How to use Elasticsearch’s range types with Spring Data Elasticsearch

Elasticsearch allows the data, that is stored in a document, to be not only of elementary types, but also of a range of types, see the documentation. With a short example I will explain this range type and how to use it in Spring Data Elasticsearch (the current version being 4.0.3).

For this example we want be able to answer the question: “Who was president of the United States of America in the year X?”. We will store in Elasticsearch a document describing a president with the name and his term, defined be a range of years, defined by a from and to value. We will then query the index for documents where this term range contains a given value

The first thing we need to define is our entity. I named it President:

@Document(indexName = "presidents")
public class President {
    @Id
    private String id;

    @Field(type = FieldType.Text)
    private String name;

    @Field(type = FieldType.Integer_Range)
    private Term term;

    static President of(String name, Integer from, Integer to) {
        return new President(name, new Term(from, to));
    }

    public President() {
    }

    public President(String name, Term term) {
        this(UUID.randomUUID().toString(), name, term);
    }

    public President(String id, String name, Term term) {
        this.id = id;
        this.name = name;
        this.term = term;
    }

    // getter/setter

    static class Term {
        @Field(name = "gte")
        private Integer from;
        @Field(name = "lte")
        private Integer to;

        public Term() {
        }

        public Term(Integer from, Integer to) {
            this.from = from;
            this.to = to;
        }

        // getter/setter
    }
}

There are the standard annotations for a Spring Data Elasticsearch entity like @Document and @Id, but in addition there is the property term that is annotated with @Field(type = FieldType.Integer_Range) (line 9). This marks it as an integer range property. The Term class is defined as inner class at line 31 (not to be confused with the Elasticsearch Term), it defines the term of a president with the two value from and to. Elasticsearch needs for a range the fields to be named gte and lte, this we achieve by defining these names with the @Field annotations in lines 32 and 34.

The rest is just a basic repository:

public interface PresidentRepository extends ElasticsearchRepository<President, String> {
    SearchHits<President> searchByTerm(Integer year);
}

Here we use a single Integer as value because Elasticsearch does the magic by finding the corresponding entries where the searched value is in the range of the stored documents.

And of yourse we have some Controller using it. This Controller has one endpoint that loads the presidents since World War II into Elasticsearch, and a second one returns the desired results:

@RequestMapping("presidents")
public class PresidentController {

    private final PresidentRepository repository;

    public PresidentController(PresidentRepository repository) {
        this.repository = repository;
    }

    @GetMapping("/load")
    public void load() {
        repository.saveAll(Arrays.asList(
                President.of("Dwight D Eisenhower", 1953, 1961),
                President.of("Lyndon B Johnson", 1963, 1969),
                President.of("Richard Nixon", 1969, 1974),
                President.of("Gerald Ford", 1974, 1977),
                President.of("Jimmy Carter", 1977, 1981),
                President.of("Ronald Reagen", 1981, 1989),
                President.of("George Bush", 1989, 1993),
                President.of("Bill Clinton", 1993, 2001),
                President.of("George W Bush", 2001, 2009),
                President.of("Barack Obama", 2009, 2017),
                President.of("Donald Trump", 2017, 2021)));
    }

    @GetMapping("/term/{year}")
    public SearchHits<President> searchByTerm(@PathVariable Integer year) {
        return repository.searchByTerm(year);
    }
}

See it in action (I am using HTTPie), my application is listening on port 9090:

$ http -b :9090/presidents/term/2009
{
    "aggregations": null,
    "empty": false,
    "maxScore": 1.0,
    "scrollId": null,
    "searchHits": [
        {
            "content": {
                "id": "c3a3a0d0-d835-4a02-a2e8-20cc1c0e9285",
                "name": "George W Bush",
                "term": {
                    "from": 2001,
                    "to": 2009
                }
            },
            "highlightFields": {},
            "id": "c3a3a0d0-d835-4a02-a2e8-20cc1c0e9285",
            "score": 1.0,
            "sortValues": []
        },
        {
            "content": {
                "id": "36416746-ff11-4243-a4f3-a6bb0cff9a93",
                "name": "Barack Obama",
                "term": {
                    "from": 2009,
                    "to": 2017
                }
            },
            "highlightFields": {},
            "id": "36416746-ff11-4243-a4f3-a6bb0cff9a93",
            "score": 1.0,
            "sortValues": []
        }
    ],
    "totalHits": 2,
    "totalHitsRelation": "EQUAL_TO"
}

