Source input formats

Apache Druid can ingest denormalized data in JSON, CSV, or a delimited form such as TSV, or any custom format. While most examples in the documentation use data in JSON format, it is not difficult to configure Druid to ingest any other delimited data. We welcome any contributions to new formats.

This page lists all default and core extension data formats supported by Druid. For additional data formats supported with community extensions, please see our community extensions list.

Formatting data

The following samples show data formats that are natively supported in Druid:

JSON

{"timestamp": "2013-08-31T01:02:33Z", "page": "Gypsy Danger", "language" : "en", "user" : "nuclear", "unpatrolled" : "true", "newPage" : "true", "robot": "false", "anonymous": "false", "namespace":"article", "continent":"North America", "country":"United States", "region":"Bay Area", "city":"San Francisco", "added": 57, "deleted": 200, "delta": -143}
{"timestamp": "2013-08-31T03:32:45Z", "page": "Striker Eureka", "language" : "en", "user" : "speed", "unpatrolled" : "false", "newPage" : "true", "robot": "true", "anonymous": "false", "namespace":"wikipedia", "continent":"Australia", "country":"Australia", "region":"Cantebury", "city":"Syndey", "added": 459, "deleted": 129, "delta": 330}
{"timestamp": "2013-08-31T07:11:21Z", "page": "Cherno Alpha", "language" : "ru", "user" : "masterYi", "unpatrolled" : "false", "newPage" : "true", "robot": "true", "anonymous": "false", "namespace":"article", "continent":"Asia", "country":"Russia", "region":"Oblast", "city":"Moscow", "added": 123, "deleted": 12, "delta": 111}
{"timestamp": "2013-08-31T11:58:39Z", "page": "Crimson Typhoon", "language" : "zh", "user" : "triplets", "unpatrolled" : "true", "newPage" : "false", "robot": "true", "anonymous": "false", "namespace":"wikipedia", "continent":"Asia", "country":"China", "region":"Shanxi", "city":"Taiyuan", "added": 905, "deleted": 5, "delta": 900}
{"timestamp": "2013-08-31T12:41:27Z", "page": "Coyote Tango", "language" : "ja", "user" : "cancer", "unpatrolled" : "true", "newPage" : "false", "robot": "true", "anonymous": "false", "namespace":"wikipedia", "continent":"Asia", "country":"Japan", "region":"Kanto", "city":"Tokyo", "added": 1, "deleted": 10, "delta": -9}

CSV

2013-08-31T01:02:33Z,"Gypsy Danger","en","nuclear","true","true","false","false","article","North America","United States","Bay Area","San Francisco",57,200,-143
2013-08-31T03:32:45Z,"Striker Eureka","en","speed","false","true","true","false","wikipedia","Australia","Australia","Cantebury","Syndey",459,129,330
2013-08-31T07:11:21Z,"Cherno Alpha","ru","masterYi","false","true","true","false","article","Asia","Russia","Oblast","Moscow",123,12,111
2013-08-31T11:58:39Z,"Crimson Typhoon","zh","triplets","true","false","true","false","wikipedia","Asia","China","Shanxi","Taiyuan",905,5,900
2013-08-31T12:41:27Z,"Coyote Tango","ja","cancer","true","false","true","false","wikipedia","Asia","Japan","Kanto","Tokyo",1,10,-9

TSV (Delimited)

2013-08-31T01:02:33Z  "Gypsy Danger"  "en"  "nuclear" "true"  "true"  "false" "false" "article" "North America" "United States" "Bay Area"  "San Francisco" 57  200 -143
2013-08-31T03:32:45Z  "Striker Eureka"  "en"  "speed" "false" "true"  "true"  "false" "wikipedia" "Australia" "Australia" "Cantebury" "Syndey"  459 129 330
2013-08-31T07:11:21Z  "Cherno Alpha"  "ru"  "masterYi"  "false" "true"  "true"  "false" "article" "Asia"  "Russia"  "Oblast"  "Moscow"  123 12  111
2013-08-31T11:58:39Z  "Crimson Typhoon" "zh"  "triplets"  "true"  "false" "true"  "false" "wikipedia" "Asia"  "China" "Shanxi"  "Taiyuan" 905 5 900
2013-08-31T12:41:27Z  "Coyote Tango"  "ja"  "cancer"  "true"  "false" "true"  "false" "wikipedia" "Asia"  "Japan" "Kanto" "Tokyo" 1 10  -9

Note that the CSV and TSV data do not contain column heads. This becomes important when you specify the data for ingesting.

Besides text formats, Druid also supports binary formats such as Orc and Parquet formats.

Custom formats

Druid supports custom data formats and can use the Regex parser or the JavaScript parsers to parse these formats. Using any of these parsers for parsing data is less efficient than writing a native Java parser or using an external stream processor. We welcome contributions of new parsers.

Input format

You can use the inputFormat field to specify the data format for your input data.

info

inputFormat doesn’t support all data formats or ingestion methods supported by Druid.

Especially if you want to use the Hadoop ingestion, you still need to use the Parser. If your data is formatted in some format not listed in this section, please consider using the Parser instead.

All forms of Druid ingestion require some form of schema object. The format of the data to be ingested is specified using the inputFormat entry in your ioConfig.

JSON

Configure the JSON inputFormat to load JSON data as follows:

The following properties are specialized properties that only apply when the JSON inputFormat is used in streaming ingestion, and they are related to how parsing exceptions are handled. In streaming ingestion, multi-line JSON events can be ingested (i.e. where a single JSON event spans multiple lines). However, if a parsing exception occurs, all JSON events that are present in the same streaming record will be discarded.

