search

jillesvangurp / kt-search

Multi platform kotlin client for Elasticsearch & Opensearch with easily extendable Kotlin DSLs for queries, mappings, bulk, and more.
MIT License
95 stars 23 forks source link
client-library domain-specific-languages elasticsearch kotlin kotlin-multiplatform kotlinx-coroutines kotlinx-serialization ktor-client opensearch

readme

KT Search Client

matrix-test-and-deploy-docs

Kt-search is a Kotlin Multi Platform library to search across the Opensearch and Elasticsearch ecosystem on any platform that kotlin can compile to. It provides Kotlin DSLs for querying, defining mappings, bulk indexing, index templates, index life cycle management, index aliases, and much more. The key goal for this library is to provide a best in class developer experience for using Elasticsearch and Opensearch.

Why Kt-search?

If you develop software in Kotlin and would like to use Opensearch or Elasticsearch, you have a few choices to make. There are multiple clients to choose from and not all of them work for each version. And then there is Kotlin multi platform to consider as well. Maybe you are running spring boot on the jvm. Or maybe you are using ktor compiled to native or wasm and using that to run lambda functions.

Kt-search has you covered for all of those. The official Elastic or Opensearch clients are Java clients. You can use them from Kotlin but only on the JVM. And they are not source compatible with each other. The Opensearch client is based on a fork of the old Java client which after the fork was deprecated. On top of that, it uses opensearch specific package names.

Kt-search solves a few important problems here:

License

This project is licensed under the MIT license.

Learn more

Use cases

Integrate advanced search capabilities in your Kotlin applications. Whether you want to build a web based dashboard, an advanced ETL pipeline or simply expose a search endpoint as a microservice, this library has you covered.

The goal for kt-search is to be the most convenient way to use opensearch and elasticsearch from Kotlin on any platform where Kotlin is usable.

Kt-search is extensible and modular. You can easily add your own custom DSLs for e.g. things not covered by this library or any custom plugins you use. And while it is opinionated about using e.g. kotlinx.serialization, you can also choose to use alternative serialization frameworks, or even use your own http client and just use the search-dsl.

Gradle

Add the maven.tryformation.com repository:

repositories {
    mavenCentral()
    maven("https://maven.tryformation.com/releases") {
        content {
            includeGroup("com.jillesvangurp")
        }
    }
}

And then the dependency to commonsMain or main:

    // check the latest release tag for the latest version
    implementation("com.jillesvangurp:search-client:2.x.y")

About maven central ... I've switched maven repositories a couple of times now. Jitpack and multiplatform just doesn't work. Of course I would have liked to get this on maven central. However, after repeated attempts to get that done, I've decided to not sacrifice more time on this. The (lack of) documentation, the Jira bureaucracy, the uninformative errors, the gradle plugin, etc. just doesn't add up to something that works for a multi module, multi platform project. I'm sure it can be done but I'm not taking more time out my schedule to find out.

If somebody decides to fix a proper, modern solution for hosting packages, I'll consider using it but I'm done with maven central for now. Google buckets work fine for hosting. So does ssh or any old web server. So does aws. It's just maven central that's a huge PITA.

Maven

If you have maven based kotlin project targeting jvm and can't use kotlin multiplatform dependency, you would need to add jvm targeting artifacts.

Add the maven.tryformation.com repository:

<repositories>
    <repository>
        <id>try-formation</id>
        <name>kt search repository</name>
        <url>https://maven.tryformation.com/releases</url>
    </repository>
</repositories>

And then add dependencies to jvm targets:

<dependencies>
    <dependency>
        <groupId>com.jillesvangurp</groupId>
        <artifactId>search-client-jvm</artifactId>
        <version>2.1.25</version>
    </dependency>
    <dependency>
        <groupId>com.jillesvangurp</groupId>
        <artifactId>search-dsls-jvm</artifactId>
        <version>2.1.25</version>
    </dependency>
    <dependency>
        <groupId>com.jillesvangurp</groupId>
        <artifactId>json-dsl-jvm</artifactId>
        <version>3.0.0</version>
    </dependency>
</dependencies>

