here",
"last example key": "and its value",
[...]
}
}
What I want is to find all documents which one's context contains a given value. For example "a value here".
Using aggregation, I was able to transform all the values from the context into a single text and check if it contains that value. This just works fine, but it is slow as hell.
val filterExistsStage = Aggregation.match(Criteria("context").exists(true))
val flatVariablesStage = Aggregation.addFields().addField("text").withValue(ObjectToArray.valueOfToArray("context")).build()
val extractValuesStage = Aggregation.addFields().addField("text").withValue("\$text.v").build()
val condition = FilterCondition("text", filter.context, FilterOperation.CONTAINS)
val criteria = CRITERIA[condition.operator]!!(condition)
val matchStage = Aggregation.match(criteria)
return listOf(
filterExistsStage,
flatVariablesStage,
extractValuesStage,
matchStage
)
val CRITERIA = mapOf<FilterOperation, (condition: FilterCondition) -> Criteria>(
FilterOperation.CONTAINS to { Criteria.where(it.field).regex(Regex.escape(it.value as String)) },
)
Another solution would be the so called wildcard text indices. This is really fast but unfortunately DocumentDB (compatibile with Mongo 3.6 and 4.0) does not support it. It is only available from MongoDB v4.2.
mongoTemplate.indexOps(MyEntity::class.java)
.ensureIndex(TextIndexDefinition.forAllFields())
val textCriteria = TextCriteria.forDefaultLanguage().matching(filter.context)
val textMatchStage = Aggregation.match(textCriteria)
I also thought to save all these values into a single field as text and do simple text search. But I try to avoid this kind of optimization if there are better solutions.
I wonder if there is any solution for this in DocumentDB or better to use some kind of text search engine, like ElasticSearch?
It has to be fast on millions of documents.