Clustered deployment

Clustered deployment

Apache Druid is designed to be deployed as a scalable, fault-tolerant cluster.

In this document, we’ll set up a simple cluster and discuss how it can be further configured to meet your needs.

This simple cluster will feature:

A Master server to host the Coordinator and Overlord processes
Two scalable, fault-tolerant Data servers running Historical and MiddleManager processes
A query server, hosting the Druid Broker and Router processes

In production, we recommend deploying multiple Master servers and multiple Query servers in a fault-tolerant configuration based on your specific fault-tolerance needs, but you can get started quickly with one Master and one Query server and add more servers later.

Select hardware

Fresh Deployment

If you do not have an existing Druid cluster, and wish to start running Druid in a clustered deployment, this guide provides an example clustered deployment with pre-made configurations.

Master server

The Coordinator and Overlord processes are responsible for handling the metadata and coordination needs of your cluster. They can be colocated together on the same server.

In this example, we will be deploying the equivalent of one AWS m5.2xlarge instance.

This hardware offers:

8 vCPUs
32 GiB RAM

Example Master server configurations that have been sized for this hardware can be found under conf/druid/cluster/master.

Data server

Historicals and MiddleManagers can be colocated on the same server to handle the actual data in your cluster. These servers benefit greatly from CPU, RAM, and SSDs.

In this example, we will be deploying the equivalent of two AWS i3.4xlarge instances.

This hardware offers:

16 vCPUs
122 GiB RAM
2 * 1.9TB SSD storage

Example Data server configurations that have been sized for this hardware can be found under conf/druid/cluster/data.

Query server

Druid Brokers accept queries and farm them out to the rest of the cluster. They also optionally maintain an in-memory query cache. These servers benefit greatly from CPU and RAM.

In this example, we will be deploying the equivalent of one AWS m5.2xlarge instance.

This hardware offers:

8 vCPUs
32 GiB RAM

You can consider co-locating any open source UIs or query libraries on the same server that the Broker is running on.

Example Query server configurations that have been sized for this hardware can be found under conf/druid/cluster/query.

Other Hardware Sizes

The example cluster above is chosen as a single example out of many possible ways to size a Druid cluster.

You can choose smaller/larger hardware or less/more servers for your specific needs and constraints.

If your use case has complex scaling requirements, you can also choose to not co-locate Druid processes (e.g., standalone Historical servers).

The information in the basic cluster tuning guide can help with your decision-making process and with sizing your configurations.

Migrating from a single-server deployment

If you have an existing single-server deployment, such as the ones from the single-server deployment examples, and you wish to migrate to a clustered deployment of similar scale, the following section contains guidelines for choosing equivalent hardware using the Master/Data/Query server organization.

Master server

The main considerations for the Master server are available CPUs and RAM for the Coordinator and Overlord heaps.

Sum up the allocated heap sizes for your Coordinator and Overlord from the single-server deployment, and choose Master server hardware with enough RAM for the combined heaps, with some extra RAM for other processes on the machine.

For CPU cores, you can choose hardware with approximately 1/4th of the cores of the single-server deployment.

Data server

When choosing Data server hardware for the cluster, the main considerations are available CPUs and RAM, and using SSD storage if feasible.

In a clustered deployment, having multiple Data servers is a good idea for fault-tolerance purposes.

When choosing the Data server hardware, you can choose a split factor N, divide the original CPU/RAM of the single-server deployment by N, and deploy N Data servers of reduced size in the new cluster.

Instructions for adjusting the Historical/MiddleManager configs for the split are described in a later section in this guide.

Query server

The main considerations for the Query server are available CPUs and RAM for the Broker heap + direct memory, and Router heap.

Sum up the allocated memory sizes for your Broker and Router from the single-server deployment, and choose Query server hardware with enough RAM to cover the Broker/Router, with some extra RAM for other processes on the machine.

For CPU cores, you can choose hardware with approximately 1/4th of the cores of the single-server deployment.

The basic cluster tuning guide has information on how to calculate Broker/Router memory usage.

Select OS

We recommend running your favorite Linux distribution. You will also need

Java 8u92+, 11, or 17
Python 2 or Python 3

info

If needed, you can specify where to find Java using the environment variables DRUID_JAVA_HOME or JAVA_HOME. For more details run the bin/verify-java script.

For information about installing Java, see the documentation for your OS package manager. If your Ubuntu-based OS does not have a recent enough version of Java, WebUpd8 offers packages for those OSes.

Download the distribution

First, download and unpack the release archive. It’s best to do this on a single machine at first, since you will be editing the configurations and then copying the modified distribution out to all of your servers.

Download the 30.0.1 release.