For example:

"ioConfig": {
  "inputFormat": {
    "type": "json"
  },
  ...
}

CSV

Configure the CSV inputFormat to load CSV data as follows:

For example:

"ioConfig": {
  "inputFormat": {
    "type": "csv",
    "columns" : ["timestamp","page","language","user","unpatrolled","newPage","robot","anonymous","namespace","continent","country","region","city","added","deleted","delta"]
  },
  ...
}

TSV (Delimited)

Configure the TSV inputFormat to load TSV data as follows:

Be sure to change the delimiter to the appropriate delimiter for your data. Like CSV, you must specify the columns and which subset of the columns you want indexed.

For example:

"ioConfig": {
  "inputFormat": {
    "type": "tsv",
    "columns" : ["timestamp","page","language","user","unpatrolled","newPage","robot","anonymous","namespace","continent","country","region","city","added","deleted","delta"],
    "delimiter":"|"
  },
  ...
}

ORC

To use the ORC input format, load the Druid Orc extension ( druid-orc-extensions).

info

To upgrade from versions earlier than 0.15.0 to 0.15.0 or new, read Migration from ‘contrib’ extension.

Configure the ORC inputFormat to load ORC data as follows:

For example:

"ioConfig": {
  "inputFormat": {
    "type": "orc",
    "flattenSpec": {
      "useFieldDiscovery": true,
      "fields": [
        {
          "type": "path",
          "name": "nested",
          "expr": "$.path.to.nested"
        }
      ]
    },
    "binaryAsString": false
  },
  ...
}

Parquet

To use the Parquet input format load the Druid Parquet extension (druid-parquet-extensions).

Configure the Parquet inputFormat to load Parquet data as follows:

For example:

"ioConfig": {
  "inputFormat": {
    "type": "parquet",
    "flattenSpec": {
      "useFieldDiscovery": true,
      "fields": [
        {
          "type": "path",
          "name": "nested",
          "expr": "$.path.to.nested"
        }
      ]
    },
    "binaryAsString": false
  },
  ...
}

Avro Stream

To use the Avro Stream input format load the Druid Avro extension (druid-avro-extensions).

For more information on how Druid handles Avro types, see Avro Types section for

Configure the Avro inputFormat to load Avro data as follows:

For example:

"ioConfig": {
  "inputFormat": {
    "type": "avro_stream",
    "avroBytesDecoder": {
      "type": "schema_inline",
      "schema": {
        //your schema goes here, for example
        "namespace": "org.apache.druid.data",
        "name": "User",
        "type": "record",
        "fields": [
          { "name": "FullName", "type": "string" },
          { "name": "Country", "type": "string" }
        ]
      }
    },
    "flattenSpec": {
      "useFieldDiscovery": true,
      "fields": [
        {
          "type": "path",
          "name": "someRecord_subInt",
          "expr": "$.someRecord.subInt"
        }
      ]
    },
    "binaryAsString": false
  },
  ...
}

Avro Bytes Decoder

If type is not included, the avroBytesDecoder defaults to schema_repo.

Inline Schema Based Avro Bytes Decoder

info

The “schema_inline” decoder reads Avro records using a fixed schema and does not support schema migration. If you may need to migrate schemas in the future, consider one of the other decoders, all of which use a message header that allows the parser to identify the proper Avro schema for reading records.

This decoder can be used if all the input events can be read using the same schema. In this case, specify the schema in the input task JSON itself, as described below.

...
"avroBytesDecoder": {
  "type": "schema_inline",
  "schema": {
    //your schema goes here, for example
    "namespace": "org.apache.druid.data",
    "name": "User",
    "type": "record",
    "fields": [
      { "name": "FullName", "type": "string" },
      { "name": "Country", "type": "string" }
    ]
  }
}
...

Multiple Inline Schemas Based Avro Bytes Decoder

Use this decoder if different input events can have different read schemas. In this case, specify the schema in the input task JSON itself, as described below.

...
"avroBytesDecoder": {
  "type": "multiple_schemas_inline",
  "schemas": {
    //your id -> schema map goes here, for example
    "1": {
      "namespace": "org.apache.druid.data",
      "name": "User",
      "type": "record",
      "fields": [
        { "name": "FullName", "type": "string" },
        { "name": "Country", "type": "string" }
      ]
    },
    "2": {
      "namespace": "org.apache.druid.otherdata",
      "name": "UserIdentity",
      "type": "record",
      "fields": [
        { "name": "Name", "type": "string" },
        { "name": "Location", "type": "string" }
      ]
    },
    ...
    ...
  }
}
...

Note that it is essentially a map of integer schema ID to avro schema object. This parser assumes that record has following format. first 1 byte is version and must always be 1. next 4 bytes are integer schema ID serialized using big-endian byte order. remaining bytes contain serialized avro message.

SchemaRepo Based Avro Bytes Decoder

This Avro bytes decoder first extracts subject and id from the input message bytes, and then uses them to look up the Avro schema used to decode the Avro record from bytes. For details, see the schema repo. You need an HTTP service like schema repo to hold the Avro schema. For information on registering a schema on the message producer side, see org.apache.druid.data.input.AvroStreamInputRowParserTest#testParse().

Avro-1124 Subject And Id Converter

This section describes the format of the subjectAndIdConverter object for the schema_repo Avro bytes decoder.

Avro-1124 Schema Repository

This section describes the format of the schemaRepository object for the schema_repo Avro bytes decoder.

Confluent Schema Registry-based Avro Bytes Decoder

This Avro bytes decoder first extracts a unique id from input message bytes, and then uses it to look up the schema in the Schema Registry used to decode the Avro record from bytes. For details, see the Schema Registry documentation and repository.