$http -b :9090/presidents/term/2000
{
    "aggregations": null,
    "empty": false,
    "maxScore": 1.0,
    "scrollId": null,
    "searchHits": [
        {
            "content": {
                "id": "984fdf87-a7d8-4dc2-b2e8-5dd948065147",
                "name": "Bill Clinton",
                "term": {
                    "from": 1993,
                    "to": 2001
                }
            },
            "highlightFields": {},
            "id": "984fdf87-a7d8-4dc2-b2e8-5dd948065147",
            "score": 1.0,
            "sortValues": []
        }
    ],
    "totalHits": 1,
    "totalHitsRelation": "EQUAL_TO"
}

So just with putting the right types and names into our @Field annotations we are able to use the range types of Elasticsearch in our Spring Data Elasticsearch application.

Search entities within a geographic distance with Spring Data Elasticsearch 4

A couple of months ago I published the post Using geo-distance sort in Spring Data Elasticsearch 4. In the comments there came up the question “What about searching within a distance?”

Well, this is not supported by query derivation from the method name, but it can easily be done with a custom repository implementation (see the documentation for more information about that).

I updated the example – which is available on GitHub – and will explain what is needed for this implementation. I won’t describe the entity and setup, please check the original post for that.

The custom repository interface

First we need to define a new repository interface that defines the method we want to provide:

public interface FoodPOIRepositoryCustom {

    /**
     * search all {@link FoodPOI} that are within a given distance of a point
     *
     * @param geoPoint
     *     the center point
     * @param distance
     *     the distance
     * @param unit
     *     the distance unit
     * @return the found entities
     */
    List<SearchHit<FoodPOI>> searchWithin(GeoPoint geoPoint, Double distance, String unit);
}

The custom repository implementation

Next we need to provide an implementation, important here is that this is named like the interface with the suffix “Impl”:

public class FoodPOIRepositoryCustomImpl implements FoodPOIRepositoryCustom {

    private final ElasticsearchOperations operations;

    public FoodPOIRepositoryCustomImpl(ElasticsearchOperations operations) {
        this.operations = operations;
    }

    @Override
    public List<SearchHit<FoodPOI>> searchWithin(GeoPoint geoPoint, Double distance, String unit) {

        Query query = new CriteriaQuery(
          new Criteria("location").within(geoPoint, distance.toString() + unit)
        );

        // add a sort to get the actual distance back in the sort value
        Sort sort = Sort.by(new GeoDistanceOrder("location", geoPoint).withUnit(unit));
        query.addSort(sort);

        return operations.search(query, FoodPOI.class).getSearchHits();
    }
}

In this implementation we have an ElasticsearchOperations instance injected by Spring. In the method implementation we build a NativeSearchQuery that specifies the distance query we want. In addition to that we add a GeoDistanceSort to have the actual distance of the found entities in the output. We pass this query to the ElasticsearchOperations instance and return the search result.

Adapt the repository

We need to add the new interface to our FoodPOIRepository definition, which otherwise is unchanged:

public interface FoodPOIRepository extends ElasticsearchRepository<FoodPOI, String>, FoodPOIRepositoryCustom {

    List<SearchHit<FoodPOI>> searchTop3By(Sort sort);

    List<SearchHit<FoodPOI>> searchTop3ByName(String name, Sort sort);
}

Use it in the controller

In the rest controller, there is a new method that uses the distance search:

@PostMapping("/within")
List<ResultData> withinDistance(@RequestBody RequestData requestData) {

    GeoPoint location = new GeoPoint(requestData.getLat(), requestData.getLon());

    List<SearchHit<FoodPOI>> searchHits
        = repository.searchWithin(location, requestData.distance, requestData.unit);

    return toResultData(searchHits);
}

private List<ResultData> toResultData(List<SearchHit<FoodPOI>> searchHits) {
    return searchHits.stream()
        .map(searchHit -> {
            Double distance = (Double) searchHit.getSortValues().get(0);
            FoodPOI foodPOI = searchHit.getContent();
            return new ResultData(foodPOI.getName(), foodPOI.getLocation(), distance);
        }).collect(Collectors.toList());
}

We extract the needed parameters from the requestData that came in, call our repository method and convert the results to our output format.

