ing-bank / scruid   2.5.0

Apache License 2.0 GitHub

Scala + Druid: Scruid. A library that allows you to compose queries in Scala, and parse the result back into typesafe classes.

Scala versions: 2.13 2.12 2.11

Build Codacy Badge

Scruid

Scruid (Scala+Druid) is an open source library that allows you to compose Druid queries easily in Scala. The library will take care of the translation of the query into json, parse the result in the case class that you define.

Currently, the API is under heavy development, so changes might occur.

Release Notes

Please view the Releases page on GitHub.

Installation

The binaries are hosted on Maven Central. We publish builds for Scala 2.11, 2.12 and 2.13.

libraryDependencies += "com.ing.wbaa.druid" %% "scruid" % "2.5.0"

Example queries:

Scruid provides query constructors for TopNQuery, GroupByQuery, TimeSeriesQuery, ScanQuery and SearchQuery (see below for details). You can call the execute method on a query to send the query to Druid. This will return a Future[DruidResponse]. This response contains the Circe JSON data without having it parsed to a specific case class yet. To interpret this JSON data you can run two methods on a DruidResponse:

  • .list[T](implicit decoder: Decoder[T]): List[T] : This decodes the JSON to a list with items of type T.
  • .series[T](implicit decoder: Decoder[T]): Map[ZonedDateTime, T] : This decodes the JSON to a timeseries map with the timestamp as key and T as value.

Below the example queries supported by Scruid. For more information about how to query Druid, and what query to pick, please refer to the Druid documentation

TopN query

case class TopCountry(count: Int, countryName: String = null)

val response = TopNQuery(
  dimension = Dimension(
    dimension = "countryName"
  ),
  threshold = 5,
  metric = "count",
  aggregations = List(
    CountAggregation(name = "count")
  ),
  intervals = List("2011-06-01/2017-06-01")
).execute

val result: Future[Map[ZonedDateTime, List[TopCountry]]] = response.map(_.series[List[TopCountry]])

GroupBy query

case class GroupByIsAnonymous(isAnonymous: Boolean, count: Int)

val response = GroupByQuery(
  aggregations = List(
    CountAggregation(name = "count")
  ),
  dimensions = List("isAnonymous"),
  intervals = List("2011-06-01/2017-06-01")
).execute()

val result: Future[List[GroupByIsAnonymous]] = response.map(_.list[GroupByIsAnonymous])

The returned Future[DruidResponse] will contain json data where isAnonymouse is either true or false. Please keep in mind that Druid is only able to handle strings, and recently also numerics. So Druid will be returning a string, and the conversion from a string to a boolean is done by the json parser.

TimeSeries query

case class TimeseriesCount(count: Int)

val response = TimeSeriesQuery(
  aggregations = List(
    CountAggregation(name = "count")
  ),
  granularity = GranularityType.Hour,
  intervals = List("2011-06-01/2017-06-01")
).execute

val series: Future[Map[ZonedDateTime, TimeseriesCount]] = response.map(_.series[TimeseriesCount])

Scan query

case class ScanResult(channel: Option[String], cityName: Option[String], countryIsoCode: Option[String], user: Option[String])

val response = ScanQuery(
    granularity = GranularityType.Hour
    intervals = List("2011-06-01/2017-06-01")
    dimensions = List("channel", "cityName", "countryIsoCode", "user"),
    limit = 100
).execute() 

val result: Future[List[ScanResult]] = response.map(_.list[ScanResult])

Search query

Search query is a bit different, since it does not take type parameters as its results are of type com.ing.wbaa.druid.DruidSearchResult

val response = SearchQuery(
    granularity = GranularityType.Hour,
    intervals = List("2011-06-01/2017-06-01"),
    query = ContainsInsensitive("GR"),
    searchDimensions = List("countryIsoCode")
).execute()

val result = Future[List[DruidSearchResult]] = response.map(_.list)

Query context

Queries can be configured using Druid query context, such as timeout, queryId and groupByStrategy. All types of query contain the argument context which associates query parameter with their corresponding values. The parameter names can also be accessed by com.ing.wbaa.druid.definitions.QueryContext object. Consider, for example, a timeseries query with custom query id and priority:

TimeSeriesQuery(
  aggregations = List(
    CountAggregation(name = "count")
  ),
  granularity = GranularityType.Hour,
  intervals = List("2011-06-01/2017-06-01"),
  context = Map(
    QueryContext.QueryId -> "some_custom_id",
    QueryContext.Priority -> 1
  )
)

Druid query language (DQL)

Scruid also provides a rich Scala API for building queries using the fluent pattern.

case class GroupByIsAnonymous(isAnonymous: String, country: String, count: Int)

val query: GroupByQuery = DQL
    .granularity(GranularityType.Day)
    .interval("2011-06-01/2017-06-01")
    .agg(count as "count")
    .where(d"countryName".isNotNull)
    .groupBy(d"isAnonymous", d"countryName".extract(UpperExtractionFn()) as "country")
    .having(d"count" > 100 and d"count" < 200)
    .limit(10, d"count".desc(DimensionOrderType.Numeric))
    .build()

val response: Future[List[GroupByIsAnonymous]] = query.execute().map(_.list[GroupByIsAnonymous])

For details and examples see the DQL documentation.

Print native Druid JSON representation

For all types of queries you can call the function toDebugString, in order to get the corresponding native Druid JSON query representation.

