Optimizing GraphQL Queries optimizing-graphql-queries

These guidelines are provided to help prevent performance issues with your GraphQL queries.

GraphQL Checklist graphql-checklist

The following checklist aims to help you optimize the configuration and use of GraphQL in Adobe Experience Manager (AEM) as a Cloud Service.

First Principles first-principles

Use persisted GraphQL queries use-persisted-graphql-queries

Recommendation

Use of persisted GraphQL queries is strongly recommended.

Persisted GraphQL queries help reduce query execution performance by utilizing the Content Delivery Network (CDN). Client applications request persisted queries with GET requests for fast edge-enabled execution.

Further Reference

See:

Cache Strategy cache-strategy

Various methods of caching can also be used for optimization.

Enable AEM Dispatcher caching enable-aem-dispatcher-caching

Recommendation

AEM Dispatcher is the first level cache within the AEM service, before CDN cache.

Further Reference

See:

Use a Content Delivery Network (CDN) use-cdn

Recommendation

GraphQL queries and their JSON responses can be cached if targeted as GET requests when using a CDN. In contrast, uncached requests can be very (resource) expensive and slow to process, with the potential for further detrimental effects on the origin’s resources.

Further Reference

See:

Set HTTP cache control headers set-http-cache-control-headers

Recommendation

When using persisted GraphQL queries with a CDN, it is recommended to set appropriate HTTP cache control headers.

Each persisted query can have its own specific set of cache control headers. The headers can be set over the GraphQL API or the AEM GraphiQL IDE.

Further Reference

See:

GraphQL Query optimization graphql-query-optimization

On an AEM instance with a high number of Content Fragments that share the same model, GraphQL list queries can become costly (in terms of resources).

This is because all fragments that share a model being used within the GraphQL query have to be loaded into memory. This consumes both time and memory. Filtering, which may reduce the number of items in the (final) result set, can only be applied after loading the entire result set into memory.

This can lead to the impression that even small result sets (can) lead to bad performance. However, in reality the slowness is caused by the size of the initial result set, as it has to be handled internally before filtering can be applied.

To reduce performance and memory issues, this initial result set has to be kept as small as possible.

AEM provides two approaches for optimizing GraphQL queries:

Each approach has its own use-cases and limitations. This section provides information on Hybrid Filtering and Paging, together with some of the best practices for use in optimizing GraphQL queries.

Use AEM GraphQL hybrid filtering use-aem-graphql-hybrid-filtering

Recommendation

Hybrid filtering combines JCR filtering with AEM filtering.

It applies a JCR filter (in the form of a query constraint) before loading the result set into memory for AEM filtering. This is to reduce the result set loaded into memory, as the JCR filter removes superfluous results prior to this.

This technique keeps the flexibility that GraphQL filters provide, while delegating as much of the filtering as possible to JCR.

Further Reference

See:

Use GraphQL pagination use-aem-graphql-pagination

Recommendation

The response time of complex queries, with large result sets, can be improved by segmenting responses into chunks using pagination, a GraphQL standard.

GraphQL in AEM provides support for two types of pagination:

Further Reference

See:

Use GraphQL sorting use-graphql-sorting

Recommendation

Also a GraphQL standard, sorting enables clients to receive JSON content in sorted order. This can reduce the need for further processing on the client.

Sorting can only be efficient if all sort criteria are related to top-level fragments.

If the sorting order includes one, or more, fields that are located on a nested fragment, then all fragments sharing the top-level model must be loaded into memory. This causes a negative performance impact.

Further Reference

See:

Best Practices best-practices

The main goal of all optimization recommendations is to reduce the initial result set. The best practices listed here provide ways to do so. They can (and should) be combined.

Filter on top-level properties only filter-top-level-properties-only

Currently, filtering at the JCR level only works for top-level fragments.

If a filter addresses the fields of a nested fragment, AEM has to fall back to loading (into memory) all fragments that share the underlying model.

You can still optimize such GraphQL queries by combining filter expressions on fields of top-level fragments and those on fields of nested fragments with the AND operator.