And that’s it

So with a small custom repository implementation we were able to add the desired functionality to our repository

Use an index name defined by the entity to store data in Spring Data Elasticsearch 4.0

When using Spring Data Elasticsearch (I am referencing the current version 4.0.2), normally the name of the index where the documents are stored is taken from the @Document annotation of the entity class – here it’s books:

@Document(indexName="books")
public class Book {
  // ...
}

Recently in a discussion of a Pull Request in Spring Data Elasticsearch, someone told that she needed a possibility to extract the name from the entity itself, as entities might go to different indices.

In this post I will show how this can be done by using Spring Data Repository customization by providing a custom implementation for the save method. A complete solution would need to customize saveAll and other methods as well, but I will restrict this here to just one method.

The Hotel entity

For this post I will use an entity describing a hotel, with the idea that hotels from different countries should be stored in different Elasticsearch indices. The index name in the annotation is a wildcard name so that when searching for hotels all indices are considered.

Hotel.java

package com.sothawo.springdataelastictest;

import org.springframework.data.annotation.Id;
import org.springframework.data.elasticsearch.annotations.Document;
import org.springframework.data.elasticsearch.annotations.Field;
import org.springframework.data.elasticsearch.annotations.FieldType;
import org.springframework.lang.Nullable;

/**
 * @author P.J. Meisch (pj.meisch@sothawo.com)
 */
@Document(indexName = "hotel-*", createIndex = false)
public class Hotel {
    @Id
    @Nullable
    private String id;

    @Field(type = FieldType.Text)
    @Nullable
    private String name;

    @Field(type = FieldType.Keyword)
    @Nullable
    private String countryCode;

    // getter/setter ...
}

The custom repository

We need to define a custom repository interface that defines the methods we want to implement. Since we want to customize the save method that ElasticsearchRepository has by extending CrudRepository, we need to use the very same method signature including the generics:

CustomHotelRepository.java

package com.sothawo.springdataelastictest;

/**
 * @author P.J. Meisch (pj.meisch@sothawo.com)
 */
public interface CustomHotelRepository<T> {
    <S extends T> S save(S entity);
}

The next class to provide is an implementation of this interface. It is important that the implementation class is named like the interface with a Impl suffix:

CustomHotelRepositoryImpl.java

package com.sothawo.springdataelastictest;

import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import org.springframework.data.elasticsearch.core.ElasticsearchOperations;
import org.springframework.data.elasticsearch.core.IndexOperations;
import org.springframework.data.elasticsearch.core.document.Document;
import org.springframework.data.elasticsearch.core.mapping.IndexCoordinates;
import org.springframework.lang.NonNull;
import org.springframework.lang.Nullable;

import java.util.concurrent.ConcurrentHashMap;

/**
 * @author P.J. Meisch (pj.meisch@sothawo.com)
 */
@SuppressWarnings("unused")
public class CustomHotelRepositoryImpl implements CustomHotelRepository<Hotel> {

    private static final Logger LOG = LoggerFactory.getLogger(CustomHotelRepositoryImpl.class);

    private final ElasticsearchOperations operations;

    private final ConcurrentHashMap<String, IndexCoordinates> knownIndexCoordinates = new ConcurrentHashMap<>();
    @Nullable
    private Document mapping;

    @SuppressWarnings("unused")
    public CustomHotelRepositoryImpl(ElasticsearchOperations operations) {
        this.operations = operations;
    }

    @Override
    public <S extends Hotel> S save(S hotel) {

        IndexCoordinates indexCoordinates = getIndexCoordinates(hotel);
        LOG.info("saving {} to {}", hotel, indexCoordinates);

        S saved = operations.save(hotel, indexCoordinates);

        operations.indexOps(indexCoordinates).refresh();

        return saved;
    }

    @NonNull
    private <S extends Hotel> IndexCoordinates getIndexCoordinates(S hotel) {

        String indexName = "hotel-" + hotel.getCountryCode();
        return knownIndexCoordinates.computeIfAbsent(indexName, i -> {

                IndexCoordinates indexCoordinates = IndexCoordinates.of(i);
                IndexOperations indexOps = operations.indexOps(indexCoordinates);

                if (!indexOps.exists()) {
                    indexOps.create();

                    if (mapping == null) {
                        mapping = indexOps.createMapping(Hotel.class);
                    }

                    indexOps.putMapping(mapping);
                }
                return indexCoordinates;
            }
        );
    }
}

This implementation is a Spring Bean (no need for adding @Component) and so can use dependency injection. Let me explain the code.