For example the following:

import com.ing.wbaa.druid.dql.DSL._

val query: TopNQuery = DQL
    .from("wikipedia")
    .agg(count as "count")
    .interval("2011-06-01/2017-06-01")
    .topN(dimension = d"countryName", metric = "count", threshold = 5)
    .build()

println(query.toDebugString)

will print to the standard output:

{
  "dimension" : {
    "dimension" : "countryName",
    "outputName" : "countryName",
    "outputType" : null,
    "type" : "default"
  },
  "threshold" : 5,
  "metric" : "count",
  "aggregations" : [
    {
      "name" : "count",
      "type" : "count"
    }
  ],
  "intervals" : [
    "2011-06-01/2017-06-01"
  ],
  "granularity" : "all",
  "filter" : null,
  "postAggregations" : [
  ],
  "context" : {

  },
  "queryType" : "topN",
  "dataSource" : "wikipedia"
}

Handling large payloads with Akka Streams

For queries with large payload of results (e.g., half a million of records), Scruid can transform the corresponding response into an Akka Stream Source. The results can be processed, filtered and transformed using Flows and/or output to Sinks, as a continuous stream, without collecting the entire payload first. To process the results with Akka Stream, you can call one of the following methods:

  • .stream: gives a Source of DruidResult.
  • .streamAs[T](implicit decoder: Decoder[T]): gives a Source where each JSON record is being decoded to the type of T.
  • .streamSeriesAs[T](implicit decoder: Decoder[T]): gives a Source where each JSON record is being decoded to the type of T and it is accompanied by its corresponding timestamp.

All the methods above can be applied to any timeseries, group-by or top-N query created either directly by using query constructors or by DQL.

Druid SQL support

Instead of using the Druid native API, Scruid also supports Druid queries via SQL.

import com.ing.wbaa.druid.SQL._

val query = dsql"""SELECT COUNT(*) as "count" FROM wikipedia WHERE "__time" >= TIMESTAMP '2015-09-12 00:00:00'"""

val response = query.execute()

For details see the SQL documentation.

Example

implicit val mat = DruidClient.materializer

case class TimeseriesCount(count: Int)

val query = TimeSeriesQuery(
  aggregations = List(
    CountAggregation(name = "count")
  ),
  granularity = GranularityType.Hour,
  intervals = List("2011-06-01/2017-06-01")
)

// Decode each record into the type of `TimeseriesCount` and sum all `count` results
val result: Future[Int] = query
        .streamAs[TimeseriesCount]
        .map(_.count)
        .runWith(Sink.fold(0)(_ + _))

Configuration

The configuration is done by Typesafe config. The configuration can be overridden by using environment variables, e.g. DRUID_HOSTS (DRUID_HOST and DRUID_PORT are still supported for backward compatibility) and DRUID_DATASOURCE. Or by placing an application.conf in your own project and this will override the reference.conf of the scruid library.

druid = {
  host = "localhost"
  host = ${?DRUID_HOST}
  port = 8082
  port = ${?DRUID_PORT}
  hosts = ${druid.host}":"${druid.port}
  hosts = ${?DRUID_HOSTS}
  secure = false
  secure = ${?DRUID_USE_SECURE_CONNECTION}
  url = "/druid/v2/"
  url = ${?DRUID_URL}
  health-endpoint = "/status/health"
  health-endpoint = ${?DRUID_HEALTH_ENDPOINT}
  client-backend = "com.ing.wbaa.druid.client.DruidHttpClient"
  client-backend = ${?DRUID_CLIENT_BACKEND}

  scan-query-legacy-mode = false
  scan-query-legacy-mode = ${?DRUID_SCAN_QUERY_LEGACY_MODE}

  datasource = "wikipedia"
  datasource = ${?DRUID_DATASOURCE}

  response-parsing-timeout = 5 seconds
  response-parsing-timeout = ${?DRUID_RESPONSE_PARSING_TIMEOUT}

  zone-id = "UTC"
}

Alternatively it can be programmatically overridden by defining an implicit instance of com.ing.wbaa.druid.DruidConfig:

import java.time.ZonedDateTime
import com.ing.wbaa.druid._
import com.ing.wbaa.druid.definitions._
import scala.concurrent.duration._


implicit val druidConf = DruidConfig(
  hosts = Seq("localhost:8082"),
  datasource = "wikipedia",
  responseParsingTimeout = 10.seconds
)

case class TimeseriesCount(count: Int)

val response = TimeSeriesQuery(
  aggregations = List(
    CountAggregation(name = "count")
  ),
  granularity = GranularityType.Week,
  intervals = List("2011-06-01/2017-06-01")
).execute

val series: Map[ZonedDateTime, TimeseriesCount] = response.series[TimeseriesCount]

All parameters of DruidConfig are optional, and in case that some parameter is missing then the default behaviour is to use the value that is defined in the configuration file.

Druid Clients

Scruid provides two client implementations, one for simple requests over a single Druid query host (default) and an advanced one with a queue, cached pool connections and, a load balancer when multiple Druid query hosts are provided. Depending on your use case, it is also possible to create a custom client. For details regarding clients, their configuration, as well the creation of a custom one see the Scruid Clients documentation.

Authentication

The Advanced client can be configured to authenticate with the Druid cluster. See the Scruid Clients document for more information.

Tests

The test suite relies on a docker-compose with supporting services. The dockerfiles for the images it uses are in the docker/ subdirectory. Dependency versions of the dockerized resources are defined in ./env.

To run the tests, please make sure that you have the Druid instance running:

./services.sh start

This command will build the local images as needed. You can manually build these using the ./services.sh build_images command or the Makefile in ./docker.