Use the content structure use-content-structure

In AEM, it is generally considered good practice to use the repository structure to narrow down the scope of content to be processed.

This approach should also be applied to GraphQL queries.

This can be done by applying a filter on the _path field of the top-level fragment:

{
  someList(filter: {
    _path: {
      _expressions: [
        {
          value: "/content/dam/some/sub/path/",
          _operator: STARTS_WITH
        }
      ]
    }
  }) {
    items {
      # ...
    }
  }
}

Use paging use-paging

You can also use paging to reduce the initial result set; especially if your requests do not use any filtering and sorting.

If you filter or sort on nested fragments, paginated queries can still be slow, as AEM may still need to load larger amounts of fragments into memory. Therefore, if you combine filtering and paging, consider the rules for filtering (as mentioned above).

For paging, sorting is equally important, as paginated results are always sorted - either in an explicit or an implicit way.

If you are primarily interested in only retrieving the first few pages, there is no significant difference between using the ...List or ...Paginated queries. However, if your application is interested in reading more than just one or two pages, you should consider the ...Paginated query, as it performs notably better with the later pages.

Logical operations in filter expressions logical-operations-in-filter-expressions

If you are filtering on nested fragments, you can still apply JCR filtering by providing an accompanying filter on a top-level field that is combined using the AND operator.

A typical use-case would be to restrict the scope of the query using a filter on the _path field of the top-level fragment, and then filter on additional fields that might be on the top-level, or on a nested fragment.

In this case, the different filter expressions would be combined with AND. Therefore, the filter on _path can effectively limit the initial result set. All other filters on top-level fields can help with reducing the initial result set as well - as long as they are combined with AND.

Filter expressions combined with OR cannot be optimized if nested fragments are involved. OR expressions can only be optimized if no nested fragments are involved.

Avoid filtering on multiline textfields avoid-filtering-multiline-textfields

The fields of a multiline textfield (html, markdown, plaintext, json) cannot be filtered through a JCR query, as the content of these fields have to be calculated on the fly.

If you still need to filter on a multiline textfield, consider limiting the size of the initial result set by adding additional filter expressions and combine them with AND. Limiting the scope through filtering on the _path field is a good approach as well.

Avoid filtering on virtual fields avoid-filtering-virtual-fields

Virtual fields (most fields starting with _) are calculated while a GraphQL query is executed, and are therefore outside the scope of JCR-based filtering.

One important exception is the _path field, which can be used effectively to reduce the size of the initial result set - if content is structured accordingly (see Use the content structure).

Filtering: Exclusions filtering-exclusions

There are several other situations where a filter expression cannot be evaluated on the JCR level (and therefore should be avoided to achieve the best performance):

Minimize Content Fragment Nesting minimize-content-fragment-nesting

Nesting Content Fragments is a great way to model custom content structures. You can even have a fragment with a nested fragment, that also has a nested fragment, that has…and so on.

However, creating a structure with too many levels can increase the processing times for a GraphQL query, as GraphQL has to traverse the entire hierarchy of all nested Content Fragments.

Deep nesting can also have adverse effects on content governance. In general, it is recommended to limit Content Fragment nesting to below five or six levels.

Do not output all formats (Multi line text elements) do-not-output-all-formats

AEM GraphQL can return text, authored in the Multi line text data type, in multiple formats: Rich Text, Simple Text, and Markdown.

Outputting all three formats increases the size of text output in JSON by a factor of three. That, combined with generally large result sets from very broad queries, can produce very large JSON responses that therefore take a long time to compute. It is better to limit the output to only the text formats required for rendering the content.

Modifying Content Fragments modifying-content-fragments

Only modify Content Fragments, and their resources, using the AEM UI or APIs. Do not make modifications directly in JCR.

Test your queries test-your-queries

Processing GraphQL queries is similar to processing search queries, and is significantly more complex than simple GET-all-content API requests.

Carefully planning, testing, and optimizing your queries in a controlled non-production environment is key for later success when used in production.