For a single schema registry instance, use Field url or urls for multi instances.

Single Instance:

...
"avroBytesDecoder" : {
   "type" : "schema_registry",
   "url" : <schema-registry-url>
}
...

Multiple Instances:

...
"avroBytesDecoder" : {
   "type" : "schema_registry",
   "urls" : [<schema-registry-url-1>, <schema-registry-url-2>, ...],
   "config" : {
        "basic.auth.credentials.source": "USER_INFO",
        "basic.auth.user.info": "fred:letmein",
        "schema.registry.ssl.truststore.location": "/some/secrets/kafka.client.truststore.jks",
        "schema.registry.ssl.truststore.password": "<password>",
        "schema.registry.ssl.keystore.location": "/some/secrets/kafka.client.keystore.jks",
        "schema.registry.ssl.keystore.password": "<password>",
        "schema.registry.ssl.key.password": "<password>",
        "schema.registry.ssl.key.password",
       ... 
   },
   "headers": {
       "traceID" : "b29c5de2-0db4-490b-b421",
       "timeStamp" : "1577191871865",
       "druid.dynamic.config.provider":{
            "type":"mapString", 
            "config":{
                 "registry.header.prop.1":"value.1", 
                 "registry.header.prop.2":"value.2"
                 }
            }
       ...
    }
}
...

Parse exceptions

The following errors when reading records will be considered parse exceptions, which can be limited and logged with ingestion task configurations such as maxParseExceptions and maxSavedParseExceptions:

• Failure to retrieve a schema due to misconfiguration or corrupt records (invalid schema IDs)
• Failure to decode an Avro message

Avro OCF

To load the Avro OCF input format, load the Druid Avro extension (druid-avro-extensions).

See the Avro Types section for how Avro types are handled in Druid

Configure the Avro OCF inputFormat to load Avro OCF data as follows:

For example:

"ioConfig": {
  "inputFormat": {
    "type": "avro_ocf",
    "flattenSpec": {
      "useFieldDiscovery": true,
      "fields": [
        {
          "type": "path",
          "name": "someRecord_subInt",
          "expr": "$.someRecord.subInt"
        }
      ]
    },
    "schema": {
      "namespace": "org.apache.druid.data.input",
      "name": "SomeDatum",
      "type": "record",
      "fields" : [
        { "name": "timestamp", "type": "long" },
        { "name": "eventType", "type": "string" },
        { "name": "id", "type": "long" },
        { "name": "someRecord", "type": {
          "type": "record", "name": "MySubRecord", "fields": [
            { "name": "subInt", "type": "int"},
            { "name": "subLong", "type": "long"}
          ]
        }}]
    },
    "binaryAsString": false
  },
  ...
}

Protobuf

info

You need to include the druid-protobuf-extensions as an extension to use the Protobuf input format.

Configure the Protobuf inputFormat to load Protobuf data as follows:

For example:

"ioConfig": {
  "inputFormat": {
    "type": "protobuf",
    "protoBytesDecoder": {
      "type": "file",
      "descriptor": "file:///tmp/metrics.desc",
      "protoMessageType": "Metrics"
    }
    "flattenSpec": {
      "useFieldDiscovery": true,
      "fields": [
        {
          "type": "path",
          "name": "someRecord_subInt",
          "expr": "$.someRecord.subInt"
        }
      ]
    }
  },
  ...
}

Kafka

The kafka input format lets you parse the Kafka metadata fields in addition to the Kafka payload value contents. It should only be used when ingesting from Apache Kafka.

The kafka input format wraps around the payload parsing input format and augments the data it outputs with the Kafka event timestamp, topic name, event headers, and the key field that itself can be parsed using any available input format.

If there are conflicts between column names in the payload and those created from the metadata, the payload takes precedence. This ensures that upgrading a Kafka ingestion to use the Kafka input format (by taking its existing input format and setting it as the valueFormat) can be done without losing any of the payload data.

Configure the Kafka inputFormat as follows:

Header format

headerFormat supports the following encoding formats:

• ISO-8859-1: ISO Latin Alphabet No. 1, that is, ISO-LATIN-1.
• US-ASCII: Seven-bit ASCII. Also known as ISO646-US. The Basic Latin block of the Unicode character set.
• UTF-8: Eight-bit UCS Transformation Format.
• UTF-16: Sixteen-bit UCS Transformation Format, byte order identified by an optional byte-order mark.
• UTF-16BE: Sixteen-bit UCS Transformation Format, big-endian byte order.
• UTF-16LE: Sixteen-bit UCS Transformation Format, little-endian byte order.
• headerColumnPrefix: Supply a prefix to the Kafka headers to avoid any conflicts with columns from the payload. The default is kafka.header..

Example

Using { "type": "json" } as the input format would only parse the payload value. To parse the Kafka metadata in addition to the payload, use the kafka input format.