Line 22: the ElasticsearchOperations object we will use to store the entity in the desired index, this is autowired by constructor injection in lines 29-31

Line 24-26: As we want to make sure that the index we write to exists and has the correct mapping, we keep track of which indices we already know. This is used in the getIndexCoordinates method explained later.

Line 34 to 44: This is the actual implementation of the save operation. First we call getIndexCoordinates which will make sure the index exists. We pass the indexCoordinates into the save method of the ElasticsearchOperations instance. If we would use ElasticsearchOperations.save(hotel), the name from the @Document annotation would be used. But when passing an IndexCoordinates as second parameter, the index name from this is used to store the entity. In line 41 there is a call to refresh, this is the behaviour of the original ElasticsearchRepository.save() method, so we do the same here. If you do not need the immediate refresh, omit this line.

Line 47 to 76: As Spring Data Elasticsearch does not yet support index templates (this will come with version 4.1) this method ensures, that when the first time that an entity is saved to an index, this index is created if necessary and writes the mappings to the new created index.

The HotelRepository to use in the application

We now need to combine our custom repository with the ElasticsearchRepository from Spring Data Elasticsearch:

HotelRepository.java

package com.sothawo.springdataelastictest;

import org.springframework.data.elasticsearch.core.SearchHits;
import org.springframework.data.elasticsearch.repository.ElasticsearchRepository;

/**
 * @author P.J. Meisch (pj.meisch@sothawo.com)
 */
public interface HotelRepository extends ElasticsearchRepository<Hotel, String>, CustomHotelRepository<Hotel> {
    SearchHits<Hotel> searchAllBy();
}

Here we combine the two interfaces and define an additional method that returns all hotels in a SearchHits object.

Use the repository in the code

The only thing that’s left is to use this repository, for example in a REST controller:

HotelController.java

package com.sothawo.springdataelastictest;

import org.springframework.data.elasticsearch.core.SearchHits;
import org.springframework.web.bind.annotation.GetMapping;
import org.springframework.web.bind.annotation.PostMapping;
import org.springframework.web.bind.annotation.RequestBody;
import org.springframework.web.bind.annotation.RequestMapping;
import org.springframework.web.bind.annotation.RestController;

/**
 * @author P.J. Meisch (pj.meisch@sothawo.com)
 */
@RestController
@RequestMapping("/hotels")
public class HotelController {

    private final HotelRepository repository;

    public HotelController(HotelRepository repository) {
        this.repository = repository;
    }

    @GetMapping()
    public SearchHits<Hotel> all() {
        return repository.searchAllBy();
    }

    @PostMapping()
    public Hotel save(@RequestBody Hotel hotel) {
        return repository.save(hotel);
    }
}

This is a standard controller which has a HotelRepository instance injected (which Spring Data Elasticsearch will create for us). This looks exactly how it would without our customization. The difference is that the call to save() ends up in our custom implementation.

Conclusion

This post shows how easy it is to provide custom implementations for the methods that are normally provided by Spring Data Repositories (not just in Spring Data Elasticsearch) if custom logic is needed.

