Data management

Schema changes

Schemas for datasources can change at any time and Apache Druid supports different schemas among segments.

Replacing segments

Druid uniquely identifies segments using the datasource, interval, version, and partition number. The partition number is only visible in the segment id if there are multiple segments created for some granularity of time. For example, if you have hourly segments, but you have more data in an hour than a single segment can hold, you can create multiple segments for the same hour. These segments will share the same datasource, interval, and version, but have linearly increasing partition numbers.

foo_2015-01-01/2015-01-02_v1_0
foo_2015-01-01/2015-01-02_v1_1
foo_2015-01-01/2015-01-02_v1_2

In the example segments above, the dataSource = foo, interval = 2015-01-01/2015-01-02, version = v1, partitionNum = 0. If at some later point in time, you reindex the data with a new schema, the newly created segments will have a higher version id.

foo_2015-01-01/2015-01-02_v2_0
foo_2015-01-01/2015-01-02_v2_1
foo_2015-01-01/2015-01-02_v2_2

Druid batch indexing (either Hadoop-based or IndexTask-based) guarantees atomic updates on an interval-by-interval basis. In our example, until all v2 segments for 2015-01-01/2015-01-02 are loaded in a Druid cluster, queries exclusively use v1 segments. Once all v2 segments are loaded and queryable, all queries ignore v1 segments and switch to the v2 segments. Shortly afterwards, the v1 segments are unloaded from the cluster.

Note that updates that span multiple segment intervals are only atomic within each interval. They are not atomic across the entire update. For example, you have segments such as the following:

foo_2015-01-01/2015-01-02_v1_0
foo_2015-01-02/2015-01-03_v1_1
foo_2015-01-03/2015-01-04_v1_2

v2 segments will be loaded into the cluster as soon as they are built and replace v1 segments for the period of time the segments overlap. Before v2 segments are completely loaded, your cluster may have a mixture of v1 and v2 segments.

foo_2015-01-01/2015-01-02_v1_0
foo_2015-01-02/2015-01-03_v2_1
foo_2015-01-03/2015-01-04_v1_2

In this case, queries may hit a mixture of v1 and v2 segments.

Different schemas among segments

Druid segments for the same datasource may have different schemas. If a string column (dimension) exists in one segment but not another, queries that involve both segments still work. Queries for the segment missing the dimension will behave as if the dimension has only null values. Similarly, if one segment has a numeric column (metric) but another does not, queries on the segment missing the metric will generally “do the right thing”. Aggregations over this missing metric behave as if the metric were missing.

Compaction and reindexing

Compaction is a type of overwrite operation, which reads an existing set of segments, combines them into a new set with larger but fewer segments, and overwrites the original set with the new compacted set, without changing the data that is stored.

For performance reasons, it is sometimes beneficial to compact a set of segments into a set of larger but fewer segments, as there is some per-segment processing and memory overhead in both the ingestion and querying paths.

Compaction tasks merge all segments of the given interval. The syntax is:

{
    "type": "compact",
    "id": <task_id>,
    "dataSource": <task_datasource>,
    "ioConfig": <IO config>,
    "dimensionsSpec" <custom dimensionsSpec>,
    "metricsSpec" <custom metricsSpec>,
    "segmentGranularity": <segment granularity after compaction>,
    "tuningConfig" <parallel indexing task tuningConfig>,
    "context": <task context>
}

An example of compaction task is

{
  "type" : "compact",
  "dataSource" : "wikipedia",
  "ioConfig" : {
    "type": "compact",
    "inputSpec": {
      "type": "interval",
      "interval": "2017-01-01/2018-01-01"
    }
  }
}

This compaction task reads all segments of the interval 2017-01-01/2018-01-01 and results in new segments. Since segmentGranularity is null, the original segment granularity will be remained and not changed after compaction. To control the number of result segments per time chunk, you can set maxRowsPerSegment or numShards. Please note that you can run multiple compactionTasks at the same time. For example, you can run 12 compactionTasks per month instead of running a single task for the entire year.

A compaction task internally generates an index task spec for performing compaction work with some fixed parameters. For example, its inputSource is always the DruidInputSource, and dimensionsSpec and metricsSpec include all dimensions and metrics of the input segments by default.

Compaction tasks will exit with a failure status code, without doing anything, if the interval you specify has no data segments loaded in it (or if the interval you specify is empty).

The output segment can have different metadata from the input segments unless all input segments have the same metadata.