For example, consider the following structure for a Kafka message that represents an edit in a development environment:

• Kafka timestamp: 1680795276351
• Kafka topic: wiki-edits
• Kafka headers:
  • env=development
  • zone=z1
• Kafka key: wiki-edit
• Kafka payload value: {"channel":"#sv.wikipedia","timestamp":"2016-06-27T00:00:11.080Z","page":"Salo Toraut","delta":31,"namespace":"Main"}

You would configure it as follows:

"ioConfig": {
  "inputFormat": {
    "type": "kafka",
    "valueFormat": {
      "type": "json"
    },
    "timestampColumnName": "kafka.timestamp",
    "topicColumnName": "kafka.topic",
    "headerFormat": {
      "type": "string",
      "encoding": "UTF-8"
    },
    "headerColumnPrefix": "kafka.header.",
    "keyFormat": {
      "type": "tsv",
      "findColumnsFromHeader": false,
      "columns": ["x"]
    },
    "keyColumnName": "kafka.key",
  }
}

You would parse the example message as follows:

{
  "channel": "#sv.wikipedia",
  "timestamp": "2016-06-27T00:00:11.080Z",
  "page": "Salo Toraut",
  "delta": 31,
  "namespace": "Main",
  "kafka.timestamp": 1680795276351,
  "kafka.topic": "wiki-edits",
  "kafka.header.env": "development",
  "kafka.header.zone": "z1",
  "kafka.key": "wiki-edit"
}

If you want to use kafka.timestamp as Druid’s primary timestamp (__time), specify it as the value for column in the timestampSpec:

"timestampSpec": {
  "column": "kafka.timestamp",
  "format": "millis"
}

Similarly, if you want to use a timestamp extracted from the Kafka header:

"timestampSpec": {
  "column": "kafka.header.myTimestampHeader",
  "format": "millis"
}

Finally, add these Kafka metadata columns to the dimensionsSpec or set your dimensionsSpec to auto-detect columns.

The following supervisor spec demonstrates how to ingest the Kafka header, key, timestamp, and topic into Druid dimensions:

Click to view the example

{
  "type": "kafka",
  "spec": {
    "ioConfig": {
      "type": "kafka",
      "consumerProperties": {
        "bootstrap.servers": "localhost:9092"
      },
      "topic": "wiki-edits",
      "inputFormat": {
        "type": "kafka",
        "valueFormat": {
          "type": "json"
        },
        "headerFormat": {
          "type": "string"
        },
        "keyFormat": {
          "type": "tsv",
          "findColumnsFromHeader": false,
          "columns": ["x"]
        }
      },
      "useEarliestOffset": true
    },
    "dataSchema": {
      "dataSource": "wikiticker",
      "timestampSpec": {
        "column": "timestamp",
        "format": "posix"
      },
      "dimensionsSpec":  "dimensionsSpec": {
        "useSchemaDiscovery": true,
        "includeAllDimensions": true
      },
      "granularitySpec": {
        "queryGranularity": "none",
        "rollup": false,
        "segmentGranularity": "day"
      }
    },
    "tuningConfig": {
      "type": "kafka"
    }
  }
}

After Druid ingests the data, you can query the Kafka metadata columns as follows:

SELECT
  "kafka.header.env",
  "kafka.key",
  "kafka.timestamp",
  "kafka.topic"
FROM "wikiticker"

This query returns:

FlattenSpec

You can use the flattenSpec object to flatten nested data, as an alternative to the Druid nested columns feature, and for nested input formats unsupported by the feature. It is an object within the inputFormat object.

See Nested columns for information on ingesting and storing nested data in an Apache Druid column as a COMPLEX<json> data type.

Configure your flattenSpec as follows:

For example:

"flattenSpec": {
  "useFieldDiscovery": true,
  "fields": [
    { "name": "baz", "type": "root" },
    { "name": "foo_bar", "type": "path", "expr": "$.foo.bar" },
    { "name": "foo_other_bar", "type": "tree", "nodes": ["foo", "other", "bar"] },
    { "name": "first_food", "type": "jq", "expr": ".thing.food[1]" }
  ]
}

After Druid reads the input data records, it applies the flattenSpec before applying any other specs such as timestampSpec, transformSpec, dimensionsSpec, or metricsSpec. This makes it possible to extract timestamps from flattened data, for example, and to refer to flattened data in transformations, in your dimension list, and when generating metrics.

Flattening is only supported for data formats that support nesting, including avro, json, orc, and parquet.

Field flattening specifications

Each entry in the fields list can have the following components:

Notes on flattening

• For convenience, when defining a root-level field, it is possible to define only the field name, as a string, instead of a JSON object. For example, {"name": "baz", "type": "root"} is equivalent to "baz".

• Enabling useFieldDiscovery will only automatically detect “simple” fields at the root level that correspond to data types that Druid supports. This includes strings, numbers, and lists of strings or numbers. Other types will not be automatically detected, and must be specified explicitly in the fields list.

• Duplicate field names are not allowed. An exception will be thrown.

• If useFieldDiscovery is enabled, any discovered field with the same name as one already defined in the fields list will be skipped, rather than added twice.

• JSONPath evaluator is useful for testing path-type expressions.

• jackson-jq supports a subset of the full jq syntax. Please refer to the jackson-jq documentation for details.

• JsonPath supports a bunch of functions, but not all of these functions are supported by Druid now. Following matrix shows the current supported JsonPath functions and corresponding data formats. Please also note the output data type of these functions.

  Function	Description	Output type	json	orc	avro	parquet
  min()	Provides the min value of an array of numbers	Double	✓	✓	✓	✓
  max()	Provides the max value of an array of numbers	Double	✓	✓	✓	✓
  avg()	Provides the average value of an array of numbers	Double	✓	✓	✓	✓
  stddev()	Provides the standard deviation value of an array of numbers	Double	✓	✓	✓	✓
  length()	Provides the length of an array	Integer	✓	✓	✓	✓
  sum()	Provides the sum value of an array of numbers	Double	✓	✓	✓	✓
  concat(X)	Provides a concatenated version of the path output with a new item	like input	✓	✗	✗	✗
  append(X)	add an item to the json path output array	like input	✓	✗	✗	✗
  keys()	Provides the property keys (An alternative for terminal tilde ~)	Set<E>	✗	✗	✗	✗

Parser

info

The Parser is deprecated for native batch tasks, Kafka indexing service, and Kinesis indexing service. Consider using the input format instead for these types of ingestion.

