Table of Contents

Hadoop-based Batch Ingestion

Apache Hadoop-based batch ingestion in Apache Druid (incubating) is supported via a Hadoop-ingestion task. These tasks can be posted to a running instance of a Druid Overlord.

Please check Hadoop-based Batch Ingestion VS Native Batch Ingestion for differences between native batch ingestion and Hadoop-based ingestion.

Command Line Hadoop Indexer

If you don't want to use a full indexing service to use Hadoop to get data into Druid, you can also use the standalone command line Hadoop indexer. See here for more info.

Task syntax

A sample task is shown below:

  "type" : "index_hadoop",
  "spec" : {
    "dataSchema" : {
      "dataSource" : "wikipedia",
      "parser" : {
        "type" : "hadoopyString",
        "parseSpec" : {
          "format" : "json",
          "timestampSpec" : {
            "column" : "timestamp",
            "format" : "auto"
          "dimensionsSpec" : {
            "dimensions": ["page","language","user","unpatrolled","newPage","robot","anonymous","namespace","continent","country","region","city"],
            "dimensionExclusions" : [],
            "spatialDimensions" : []
      "metricsSpec" : [
          "type" : "count",
          "name" : "count"
          "type" : "doubleSum",
          "name" : "added",
          "fieldName" : "added"
          "type" : "doubleSum",
          "name" : "deleted",
          "fieldName" : "deleted"
          "type" : "doubleSum",
          "name" : "delta",
          "fieldName" : "delta"
      "granularitySpec" : {
        "type" : "uniform",
        "segmentGranularity" : "DAY",
        "queryGranularity" : "NONE",
        "intervals" : [ "2013-08-31/2013-09-01" ]
    "ioConfig" : {
      "type" : "hadoop",
      "inputSpec" : {
        "type" : "static",
        "paths" : "/MyDirectory/example/wikipedia_data.json"
    "tuningConfig" : {
      "type": "hadoop"
  "hadoopDependencyCoordinates": <my_hadoop_version>
property description required?
type The task type, this should always be "index_hadoop". yes
spec A Hadoop Index Spec. See Ingestion yes
hadoopDependencyCoordinates A JSON array of Hadoop dependency coordinates that Druid will use, this property will override the default Hadoop coordinates. Once specified, Druid will look for those Hadoop dependencies from the location specified by druid.extensions.hadoopDependenciesDir no
classpathPrefix Classpath that will be pre-appended for the Peon process. no

also note that, druid automatically computes the classpath for hadoop job containers that run in hadoop cluster. But, in case of conflicts between hadoop and druid's dependencies, you can manually specify the classpath by setting druid.extensions.hadoopContainerDruidClasspath property. See the extensions config in base druid configuration.


This field is required. See Ingestion Spec DataSchema.


This field is required.

Field Type Description Required
type String This should always be 'hadoop'. yes
inputSpec Object A specification of where to pull the data in from. See below. yes
segmentOutputPath String The path to dump segments into. Only used by the CLI Hadoop Indexer. This field must be null otherwise.
metadataUpdateSpec Object A specification of how to update the metadata for the druid cluster these segments belong to. Only used by the CLI Hadoop Indexer. This field must be null otherwise.

InputSpec specification

There are multiple types of inputSpecs:


A type of inputSpec where a static path to the data files is provided.

Field Type Description Required
inputFormat String Specifies the Hadoop InputFormat class to use. e.g. org.apache.hadoop.mapreduce.lib.input.SequenceFileInputFormat no
paths Array of String A String of input paths indicating where the raw data is located. yes

For example, using the static input paths:

"paths" : "s3n://billy-bucket/the/data/is/here/data.gz,s3n://billy-bucket/the/data/is/here/moredata.gz,s3n://billy-bucket/the/data/is/here/evenmoredata.gz"


A type of inputSpec that expects data to be organized in directories according to datetime using the path format: y=XXXX/m=XX/d=XX/H=XX/M=XX/S=XX (where date is represented by lowercase and time is represented by uppercase).

Field Type Description Required
dataGranularity String Specifies the granularity to expect the data at, e.g. hour means to expect directories y=XXXX/m=XX/d=XX/H=XX. yes
inputFormat String Specifies the Hadoop InputFormat class to use. e.g. org.apache.hadoop.mapreduce.lib.input.SequenceFileInputFormat no
inputPath String Base path to append the datetime path to. yes
filePattern String Pattern that files should match to be included. yes
pathFormat String Joda datetime format for each directory. Default value is "'y'=yyyy/'m'=MM/'d'=dd/'H'=HH", or see Joda documentation no

For example, if the sample config were run with the interval 2012-06-01/2012-06-02, it would expect data at the paths:



Read Druid segments. See here for more information.


Read multiple sources of data. See here for more information.


The tuningConfig is optional and default parameters will be used if no tuningConfig is specified.