Note: The json-dsl is moved to separate repository. To find the latest version number, check releases: https://github.com/jillesvangurp/json-dsl/releases

Usage

Create a client

First we create a client. 

val client = SearchClient()

Kotlin has default values for parameters. So, we use sensible defaults for the host and port variables to connect to localhost and 9200.

val client = SearchClient(
  KtorRestClient(host="localhost", port=9200)
)

If you need ro, you can also configure multiple hosts, add ssl and basic authentication to connect to managed Opensearch or Elasticsearch clusters. If you use multiple hosts, you can also configure a strategy for selecting the host to connect to. For more on this, read the manual.

Documents and data classes

In Kotlin, the preferred way to deal with data would be a data class. This is a simple data class that we will use below.

@Serializable
data class TestDocument(
  val name: String,
  val tags: List<String>? = null
)

In the example below we will use this TestDocument, which we can serialize using the kotlinx.serialization framework. You can also pass in your own serialized json in requests, so if you want to use e.g. jackson or gson instead, you can do so easily.

Creating an index

val indexName = "readme-index"

// create an index and use our mappings dsl
client.createIndex(indexName) {
  settings {
    replicas = 0
    shards = 3
    refreshInterval = 10.seconds
  }
  mappings(dynamicEnabled = false) {
    text(TestDocument::name)
    keyword(TestDocument::tags)
  }
}

This creates the index and uses the mappings and settings DSL. With this DSL, you can map fields, configure analyzers, etc. This is optional of course; you can just call it without the block and use the defaults and rely on dynamic mapping. You can read more about that here

Adding documents

To fill the index with some content, we need to use bulk operations.

In kt-search this is made very easy with a DSL that abstracts away the book keeping that you need to do for this. The bulk block below creates a BulkSession, which does this for you and flushes operations to Elasticsearch. You can configure and tailor how this works via parameters that have sensible defaults. For example the number of operations that is flushed is something that you'd want to probably configure.

The optional refresh parameter uses WaitFor as the default. This means that after the block exits, the documents will have been indexed and are available for searching. 

client.bulk(
  refresh = Refresh.WaitFor,
  // send operations every 2 ops
  // default would be 100
  bulkSize = 2,
) {
  index(
    doc = TestDocument(
      name = "apple",
      tags = listOf("fruit")
    ),
    index = indexName
  )
  index(
    doc = TestDocument(
      name = "orange",
      tags = listOf("fruit", "citrus")
    ),
    index = indexName,
  )
  index(
    // you can also provide raw json
    // but it has to be a single line in the bulk request
    source =
    """{"name":"banana","tags":["fruit","tropical"]}""",
    index = indexName
  )
}

You can read more about bulk operations in the manual.

Search

Now that we have some documents in an index, we can do some queries:

// search for some fruit
val results = client.search(indexName) {
  query = bool {
    must(
      // note how we can use property references here
      term(TestDocument::tags, "fruit"),
      matchPhrasePrefix(TestDocument::name, "app")
    )
  }
}

println("found ${results.total} hits")
results
  // extension function that deserializes
  // the hits using kotlinx.serialization
  .parseHits<TestDocument>()
  .first()
  .let {
    // we feel lucky
    println("doc ${it.name}")
  }
// you can also get the JsonObject if you don't
// have a model class
println(results.hits?.hits?.first()?.source)

This prints:

found 1 hits
doc apple
{"name":"apple","tags":["fruit"]}

You can also construct complex aggregations with the query DSL:

val resp = client.search(indexName) {
  // we don't care about retrieving hits
  resultSize = 0
  agg("by-tag", TermsAgg(TestDocument::tags) {
    // simple terms agg on the tags field
    aggSize = 50
    minDocCount = 1
  })
}
// picking the results apart is just as easy.
resp.aggregations
  .termsResult("by-tag")
  .parsedBuckets.forEach { bucket ->
    println("${bucket.parsed.key}: ${bucket.parsed.docCount}")
  }

This prints:

fruit: 3
citrus: 1
tropical: 1

These examples show off a few nice features of this library:

Manual

There are of course a lot more features that this library supports. The manual covers all of those.