This section lists all default and core extension parsers. For community extension parsers, please see our community extensions list.

String Parser

string typed parsers operate on text based inputs that can be split into individual records by newlines. Each line can be further parsed using parseSpec.

Avro Hadoop Parser

info

You need to include the druid-avro-extensions as an extension to use the Avro Hadoop Parser.

info

See the Avro Types section for how Avro types are handled in Druid

This parser is for Hadoop batch ingestion. The inputFormat of inputSpec in ioConfig must be set to "org.apache.druid.data.input.avro.AvroValueInputFormat". You may want to set Avro reader’s schema in jobProperties in tuningConfig, e.g.: "avro.schema.input.value.path": "/path/to/your/schema.avsc" or "avro.schema.input.value": "your_schema_JSON_object". If the Avro reader’s schema is not set, the schema in Avro object container file will be used. See Avro specification for more information.

An Avro parseSpec can contain a flattenSpec using either the “root” or “path” field types, which can be used to read nested Avro records. The “jq” and “tree” field type is not currently supported for Avro.

For example, using Avro Hadoop parser with custom reader’s schema file:

{
  "type" : "index_hadoop",
  "spec" : {
    "dataSchema" : {
      "dataSource" : "",
      "parser" : {
        "type" : "avro_hadoop",
        "parseSpec" : {
          "format": "avro",
          "timestampSpec": <standard timestampSpec>,
          "dimensionsSpec": <standard dimensionsSpec>,
          "flattenSpec": <optional>
        }
      }
    },
    "ioConfig" : {
      "type" : "hadoop",
      "inputSpec" : {
        "type" : "static",
        "inputFormat": "org.apache.druid.data.input.avro.AvroValueInputFormat",
        "paths" : ""
      }
    },
    "tuningConfig" : {
       "jobProperties" : {
          "avro.schema.input.value.path" : "/path/to/my/schema.avsc"
      }
    }
  }
}

ORC Hadoop Parser

info

You need to include the druid-orc-extensions as an extension to use the ORC Hadoop Parser.

info

If you are considering upgrading from earlier than 0.15.0 to 0.15.0 or a higher version, please read Migration from ‘contrib’ extension carefully.

This parser is for Hadoop batch ingestion. The inputFormat of inputSpec in ioConfig must be set to "org.apache.orc.mapreduce.OrcInputFormat".

The parser supports two parseSpec formats: orc and timeAndDims.

orc supports auto field discovery and flattening, if specified with a flattenSpec. If no flattenSpec is specified, useFieldDiscovery will be enabled by default. Specifying a dimensionSpec is optional if useFieldDiscovery is enabled: if a dimensionSpec is supplied, the list of dimensions it defines will be the set of ingested dimensions, if missing the discovered fields will make up the list.

timeAndDims parse spec must specify which fields will be extracted as dimensions through the dimensionSpec.

All column types are supported, with the exception of union types. Columns of list type, if filled with primitives, may be used as a multi-value dimension, or specific elements can be extracted with flattenSpec expressions. Likewise, primitive fields may be extracted from map and struct types in the same manner. Auto field discovery will automatically create a string dimension for every (non-timestamp) primitive or list of primitives, as well as any flatten expressions defined in the flattenSpec.

Hadoop job properties

Like most Hadoop jobs, the best outcomes will add "mapreduce.job.user.classpath.first": "true" or "mapreduce.job.classloader": "true" to the jobProperties section of tuningConfig. Note that it is likely if using "mapreduce.job.classloader": "true" that you will need to set mapreduce.job.classloader.system.classes to include -org.apache.hadoop.hive. to instruct Hadoop to load org.apache.hadoop.hive classes from the application jars instead of system jars, e.g.

...
    "mapreduce.job.classloader": "true",
    "mapreduce.job.classloader.system.classes" : "java., javax.accessibility., javax.activation., javax.activity., javax.annotation., javax.annotation.processing., javax.crypto., javax.imageio., javax.jws., javax.lang.model., -javax.management.j2ee., javax.management., javax.naming., javax.net., javax.print., javax.rmi., javax.script., -javax.security.auth.message., javax.security.auth., javax.security.cert., javax.security.sasl., javax.sound., javax.sql., javax.swing., javax.tools., javax.transaction., -javax.xml.registry., -javax.xml.rpc., javax.xml., org.w3c.dom., org.xml.sax., org.apache.commons.logging., org.apache.log4j., -org.apache.hadoop.hbase., -org.apache.hadoop.hive., org.apache.hadoop., core-default.xml, hdfs-default.xml, mapred-default.xml, yarn-default.xml",
...

This is due to the hive-storage-api dependency of the orc-mapreduce library, which provides some classes under the org.apache.hadoop.hive package. If instead using the setting "mapreduce.job.user.classpath.first": "true", then this will not be an issue.