How to provide a dynamic index name in Spring Data Elasticsearch using SpEL

In Spring Data Elasticsearch – at the time of writing, version 4.0 is the current version – the name of an index is normally defined by the @Document annotation on the entity class. For the following examples let’s assume we want to write some log entries to Elasticsearch with our application. We use the following entity:

@Document(indexName = "log")
public class LogEntity {
    @Id
    private String id = UUID.randomUUID().toString();

    @Field(type = FieldType.Text)
    private String text;

    @Field(name = "log-time", type = FieldType.Date, format = DateFormat.basic_date_time)
    private ZonedDateTime logTime = ZonedDateTime.now();

    public String getId() {
        return id;
    }

    public String getText() {
        return text;
    }

    public void setText(String text) {
        this.text = text;
    }

    public ZonedDateTime getLogTime() {
        return logTime;
    }

    public void setLogTime(ZonedDateTime logTime) {
        this.logTime = logTime;
    }
}

Here the index name is the fixed name log.

it is possible to use a dynamically defined name for an index by using Spring Expression Language SpEL. Important: We need to use a SpEL template expression, that is an expression enclosed in #{}. This allows for the following setups:

Use a value from the application configuration

Let’s assume we have the following entry in the application.properties file:

index.prefix=test

We then use this code

@Document(indexName = "#{@environment.getProperty('index.prefix')}-log")

and the index name to use changes to test-log.

Use a value provided by a static method of some class

The second example shows how to call a static function to get a dynamic value. We use the following definition to add the current date to the index name:

@Document(indexName = "log-#{T(java.time.LocalDate).now().toString()}")

Currently this would provide an index name of log-2020-07-28.

Use a value provided by a Spring bean

For the third case we provide a bean that will give us a dynamically created string to be used as part of the index name.

@Component
public class LogIndexNameProvider {

    public String timeSuffix() {
        return LocalTime.now().truncatedTo(ChronoUnit.MINUTES).toString().replace(':', '-');
    }
}

This bean, named logIndexNameProvider, can return a String that contains the current time as hh-mm (I would not use this for naming indices, but this is just an example).

Changing the definition to

@Document(indexName = "log-#{@logIndexNameProvider.timeSuffix()}")

will now create index names like log-08-25 or log-22-07.

Of course we can mix all of these together: add a prefix from the configuration, append the current date.

Important notice:

The evaluation of SpEL for index names is only done for the index names defined in the @Document annotation. It is not done for index names that are passed as a IndexCoordinates parameter in the different methods of the ElasticsearchOperations or IndexOperations interfaces. If it were allowed on these, it would be easy to set up a scenario, where an expression is read from some outside source. And then someone might send something like "log-#{T(java.lang.Runtime).getRuntime().exec(new String[]{'/bin/rm', '/tmp/somefile'})}" which will not provide an index name, but delete files on your computer.

Using geo-distance sort in Spring Data Elasticsearch 4

The release of Spring Data Elasticsearch in version 4.0 (see the documentation) brings two new features that now enable users to use geo-distance sorts in repository queries: The first is a new class GeoDistanceOrder and the second is a new return type for repository methods SearchHit<T>. In this post I will show how easy it is to use these classes to answer questions like “Which pubs are the nearest to a given location?”.

The source code

The complete runnable code used for this post is available on GitHub. In order to run the application you will need Java 8 or higher and a running instance of Elasticsearch. If this is not accessible at localhost:9200 you need to set the correct value in the src/main/resources/application.yaml file.

Update 12.09.2020

The original code was a little extended for the follow-up post Search entities within a geographic distance with Spring Data Elasticsearch 4

The sample data

For this sample application I use a csv file with POI data from OpenStreetMap that contains POIs in Germany which are categorized as kind of food, like restaurants, pubs, fast-food and more. All together there are 826843 records.

When the application is started, the index in Elasticsearch is created and loaded with the data if it does not yet exist. So the first startup takes a little longer, the progress can be seen on the console. Within the application, these POIs are modelled by the following entity:

@Document(indexName = "foodpois")
public class FoodPOI {
    @Id
    private String id;
    @Field(type = FieldType.Text)
    private String name;
    @Field(type = FieldType.Integer)
    private Integer category;
    private GeoPoint location;
    // constructors, getter/setter left out for brevity
}

The interesting properties for this blog post are the location and the name.

The Repository

In order to search the data we need a Repository Definition:

public interface FoodPOIRepository extends ElasticsearchRepository<FoodPOI, String> {
    List<SearchHit<FoodPOI>> searchTop3By(Sort sort);
    List<SearchHit<FoodPOI>> searchTop3ByName(String name, Sort sort);
}

We have two functions defined, the first we will use to search any POI near a given point, with the second on we can search for the POIs with a name. Defining these methods in the interface is all we need as Spring Data Elasticsearch will under the hood create the implementation for these methods by analyzing the method names and parameters.