Introduction

Search

Indices and Documents

Advanced Topics

Related projects

There are several libraries that build on kt-search:

Additionally, I also maintain a few other search related projects that you might find interesting.

Setting up a development environment

Any recent version of Intellij should be able to import this project as is. This project uses docker for testing and to avoid having the tests create a mess in your existing elasticsearch cluster, it uses a different port than the default Elasticsearch port.

If you want to save some time while developing, it helps to start docker manually. Otherwise you have to wait for the container to stop and start every time you run a test.

docker-compose -f docker-compose-es-8.yml up -d

For additional details, refer to the build file.

Compatibility

The integration tests on GitHub Actions use a matrix build that tests everything against Elasticsearch 7 & 8 and Opensearch 1 & 2.

It may work fine with earlier Elasticsearch versions as well. But we don't actively test this and the tests are known to not pass with Elasticsearch 6 due to some changes in the mapping dsl. You may be able to work around some of this, however.

There is an annotation that is used to restrict APIs when needed. E.g. search-after only works with Elasticsearch and and has the following annotation to indicate that:

@VariantRestriction(SearchEngineVariant.ES7,SearchEngineVariant.ES8)
suspend fun SearchClient.searchAfter(target: String, keepAlive: Duration, query: SearchDSL): Pair<SearchResponse,Flow<SearchResponse.Hit>> {
    validateEngine("search_after does not work on OS1",
        SearchEngineVariant.ES7, 
        SearchEngineVariant.ES8)

    // ...
}

The annotation is informational only for now. In our tests, we use onlyon to prevent tests from failing on unsupported engines For example, this is added to the test for search_after:

onlyOn("opensearch has search_after but it works a bit different",
    SearchEngineVariant.ES7,
    SearchEngineVariant.ES8)

Module Overview

This repository contains several kotlin modules that each may be used independently.

Module Description
json-dsl Kotlin framework for creating kotlin DSLs for JSON dialects. Such as those in Elasticsearch.
search-dsls DSLs for search and mappings based on json-dsl.
search-client Multiplatform REST client for Elasticsearch 7 & 8 and Opensearch 1. This is what you would want to use in your projects.
docs Contains the code that generates the manual and this readme..

The search client module is the main module of this library. I extracted the json-dsl module and search-dsls module with the intention of eventually moving these to separate libraries. Json-dsl is actually useful for pretty much any kind of json dialect and I have a few APIs in mind where I might like to use it. The choice to not impose kotlinx.serialization on json dsl also means that both that and the search dsl are very portable and only depend on the Kotlin standard library.

Contributing

Pull requests are very welcome! Please consider communicating your intentions in advance to avoid conflicts, or redundant work.

Some suggestions of things you could work on:

Support and Community

Please file issues if you find any or have any suggestions for changes.

Within reason, I can help with simple issues. Beyond that, I offer my services as a consultant as well if you need some more help with getting started or just using Elasticsearch/Opensearch in general with just about any tech stack. I can help with discovery projects, trainings, architecture analysis, query and mapping optimizations, or just generally help you get the most out of your search setup and your product roadmap.

The best way to reach me is via email if you wish to use my services professionally. Please refer to my website for that.

I also try to respond quickly to issues. And I also lurk in the amazing Kotlin, Elastic, and Search Relevancy Slack communities.

About this README

This readme is generated using my kotlin4example library. I started developing that a few years ago when I realized that I was going to have to write a lot of documentation with code examples for kt-search. By now, both the manual and this readme heavily depend on this and it makes maintaining and adding documentation super easy.

The way it works is that it provides a dsl for writing markdown that you use to write documentation. It allows you to include runnable code blocks and when it builds the documentation it figures out how to extract those from the kotlin source files and adds them as markdown code snippets. It can also intercept printed output and the return values of the blocks.

If you have projects of your own that need documentation, you might get some value out of using this as well.