Field Type Description Required
workingPath String The working path to use for intermediate results (results between Hadoop jobs). Only used by the CLI Hadoop Indexer. The default is '/tmp/druid-indexing'. This field must be null otherwise.
version String The version of created segments. Ignored for HadoopIndexTask unless useExplicitVersion is set to true no (default == datetime that indexing starts at)
partitionsSpec Object A specification of how to partition each time bucket into segments. Absence of this property means no partitioning will occur. See 'Partitioning specification' below. no (default == 'hashed')
maxRowsInMemory Integer The number of rows to aggregate before persisting. Note that this is the number of post-aggregation rows which may not be equal to the number of input events due to roll-up. This is used to manage the required JVM heap size. Normally user does not need to set this, but depending on the nature of data, if rows are short in terms of bytes, user may not want to store a million rows in memory and this value should be set. no (default == 1000000)
maxBytesInMemory Long The number of bytes to aggregate in heap memory before persisting. Normally this is computed internally and user does not need to set it. This is based on a rough estimate of memory usage and not actual usage. The maximum heap memory usage for indexing is maxBytesInMemory * (2 + maxPendingPersists). no (default == One-sixth of max JVM memory)
leaveIntermediate Boolean Leave behind intermediate files (for debugging) in the workingPath when a job completes, whether it passes or fails. no (default == false)
cleanupOnFailure Boolean Clean up intermediate files when a job fails (unless leaveIntermediate is on). no (default == true)
overwriteFiles Boolean Override existing files found during indexing. no (default == false)
ignoreInvalidRows Boolean DEPRECATED. Ignore rows found to have problems. If false, any exception encountered during parsing will be thrown and will halt ingestion; if true, unparseable rows and fields will be skipped. If maxParseExceptions is defined, this property is ignored. no (default == false)
combineText Boolean Use CombineTextInputFormat to combine multiple files into a file split. This can speed up Hadoop jobs when processing a large number of small files. no (default == false)
useCombiner Boolean Use Hadoop combiner to merge rows at mapper if possible. no (default == false)
jobProperties Object A map of properties to add to the Hadoop job configuration, see below for details. no (default == null)
indexSpec Object Tune how data is indexed. See below for more information. no
numBackgroundPersistThreads Integer The number of new background threads to use for incremental persists. Using this feature causes a notable increase in memory pressure and cpu usage but will make the job finish more quickly. If changing from the default of 0 (use current thread for persists), we recommend setting it to 1. no (default == 0)
forceExtendableShardSpecs Boolean Forces use of extendable shardSpecs. Hash-based partitioning always uses an extendable shardSpec. For single-dimension partitioning, this option should be set to true to use an extendable shardSpec. For partitioning, please check Partitioning specification. This option can be useful when you need to append more data to existing dataSource. no (default = false)
useExplicitVersion Boolean Forces HadoopIndexTask to use version. no (default = false)
logParseExceptions Boolean If true, log an error message when a parsing exception occurs, containing information about the row where the error occurred. false
maxParseExceptions Integer The maximum number of parse exceptions that can occur before the task halts ingestion and fails. Overrides ignoreInvalidRows if maxParseExceptions is defined. unlimited

jobProperties field of TuningConfig

   "tuningConfig" : {
     "type": "hadoop",
     "jobProperties": {
       "<hadoop-property-a>": "<value-a>",
       "<hadoop-property-b>": "<value-b>"

Hadoop's MapReduce documentation lists the possible configuration parameters.

With some Hadoop distributions, it may be necessary to set mapreduce.job.classpath or mapreduce.job.user.classpath.first to avoid class loading issues. See the working with different Hadoop versions documentation for more details.


Field Type Description Required
bitmap Object Compression format for bitmap indexes. Should be a JSON object; see below for options. no (defaults to Concise)
dimensionCompression String Compression format for dimension columns. Choose from LZ4, LZF, or uncompressed. no (default == LZ4)
metricCompression String Compression format for metric columns. Choose from LZ4, LZF, uncompressed, or none. no (default == LZ4)
longEncoding String Encoding format for metric and dimension columns with type long. Choose from auto or longs. auto encodes the values using offset or lookup table depending on column cardinality, and store them with variable size. longs stores the value as is with 8 bytes each. no (default == longs)

Bitmap types

For Concise bitmaps:

Field Type Description Required
type String Must be concise. yes

For Roaring bitmaps:

Field Type Description Required
type String Must be roaring. yes
compressRunOnSerialization Boolean Use a run-length encoding where it is estimated as more space efficient. no (default == true)

Partitioning specification

Segments are always partitioned based on timestamp (according to the granularitySpec) and may be further partitioned in some other way depending on partition type. Druid supports two types of partitioning strategies: "hashed" (based on the hash of all dimensions in each row), and "dimension" (based on ranges of a single dimension).