Extract Druid by running the following commands in your terminal:

tar -xzf apache-druid-30.0.1-bin.tar.gz
cd apache-druid-30.0.1

In the package, you should find:

LICENSE and NOTICE files
bin/* - scripts related to the single-machine quickstart
conf/druid/cluster/* - template configurations for a clustered setup
extensions/* - core Druid extensions
hadoop-dependencies/* - Druid Hadoop dependencies
lib/* - libraries and dependencies for core Druid
quickstart/* - files related to the single-machine quickstart

We’ll be editing the files in conf/druid/cluster/ in order to get things running.

Migrating from Single-Server Deployments

In the following sections we will be editing the configs under conf/druid/cluster.

If you have an existing single-server deployment, please copy your existing configs to conf/druid/cluster to preserve any config changes you have made.

Configure metadata storage and deep storage

Migrating from Single-Server Deployments

If you have an existing single-server deployment and you wish to preserve your data across the migration, please follow the instructions at metadata migration and deep storage migration before updating your metadata/deep storage configs.

These guides are targeted at single-server deployments that use the Derby metadata store and local deep storage. If you are already using a non-Derby metadata store in your single-server cluster, you can reuse the existing metadata store for the new cluster.

These guides also provide information on migrating segments from local deep storage. A clustered deployment requires distributed deep storage like S3 or HDFS. If your single-server deployment was already using distributed deep storage, you can reuse the existing deep storage for the new cluster.

Metadata storage

In conf/druid/cluster/_common/common.runtime.properties, replace “metadata.storage.*“ with the address of the machine that you will use as your metadata store:

druid.metadata.storage.connector.connectURI
druid.metadata.storage.connector.host

In a production deployment, we recommend running a dedicated metadata store such as MySQL or PostgreSQL with replication, deployed separately from the Druid servers.

The MySQL extension and PostgreSQL extension docs have instructions for extension configuration and initial database setup.

Deep storage

Druid relies on a distributed filesystem or large object (blob) store for data storage. The most commonly used deep storage implementations are S3 (popular for those on AWS) and HDFS (popular if you already have a Hadoop deployment).

S3

In conf/druid/cluster/_common/common.runtime.properties,

Add “druid-s3-extensions” to druid.extensions.loadList.
Comment out the configurations for local storage under “Deep Storage” and “Indexing service logs”.
Uncomment and configure appropriate values in the “For S3” sections of “Deep Storage” and “Indexing service logs”.

After this, you should have made the following changes:

druid.extensions.loadList=["druid-s3-extensions"]
#druid.storage.type=local
#druid.storage.storageDirectory=var/druid/segments
druid.storage.type=s3
druid.storage.bucket=your-bucket
druid.storage.baseKey=druid/segments
druid.s3.accessKey=...
druid.s3.secretKey=...
#druid.indexer.logs.type=file
#druid.indexer.logs.directory=var/druid/indexing-logs
druid.indexer.logs.type=s3
druid.indexer.logs.s3Bucket=your-bucket
druid.indexer.logs.s3Prefix=druid/indexing-logs

Please see the S3 extension documentation for more info.

HDFS

In conf/druid/cluster/_common/common.runtime.properties,

Add “druid-hdfs-storage” to druid.extensions.loadList.
Comment out the configurations for local storage under “Deep Storage” and “Indexing service logs”.
Uncomment and configure appropriate values in the “For HDFS” sections of “Deep Storage” and “Indexing service logs”.

After this, you should have made the following changes:

druid.extensions.loadList=["druid-hdfs-storage"]
#druid.storage.type=local
#druid.storage.storageDirectory=var/druid/segments
druid.storage.type=hdfs
druid.storage.storageDirectory=/druid/segments
#druid.indexer.logs.type=file
#druid.indexer.logs.directory=var/druid/indexing-logs
druid.indexer.logs.type=hdfs
druid.indexer.logs.directory=/druid/indexing-logs

Also,

Place your Hadoop configuration XMLs (core-site.xml, hdfs-site.xml, yarn-site.xml, mapred-site.xml) on the classpath of your Druid processes. You can do this by copying them into conf/druid/cluster/_common/.

Please see the HDFS extension documentation for more info.

Configure for connecting to Hadoop (optional)

If you will be loading data from a Hadoop cluster, then at this point you should configure Druid to be aware of your cluster:

Update druid.indexer.task.hadoopWorkingPath in conf/druid/cluster/middleManager/runtime.properties to a path on HDFS that you’d like to use for temporary files required during the indexing process. druid.indexer.task.hadoopWorkingPath=/tmp/druid-indexing is a common choice.
Place your Hadoop configuration XMLs (core-site.xml, hdfs-site.xml, yarn-site.xml, mapred-site.xml) on the classpath of your Druid processes. You can do this by copying them into conf/druid/cluster/_common/core-site.xml, conf/druid/cluster/_common/hdfs-site.xml, and so on.

Note that you don’t need to use HDFS deep storage in order to load data from Hadoop. For example, if your cluster is running on Amazon Web Services, we recommend using S3 for deep storage even if you are loading data using Hadoop or Elastic MapReduce.