Examples

orc parser, orc parseSpec, auto field discovery, flatten expressions

{
  "type": "index_hadoop",
  "spec": {
    "ioConfig": {
      "type": "hadoop",
      "inputSpec": {
        "type": "static",
        "inputFormat": "org.apache.orc.mapreduce.OrcInputFormat",
        "paths": "path/to/file.orc"
      },
      ...
    },
    "dataSchema": {
      "dataSource": "example",
      "parser": {
        "type": "orc",
        "parseSpec": {
          "format": "orc",
          "flattenSpec": {
            "useFieldDiscovery": true,
            "fields": [
              {
                "type": "path",
                "name": "nestedDim",
                "expr": "$.nestedData.dim1"
              },
              {
                "type": "path",
                "name": "listDimFirstItem",
                "expr": "$.listDim[1]"
              }
            ]
          },
          "timestampSpec": {
            "column": "timestamp",
            "format": "millis"
          }
        }
      },
      ...
    },
    "tuningConfig": <hadoop-tuning-config>
    }
  }
}

orc parser, orc parseSpec, field discovery with no flattenSpec or dimensionSpec

{
  "type": "index_hadoop",
  "spec": {
    "ioConfig": {
      "type": "hadoop",
      "inputSpec": {
        "type": "static",
        "inputFormat": "org.apache.orc.mapreduce.OrcInputFormat",
        "paths": "path/to/file.orc"
      },
      ...
    },
    "dataSchema": {
      "dataSource": "example",
      "parser": {
        "type": "orc",
        "parseSpec": {
          "format": "orc",
          "timestampSpec": {
            "column": "timestamp",
            "format": "millis"
          }
        }
      },
      ...
    },
    "tuningConfig": <hadoop-tuning-config>
    }
  }
}

orc parser, orc parseSpec, no autodiscovery

{
  "type": "index_hadoop",
  "spec": {
    "ioConfig": {
      "type": "hadoop",
      "inputSpec": {
        "type": "static",
        "inputFormat": "org.apache.orc.mapreduce.OrcInputFormat",
        "paths": "path/to/file.orc"
      },
      ...
    },
    "dataSchema": {
      "dataSource": "example",
      "parser": {
        "type": "orc",
        "parseSpec": {
          "format": "orc",
          "flattenSpec": {
            "useFieldDiscovery": false,
            "fields": [
              {
                "type": "path",
                "name": "nestedDim",
                "expr": "$.nestedData.dim1"
              },
              {
                "type": "path",
                "name": "listDimFirstItem",
                "expr": "$.listDim[1]"
              }
            ]
          },
          "timestampSpec": {
            "column": "timestamp",
            "format": "millis"
          },
          "dimensionsSpec": {
            "dimensions": [
              "dim1",
              "dim3",
              "nestedDim",
              "listDimFirstItem"
            ],
            "dimensionExclusions": [],
            "spatialDimensions": []
          }
        }
      },
      ...
    },
    "tuningConfig": <hadoop-tuning-config>
    }
  }
}

orc parser, timeAndDims parseSpec

{
  "type": "index_hadoop",
  "spec": {
    "ioConfig": {
      "type": "hadoop",
      "inputSpec": {
        "type": "static",
        "inputFormat": "org.apache.orc.mapreduce.OrcInputFormat",
        "paths": "path/to/file.orc"
      },
      ...
    },
    "dataSchema": {
      "dataSource": "example",
      "parser": {
        "type": "orc",
        "parseSpec": {
          "format": "timeAndDims",
          "timestampSpec": {
            "column": "timestamp",
            "format": "auto"
          },
          "dimensionsSpec": {
            "dimensions": [
              "dim1",
              "dim2",
              "dim3",
              "listDim"
            ],
            "dimensionExclusions": [],
            "spatialDimensions": []
          }
        }
      },
      ...
    },
    "tuningConfig": <hadoop-tuning-config>
  }
}

Parquet Hadoop Parser

info

You need to include the druid-parquet-extensions as an extension to use the Parquet Hadoop Parser.

The Parquet Hadoop parser is for Hadoop batch ingestion and parses Parquet files directly. The inputFormat of inputSpec in ioConfig must be set to org.apache.druid.data.input.parquet.DruidParquetInputFormat.

The Parquet Hadoop Parser supports auto field discovery and flattening if provided with a flattenSpec with the parquet parseSpec. Parquet nested list and map logical types should operate correctly with JSON path expressions for all supported types.

When the time dimension is a DateType column, a format should not be supplied. When the format is UTF8 (String), either auto or a explicitly defined format is required.

Parquet Hadoop Parser vs Parquet Avro Hadoop Parser

Both parsers read from Parquet files, but slightly differently. The main differences are:

• The Parquet Hadoop Parser uses a simple conversion while the Parquet Avro Hadoop Parser converts Parquet data into avro records first with the parquet-avro library and then parses avro data using the druid-avro-extensions module to ingest into Druid.
• The Parquet Hadoop Parser sets a hadoop job property parquet.avro.add-list-element-records to false (which normally defaults to true), in order to ‘unwrap’ primitive list elements into multi-value dimensions.
• The Parquet Hadoop Parser supports int96 Parquet values, while the Parquet Avro Hadoop Parser does not. There may also be some subtle differences in the behavior of JSON path expression evaluation of flattenSpec.

Based on those differences, we suggest using the Parquet Hadoop Parser over the Parquet Avro Hadoop Parser to allow ingesting data beyond the schema constraints of Avro conversion. However, the Parquet Avro Hadoop Parser was the original basis for supporting the Parquet format, and as such it is a bit more mature.