In Spring Data Elasticsearch before version 4 we could only get a List<FoodPOI> from a repository method. But now there is the SearchHit<T> class, which not only contains the entity, but also other values like a score, highlights or – what we need here – the sort value. When doing a geo-distance sort, the sort value contains the actual distance of the POI to the value we passed into the search.

The Controller

We define a REST controller, so we can call our application to get the data. The request parameters will come in a POST body that will be mapped to the following class:

public class RequestData {
    private String name;
    private double lat;
    private double lon;
    // constructors, getter/setter ...
}

The result data that will be sent to the client looks like this:

public class ResultData {
    private String name;
    private GeoPoint location;
    private Double distance;

  // constructor, gette/setter ...
}

The controller has just one method:

@RestController
@RequestMapping("/foodpois")
public class FoodPOIController {

    private final FoodPOIRepository repository;

    public FoodPOIController(FoodPOIRepository repository) {
        this.repository = repository;
    }

    @PostMapping("/nearest3")
    List<ResultData> nearest3(@RequestBody RequestData requestData) {

        GeoPoint location = new GeoPoint(requestData.getLat(), requestData.getLon());
        Sort sort = Sort.by(new GeoDistanceOrder("location", location).withUnit("km"));

        List<SearchHit<FoodPOI>> searchHits;

        if (StringUtils.hasText(requestData.getName())) {
            searchHits = repository.searchTop3ByName(requestData.getName(), sort);
        } else {
            searchHits = repository.searchTop3By(sort);
        }

        return searchHits.stream()
            .map(searchHit -> {
                Double distance = (Double) searchHit.getSortValues().get(0);
                FoodPOI foodPOI = searchHit.getContent();
                return new ResultData(foodPOI.getName(), foodPOI.getLocation(), distance);
            }).collect(Collectors.toList());
    }
}

In line 15 we create a Sort object that specifies that Elasticsearch should return the data ordered by the geographical distance to the given value which we take from the request data. Then, depending if we have a name, we call the corresponding method and get back a List<SearchHit<FoodPOI>>.

We then in the lines 27 to 29 extract the information we need from the returned objects and build our result data object.

Check the result

After starting the application we can hit the endpoint. I use curl here and pipe the output through jq to have it formatted:

$curl -X "POST" "http://localhost:8080/foodpois/nearest3" \
     -H 'Content-Type: application/json; charset=utf-8' \
     -d $'{
  "lat": 49.02,
  "lon": 8.4
}'|jq

[
  {
    "name": "Cantina Majolika",
    "location": {
      "lat": 49.0190808,
      "lon": 8.4014792
    },
    "distance": 0.14860088197123017
  },
  {
    "name": "Waldgaststätte FSSV",
    "location": {
      "lat": 49.023578,
      "lon": 8.3954656
    },
    "distance": 0.5173117164589114
  },
  {
    "name": "Hatz",
    "location": {
      "lat": 49.0155358,
      "lon": 8.3975457
    },
    "distance": 0.5276800664204232
  }
]

And the Pubs?

curl -X "POST" "http://localhost:8080/foodpois/nearest3" \
     -H 'Content-Type: application/json; charset=utf-8' \
     -d $'{
  "lat": 49.02,
  "lon": 8.4,
  "name": "pub"
}'|jq
[
  {
    "name": "Scruffy's Irish Pub",
    "location": {
      "lat": 49.0116335,
      "lon": 8.3950194
    },
    "distance": 0.998711100164643
  },
  {
    "name": "Irish Pub “Sean O'Casey's”",
    "location": {
      "lat": 49.0090639,
      "lon": 8.4028365
    },
    "distance": 1.2335132790824628
  },
  {
    "name": "Oxford Pub",
    "location": {
      "lat": 49.0086149,
      "lon": 8.4129781
    },
    "distance": 1.5806674447458173
  }
]

And that’s it

Without even needing to know how these request are sent to Elasticsearch and what Elasticsearch sends back, we can easily use these features in our Spring application. Hope you enjoyed it!