• Dimensions: since Apache Druid supports schema change, the dimensions can be different across segments even if they are a part of the same dataSource. If the input segments have different dimensions, the output segment basically includes all dimensions of the input segments. However, even if the input segments have the same set of dimensions, the dimension order or the data type of dimensions can be different. For example, the data type of some dimensions can be changed from string to primitive types, or the order of dimensions can be changed for better locality. In this case, the dimensions of recent segments precede that of old segments in terms of data types and the ordering. This is because more recent segments are more likely to have the new desired order and data types. If you want to use your own ordering and types, you can specify a custom dimensionsSpec in the compaction task spec.
• Roll-up: the output segment is rolled up only when rollup is set for all input segments. See Roll-up for more details. You can check that your segments are rolled up or not by using Segment Metadata Queries.

Compaction IOConfig

The compaction IOConfig requires specifying inputSpec as seen below.

There are two supported inputSpecs for now.

The interval inputSpec is:

The segments inputSpec is:

Adding new data

Druid can insert new data to an existing datasource by appending new segments to existing segment sets. It can also add new data by merging an existing set of segments with new data and overwriting the original set.

Druid does not support single-record updates by primary key.

Updating existing data

Once you ingest some data in a dataSource for an interval and create Apache Druid segments, you might want to make changes to the ingested data. There are several ways this can be done.

Using lookups

If you have a dimension where values need to be updated frequently, try first using lookups. A classic use case of lookups is when you have an ID dimension stored in a Druid segment, and want to map the ID dimension to a human-readable String value that may need to be updated periodically.

Reingesting data

If lookup-based techniques are not sufficient, you will need to reingest data into Druid for the time chunks that you want to update. This can be done using one of the batch ingestion methods in overwrite mode (the default mode). It can also be done using streaming ingestion, provided you drop data for the relevant time chunks first.

If you do the reingestion in batch mode, Druid’s atomic update mechanism means that queries will flip seamlessly from the old data to the new data.

We recommend keeping a copy of your raw data around in case you ever need to reingest it.

With Hadoop-based ingestion

This section assumes the reader understands how to do batch ingestion using Hadoop. See Hadoop batch ingestion for more information. Hadoop batch-ingestion can be used for reindexing and delta ingestion.

Druid uses an inputSpec in the ioConfig to know where the data to be ingested is located and how to read it. For simple Hadoop batch ingestion, static or granularity spec types allow you to read data stored in deep storage.

There are other types of inputSpec to enable reindexing and delta ingestion.

Reindexing with Native Batch Ingestion

This section assumes the reader understands how to do batch ingestion without Hadoop using native batch indexing, which uses an inputSource to know where and how to read the input data. The DruidInputSource can be used to read data from segments inside Druid. Note that IndexTask is to be used for prototyping purposes only as it has to do all processing inside a single process and can’t scale. Please use Hadoop batch ingestion for production scenarios dealing with more than 1GB of data.

Deleting data

Druid supports permanent deletion of segments that are in an “unused” state (see the Segment lifecycle section of the Architecture page).

The Kill Task deletes unused segments within a specified interval from metadata storage and deep storage.

For more information, please see Kill Task.

Permanent deletion of a segment in Apache Druid has two steps:

1. The segment must first be marked as “unused”. This occurs when a segment is dropped by retention rules, and when a user manually disables a segment through the Coordinator API.
2. After segments have been marked as “unused”, a Kill Task will delete any “unused” segments from Druid’s metadata store as well as deep storage.

For documentation on retention rules, please see Data Retention.

For documentation on disabling segments using the Coordinator API, please see the Coordinator Datasources API reference.

A data deletion tutorial is available at Tutorial: Deleting data

Kill Task

Kill tasks delete all information about a segment and removes it from deep storage. Segments to kill must be unused (used==0) in the Druid segment table. The available grammar is:

{
    "type": "kill",
    "id": <task_id>,
    "dataSource": <task_datasource>,
    "interval" : <all_segments_in_this_interval_will_die!>,
    "context": <task context>
}

Retention

Druid supports retention rules, which are used to define intervals of time where data should be preserved, and intervals where data should be discarded.

Druid also supports separating Historical processes into tiers, and the retention rules can be configured to assign data for specific intervals to specific tiers.

These features are useful for performance/cost management; a common use case is separating Historical processes into a “hot” tier and a “cold” tier.

For more information, please see Load rules.