Examples

parquet parser, parquet parseSpec

{
  "type": "index_hadoop",
  "spec": {
    "ioConfig": {
      "type": "hadoop",
      "inputSpec": {
        "type": "static",
        "inputFormat": "org.apache.druid.data.input.parquet.DruidParquetInputFormat",
        "paths": "path/to/file.parquet"
      },
      ...
    },
    "dataSchema": {
      "dataSource": "example",
      "parser": {
        "type": "parquet",
        "parseSpec": {
          "format": "parquet",
          "flattenSpec": {
            "useFieldDiscovery": true,
            "fields": [
              {
                "type": "path",
                "name": "nestedDim",
                "expr": "$.nestedData.dim1"
              },
              {
                "type": "path",
                "name": "listDimFirstItem",
                "expr": "$.listDim[1]"
              }
            ]
          },
          "timestampSpec": {
            "column": "timestamp",
            "format": "auto"
          },
          "dimensionsSpec": {
            "dimensions": [],
            "dimensionExclusions": [],
            "spatialDimensions": []
          }
        }
      },
      ...
    },
    "tuningConfig": <hadoop-tuning-config>
    }
  }
}

parquet parser, timeAndDims parseSpec

{
  "type": "index_hadoop",
  "spec": {
    "ioConfig": {
      "type": "hadoop",
      "inputSpec": {
        "type": "static",
        "inputFormat": "org.apache.druid.data.input.parquet.DruidParquetInputFormat",
        "paths": "path/to/file.parquet"
      },
      ...
    },
    "dataSchema": {
      "dataSource": "example",
      "parser": {
        "type": "parquet",
        "parseSpec": {
          "format": "timeAndDims",
          "timestampSpec": {
            "column": "timestamp",
            "format": "auto"
          },
          "dimensionsSpec": {
            "dimensions": [
              "dim1",
              "dim2",
              "dim3",
              "listDim"
            ],
            "dimensionExclusions": [],
            "spatialDimensions": []
          }
        }
      },
      ...
    },
    "tuningConfig": <hadoop-tuning-config>
  }
}

Parquet Avro Hadoop Parser

info

Consider using the Parquet Hadoop Parser over this parser to ingest Parquet files. See Parquet Hadoop Parser vs Parquet Avro Hadoop Parser for the differences between those parsers.

info

You need to include both the druid-parquet-extensions [druid-avro-extensions] as extensions to use the Parquet Avro Hadoop Parser.

The Parquet Avro Hadoop Parser is for Hadoop batch ingestion. This parser first converts the Parquet data into Avro records, and then parses them to ingest into Druid. The inputFormat of inputSpec in ioConfig must be set to org.apache.druid.data.input.parquet.DruidParquetAvroInputFormat.

The Parquet Avro Hadoop Parser supports auto field discovery and flattening if provided with a flattenSpec with the avro parseSpec. Parquet nested list and map logical types should operate correctly with JSON path expressions for all supported types. This parser sets a hadoop job property parquet.avro.add-list-element-records to false (which normally defaults to true), in order to ‘unwrap’ primitive list elements into multi-value dimensions.

Note that the int96 Parquet value type is not supported with this parser.

When the time dimension is a DateType column, a format should not be supplied. When the format is UTF8 (String), either auto or an explicitly defined format is required.

Example

{
  "type": "index_hadoop",
  "spec": {
    "ioConfig": {
      "type": "hadoop",
      "inputSpec": {
        "type": "static",
        "inputFormat": "org.apache.druid.data.input.parquet.DruidParquetAvroInputFormat",
        "paths": "path/to/file.parquet"
      },
      ...
    },
    "dataSchema": {
      "dataSource": "example",
      "parser": {
        "type": "parquet-avro",
        "parseSpec": {
          "format": "avro",
          "flattenSpec": {
            "useFieldDiscovery": true,
            "fields": [
              {
                "type": "path",
                "name": "nestedDim",
                "expr": "$.nestedData.dim1"
              },
              {
                "type": "path",
                "name": "listDimFirstItem",
                "expr": "$.listDim[1]"
              }
            ]
          },
          "timestampSpec": {
            "column": "timestamp",
            "format": "auto"
          },
          "dimensionsSpec": {
            "dimensions": [],
            "dimensionExclusions": [],
            "spatialDimensions": []
          }
        }
      },
      ...
    },
    "tuningConfig": <hadoop-tuning-config>
    }
  }
}

Avro Stream Parser

info

You need to include the druid-avro-extensions as an extension to use the Avro Stream Parser.

info

See the Avro Types section for how Avro types are handled in Druid

This parser is for stream ingestion and reads Avro data from a stream directly.

An Avro parseSpec can contain a flattenSpec using either the “root” or “path” field types, which can be used to read nested Avro records. The “jq” and “tree” field type is not currently supported for Avro.

For example, using Avro stream parser with schema repo Avro bytes decoder:

"parser" : {
  "type" : "avro_stream",
  "avroBytesDecoder" : {
    "type" : "schema_repo",
    "subjectAndIdConverter" : {
      "type" : "avro_1124",
      "topic" : "${YOUR_TOPIC}"
    },
    "schemaRepository" : {
      "type" : "avro_1124_rest_client",
      "url" : "${YOUR_SCHEMA_REPO_END_POINT}",
    }
  },
  "parseSpec" : {
    "format": "avro",
    "timestampSpec": <standard timestampSpec>,
    "dimensionsSpec": <standard dimensionsSpec>,
    "flattenSpec": <optional>
  }
}

Protobuf Parser

info

You need to include the druid-protobuf-extensions as an extension to use the Protobuf Parser.

This parser is for stream ingestion and reads Protocol buffer data from a stream directly.

Sample spec:

"parser": {
  "type": "protobuf",
  "protoBytesDecoder": {
    "type": "file",
    "descriptor": "file:///tmp/metrics.desc",
    "protoMessageType": "Metrics"
  },
  "parseSpec": {
    "format": "json",
    "timestampSpec": {
      "column": "timestamp",
      "format": "auto"
    },
    "dimensionsSpec": {
      "dimensions": [
        "unit",
        "http_method",
        "http_code",
        "page",
        "metricType",
        "server"
      ],
      "dimensionExclusions": [
        "timestamp",
        "value"
      ]
    }
  }
}

See the extension description for more details and examples.

Protobuf Bytes Decoder

If type is not included, the protoBytesDecoder defaults to schema_registry.

