Searching for Array of Objects

[davidvella]

I have taken an example from Website

Lets says I have two objects in my index

{
  "id": 0,
  "patient_name": "Imogen Temult",
  "appointments": [
    {
      "date": "2022-01-01",
      "doctor": "Jester Lavorre",
      "ward": "psychiatry"
    },
    {
      "date": "2019-01-01",
      "doctor": "Dorian Storm"
    }
  ]
},
{
  "id": 1,
  "patient_name": "Brian Temult",
  "appointments": [
    {
      "date": "2022-01-01",
      "doctor": "Dorian Storm",
      "ward": "psychiatry"
    },
    {
      "date": "2019-01-01",
      "doctor": "Jester Lavorre"
    }
  ]
}

If I do a search for appointments.doctor="Dorian Storm" and appointments.date="2022-01-01" it returns the first object and second object.

I want to return all the documents where they had an appointment with "Dorian Storm" on 2022-01-01 which should only return the second.

[oluademola]

Hello @davidvella

You could transform your data as follow:

{
    "id": 0,
    "patient_name": "Imogen Temult",
    "appointments": [
       "0": {
            "date": "2022-01-01",
            "doctor": "Jester Lavorre",
            "ward": "psychiatry"
        },
       "1": {
            "date": "2019-01-01",
            "doctor": "Dorian Storm"
        }
    ]
}

And then apply this filter
"filters": appointments.1.doctor="Dorian Storm" AND appointments.1.date="2019-01-01"

[davidvella]

@oluademola: While it may work in this particular scenario. I would always want to be searching for the first record.

I don't know the order the values will be.

If I take your example and flip the values in the array:

{
    "id": 0,
    "patient_name": "Imogen Temult",
    "appointments": [
       "0": {
            "date": "2019-01-01",
            "doctor": "Dorian Storm"
        },
      "1": {
                  "date": "2022-01-01",
                  "doctor": "Jester Lavorre",
                  "ward": "psychiatry"
              },
    ]
}

And then apply this filter
"filters": appointments.1.doctor="Dorian Storm" AND appointments.1.date="2019-01-01"

It will no longer return the results.

Is there any other way or is this something Meilisearch cannot do.

[oluademola]

Hi @davidvella

I understand this isn't perfect, and this workaround would require you to know the length of the appointments and order to be able to build your search query. However, I am not sure I have a better solution. Maybe @gmourier might have some suggestions.

[madreiling]

Is there any update on this? I'm in the process of reviewing various search indexes and this is a pretty hard knock on being able to use Meilisearch. For instance, Azure's cloud index could query this via:

appointments/any(a: a/doctor eq 'Donan Storm' and a/date eq '2019-01-01')

and still is able to query the other (undesired in this case but still possible) way via:

appointments/any(a: a/doctor eq 'Donan Storm') and appointments/any(a: a/date eq '2019-01-01')

(Not to mention it fully supports native datetimes without converting to unix timestamps but that isn't that big of a deal to me)

I really like Meilisearch and will probably use it for a few personal projects, but not being able to query multiple properties in the context of the same subobject is sadly a pretty glaring missing feature.

[curquiza]

Hello @madreiling

I'm not sure I perfectly understand your use case, but you can use the multisearch feature, so querying multiple indexes

Also, Meilisearch nows handles array and object in documents: https://www.meilisearch.com/docs/learn/advanced/datatypes#array

You can filter and search in sub-fields thanks to the . notation: https://www.meilisearch.com/docs/learn/fine_tuning_results/filtering#filtering-by-nested-fields

[madreiling]

Thanks @curquiza. I understand both those features but they don't solve what @davidvella and I have asked. The use case for this is incredibly real-world and has come up in every index I have ever created in a professional setting. I wouldn't be here otherwise if it was already possible. I really want it to be so I can ditch all other indexes and just use Meilisearch!

Elasticsearch refers to them as Nested Fields; their documentation explains the issue fairly well. You (the developer) must decide before-hand if you want an Object Field or a Nested Field.

Azure Cognitive Search treats standard searches similarly to how Meilisearch currently does, where subobjects are flattened and searched together, but then treats filtering unflattened. Each subdocument in an array of objects is its own filterable object independent of the other subdocuments in that field (and you must use their any or all filtering function on them):

Queries like this are uncorrelated for full-text search, unlike filters. In filters, queries over sub-fields of a complex collection are correlated using range variables in any or all. The Lucene query above returns documents containing both "Portland, Maine" and "Portland, Oregon", along with other cities in Oregon. This happens because each clause applies to all values of its field in the entire document, so there's no concept of a "current sub-document".

Do you not see the issue? How with the current . notation would you query for all appointments with a ward of "psychiatry" AND IN THE CONTEXT OF THOSE find the ones with a doctor of "Jester Lavorre". Right now Meilisearch would give you all patients that had appointments where at least one ward was psychiatry and at least one the doctor was "Jester Lavorre". It would not only return patients where they had both together.

Forget everything and just compare the raw logic of these two statements. Its subtle but very different. The first searches appointments where both filters are true per sub-document. The other returns true if that parent object contains at least one appointment with a doctor of Jester Lavorre and one appointment in a psychiatry ward.

1. appointments/any(a: a/doctor eq 'Jester Lavorre' and a/ward eq 'psychiatry')
2. appointments/any(a: a/doctor eq 'Jester Lavorre') and appointments/any(a: a/ward eq 'psychiatry')

The question we are asking is how, with the index structure above, can Meilisearch filter for (1) instead of (2)? Is that possible currently because if so I can't figure it out.

If you'd like a more real-world example please let me know!

Edit: this seems to be another discussion on the topic: meilisearch/product#675

[curquiza]

Hello!
Indeed, in this case Meilisearch does not provide directly the feature you need. Hope the workaround described in the product discussion you linked solved your issue

[gmourier]

Hey everyone,

As mentioned by @madreiling and @curquiza, Meilisearch has yet to support this feature. The product discussion takes place here.

If you need that feature, come vote and discuss it!

Note: I'm marking that message as an answer and closing this one to direct more people to the feature request.