Hashed partitioning is recommended in most cases, as it will improve indexing performance and create more uniformly sized data segments relative to single-dimension partitioning.

Hash-based partitioning

  "partitionsSpec": {
     "type": "hashed",
     "targetPartitionSize": 5000000

Hashed partitioning works by first selecting a number of segments, and then partitioning rows across those segments according to the hash of all dimensions in each row. The number of segments is determined automatically based on the cardinality of the input set and a target partition size.

The configuration options are:

Field Description Required
type Type of partitionSpec to be used. "hashed"
targetPartitionSize Target number of rows to include in a partition, should be a number that targets segments of 500MB~1GB. either this or numShards
numShards Specify the number of partitions directly, instead of a target partition size. Ingestion will run faster, since it can skip the step necessary to select a number of partitions automatically. either this or targetPartitionSize
partitionDimensions The dimensions to partition on. Leave blank to select all dimensions. Only used with numShards, will be ignored when targetPartitionSize is set no

Single-dimension partitioning

  "partitionsSpec": {
     "type": "dimension",
     "targetPartitionSize": 5000000

Single-dimension partitioning works by first selecting a dimension to partition on, and then separating that dimension into contiguous ranges. Each segment will contain all rows with values of that dimension in that range. For example, your segments may be partitioned on the dimension "host" using the ranges "" to "" and "" to "". By default, the dimension to use is determined automatically, although you can override it with a specific dimension.

The configuration options are:

Field Description Required
type Type of partitionSpec to be used. "dimension"
targetPartitionSize Target number of rows to include in a partition, should be a number that targets segments of 500MB~1GB. yes
maxPartitionSize Maximum number of rows to include in a partition. Defaults to 50% larger than the targetPartitionSize. no
partitionDimension The dimension to partition on. Leave blank to select a dimension automatically. no
assumeGrouped Assume that input data has already been grouped on time and dimensions. Ingestion will run faster, but may choose sub-optimal partitions if this assumption is violated. no

Remote Hadoop Cluster

If you have a remote Hadoop cluster, make sure to include the folder holding your configuration *.xml files in your Druid _common configuration folder.

If you are having dependency problems with your version of Hadoop and the version compiled with Druid, please see these docs.

Using Elastic MapReduce

If your cluster is running on Amazon Web Services, you can use Elastic MapReduce (EMR) to index data from S3. To do this:

  • Create a persistent, long-running cluster.
  • When creating your cluster, enter the following configuration. If you're using the wizard, this should be in advanced mode under "Edit software settings":
classification=yarn-site,properties=[mapreduce.reduce.memory.mb=6144, -Xms2g -Xmx2g -Duser.timezone=UTC -Dfile.encoding=UTF-8 -XX:+PrintGCDetails -XX:+PrintGCTimeStamps,, -Xms512m -Xmx512m -Duser.timezone=UTC -Dfile.encoding=UTF-8 -XX:+PrintGCDetails -XX:+PrintGCTimeStamps,mapreduce.task.timeout=1800000]

Secured Hadoop Cluster

By default druid can use the exisiting TGT kerberos ticket available in local kerberos key cache. Although TGT ticket has a limited life cycle, therefore you need to call kinit command periodically to ensure validity of TGT ticket. To avoid this extra external cron job script calling kinit periodically, you can provide the principal name and keytab location and druid will do the authentication transparently at startup and job launching time.

Property Possible Values Description Default druid@EXAMPLE.COM Principal user name empty /etc/security/keytabs/druid.headlessUser.keytab Path to keytab file empty

Loading from S3 with EMR

  • In the jobProperties field in the tuningConfig section of your Hadoop indexing task, add:
"jobProperties" : {
   "fs.s3.awsAccessKeyId" : "YOUR_ACCESS_KEY",
   "fs.s3.awsSecretAccessKey" : "YOUR_SECRET_KEY",
   "fs.s3.impl" : "org.apache.hadoop.fs.s3native.NativeS3FileSystem",
   "fs.s3n.awsAccessKeyId" : "YOUR_ACCESS_KEY",
   "fs.s3n.awsSecretAccessKey" : "YOUR_SECRET_KEY",
   "fs.s3n.impl" : "org.apache.hadoop.fs.s3native.NativeS3FileSystem",
   "io.compression.codecs" : ",,,"

Note that this method uses Hadoop's built-in S3 filesystem rather than Amazon's EMRFS, and is not compatible with Amazon-specific features such as S3 encryption and consistent views. If you need to use these features, you will need to make the Amazon EMR Hadoop JARs available to Druid through one of the mechanisms described in the Using other Hadoop distributions section.

Using other Hadoop distributions

Druid works out of the box with many Hadoop distributions.

If you are having dependency conflicts between Druid and your version of Hadoop, you can try searching for a solution in the Druid user groups, or reading the Druid Different Hadoop Versions documentation.