File-based Protobuf Bytes Decoder

This Protobuf bytes decoder first read a descriptor file, and then parse it to get schema used to decode the Protobuf record from bytes.

Sample spec:

"protoBytesDecoder": {
  "type": "file",
  "descriptor": "file:///tmp/metrics.desc",
  "protoMessageType": "Metrics"
}

Inline Descriptor Protobuf Bytes Decoder

This Protobuf bytes decoder allows the user to provide the contents of a Protobuf descriptor file inline, encoded as a Base64 string, and then parse it to get schema used to decode the Protobuf record from bytes.

Sample spec:

"protoBytesDecoder": {
  "type": "inline",
  "descriptorString": <Contents of a Protobuf descriptor file encoded as Base64 string>,
  "protoMessageType": "Metrics"
}

Confluent Schema Registry-based Protobuf Bytes Decoder

This Protobuf bytes decoder first extracts a unique id from input message bytes, and then uses it to look up the schema in the Schema Registry used to decode the Avro record from bytes. For details, see the Schema Registry documentation and repository.

For a single schema registry instance, use Field url or urls for multi instances.

Single Instance:

...
"protoBytesDecoder": {
  "url": <schema-registry-url>,
  "type": "schema_registry"
}
...

Multiple Instances:

...
"protoBytesDecoder": {
  "urls": [<schema-registry-url-1>, <schema-registry-url-2>, ...],
  "type": "schema_registry",
  "capacity": 100,
  "config" : {
       "basic.auth.credentials.source": "USER_INFO",
       "basic.auth.user.info": "fred:letmein",
       "schema.registry.ssl.truststore.location": "/some/secrets/kafka.client.truststore.jks",
       "schema.registry.ssl.truststore.password": "<password>",
       "schema.registry.ssl.keystore.location": "/some/secrets/kafka.client.keystore.jks",
       "schema.registry.ssl.keystore.password": "<password>",
       "schema.registry.ssl.key.password": "<password>",
         ... 
  },
  "headers": {
      "traceID" : "b29c5de2-0db4-490b-b421",
      "timeStamp" : "1577191871865",
      "druid.dynamic.config.provider":{
           "type":"mapString", 
           "config":{
                "registry.header.prop.1":"value.1", 
                "registry.header.prop.2":"value.2"
                }
           }
      ...
  }
}
...

ParseSpec

info

The Parser is deprecated for native batch tasks, Kafka indexing service, and Kinesis indexing service. Consider using the input format instead for these types of ingestion.

ParseSpecs serve two purposes:

• The String Parser use them to determine the format (i.e., JSON, CSV, TSV) of incoming rows.
• All Parsers use them to determine the timestamp and dimensions of incoming rows.

If format is not included, the parseSpec defaults to tsv.

JSON ParseSpec

Use this with the String Parser to load JSON.

Sample spec:

"parseSpec": {
  "format" : "json",
  "timestampSpec" : {
    "column" : "timestamp"
  },
  "dimensionSpec" : {
    "dimensions" : ["page","language","user","unpatrolled","newPage","robot","anonymous","namespace","continent","country","region","city"]
  }
}

JSON Lowercase ParseSpec

info

The jsonLowercase parser is deprecated and may be removed in a future version of Druid.

This is a special variation of the JSON ParseSpec that lower cases all the column names in the incoming JSON data. This parseSpec is required if you are updating to Druid 0.7.x from Druid 0.6.x, are directly ingesting JSON with mixed case column names, do not have any ETL in place to lower case those column names, and would like to make queries that include the data you created using 0.6.x and 0.7.x.

CSV ParseSpec

Use this with the String Parser to load CSV. Strings are parsed using the com.opencsv library.

Sample spec:

"parseSpec": {
  "format" : "csv",
  "timestampSpec" : {
    "column" : "timestamp"
  },
  "columns" : ["timestamp","page","language","user","unpatrolled","newPage","robot","anonymous","namespace","continent","country","region","city","added","deleted","delta"],
  "dimensionsSpec" : {
    "dimensions" : ["page","language","user","unpatrolled","newPage","robot","anonymous","namespace","continent","country","region","city"]
  }
}

CSV Index Tasks

If your input files contain a header, the columns field is optional and you don’t need to set. Instead, you can set the hasHeaderRow field to true, which makes Druid automatically extract the column information from the header. Otherwise, you must set the columns field and ensure that field must match the columns of your input data in the same order.

Also, you can skip some header rows by setting skipHeaderRows in your parseSpec. If both skipHeaderRows and hasHeaderRow options are set, skipHeaderRows is first applied. For example, if you set skipHeaderRows to 2 and hasHeaderRow to true, Druid will skip the first two lines and then extract column information from the third line.

Note that hasHeaderRow and skipHeaderRows are effective only for non-Hadoop batch index tasks. Other types of index tasks will fail with an exception.

Other CSV Ingestion Tasks

The columns field must be included and and ensure that the order of the fields matches the columns of your input data in the same order.

TSV / Delimited ParseSpec

Use this with the String Parser to load any delimited text that does not require special escaping. By default, the delimiter is a tab, so this will load TSV.

Sample spec:

"parseSpec": {
  "format" : "tsv",
  "timestampSpec" : {
    "column" : "timestamp"
  },
  "columns" : ["timestamp","page","language","user","unpatrolled","newPage","robot","anonymous","namespace","continent","country","region","city","added","deleted","delta"],
  "delimiter":"|",
  "dimensionsSpec" : {
    "dimensions" : ["page","language","user","unpatrolled","newPage","robot","anonymous","namespace","continent","country","region","city"]
  }
}