For more info, please see the Hadoop-based ingestion page.

Configure Zookeeper connection

In a production cluster, we recommend using a dedicated ZK cluster in a quorum, deployed separately from the Druid servers.

In conf/druid/cluster/_common/common.runtime.properties, set druid.zk.service.host to a connection string containing a comma separated list of host:port pairs, each corresponding to a ZooKeeper server in your ZK quorum. (e.g. “127.0.0.1:4545” or “127.0.0.1:3000,127.0.0.1:3001,127.0.0.1:3002”)

You can also choose to run ZK on the Master servers instead of having a dedicated ZK cluster. If doing so, we recommend deploying 3 Master servers so that you have a ZK quorum.

Configuration Tuning

Migrating from a Single-Server Deployment

Master

If you are using an example configuration from single-server deployment examples, these examples combine the Coordinator and Overlord processes into one combined process.

The example configs under conf/druid/cluster/master/coordinator-overlord also combine the Coordinator and Overlord processes.

You can copy your existing coordinator-overlord configs from the single-server deployment to conf/druid/cluster/master/coordinator-overlord.

Data

Suppose we are migrating from a single-server deployment that had 32 CPU and 256GiB RAM. In the old deployment, the following configurations for Historicals and MiddleManagers were applied:

Historical (Single-server)

druid.processing.buffer.sizeBytes=500MiB
druid.processing.numMergeBuffers=8
druid.processing.numThreads=31

MiddleManager (Single-server)

druid.worker.capacity=8
druid.indexer.fork.property.druid.processing.numMergeBuffers=2
druid.indexer.fork.property.druid.processing.buffer.sizeBytes=100MiB
druid.indexer.fork.property.druid.processing.numThreads=1

In the clustered deployment, we can choose a split factor (2 in this example), and deploy 2 Data servers with 16CPU and 128GiB RAM each. The areas to scale are the following:

Historical

druid.processing.numThreads: Set to (num_cores - 1) based on the new hardware
druid.processing.numMergeBuffers: Divide the old value from the single-server deployment by the split factor
druid.processing.buffer.sizeBytes: Keep this unchanged

MiddleManager:

druid.worker.capacity: Divide the old value from the single-server deployment by the split factor
druid.indexer.fork.property.druid.processing.numMergeBuffers: Keep this unchanged
druid.indexer.fork.property.druid.processing.buffer.sizeBytes: Keep this unchanged
druid.indexer.fork.property.druid.processing.numThreads: Keep this unchanged

The resulting configs after the split:

New Historical (on 2 Data servers)

druid.processing.buffer.sizeBytes=500MiB
druid.processing.numMergeBuffers=4
druid.processing.numThreads=15

New MiddleManager (on 2 Data servers)

druid.worker.capacity=4
druid.indexer.fork.property.druid.processing.numMergeBuffers=2
druid.indexer.fork.property.druid.processing.buffer.sizeBytes=100MiB
druid.indexer.fork.property.druid.processing.numThreads=1

Query

You can copy your existing Broker and Router configs to the directories under conf/druid/cluster/query, no modifications are needed, as long as the new hardware is sized accordingly.

Fresh deployment

If you are using the example cluster described above:

1 Master server (m5.2xlarge)
2 Data servers (i3.4xlarge)
1 Query server (m5.2xlarge)

The configurations under conf/druid/cluster have already been sized for this hardware and you do not need to make further modifications for general use cases.

If you have chosen different hardware, the basic cluster tuning guide can help you size your configurations.

Open ports (if using a firewall)

If you’re using a firewall or some other system that only allows traffic on specific ports, allow inbound connections on the following:

Master Server

1527 (Derby metadata store; not needed if you are using a separate metadata store like MySQL or PostgreSQL)
2181 (ZooKeeper; not needed if you are using a separate ZooKeeper cluster)
8081 (Coordinator)
8090 (Overlord)

Data Server

8083 (Historical)
8091, 8100–8199 (Druid Middle Manager; you may need higher than port 8199 if you have a very high druid.worker.capacity)

Query Server

8082 (Broker)
8088 (Router, if used)

info

In production, we recommend deploying ZooKeeper and your metadata store on their own dedicated hardware, rather than on the Master server.

Start Master Server

Copy the Druid distribution and your edited configurations to your Master server.

If you have been editing the configurations on your local machine, you can use rsync to copy them:

rsync -az apache-druid-30.0.1/ MASTER_SERVER:apache-druid-30.0.1/

No Zookeeper on Master

From the distribution root, run the following command to start the Master server:

bin/start-cluster-master-no-zk-server

With Zookeeper on Master

If you plan to run ZK on Master servers, first update conf/zoo.cfg to reflect how you plan to run ZK. Then, you can start the Master server processes together with ZK using:

bin/start-cluster-master-with-zk-server

info

In production, we also recommend running a ZooKeeper cluster on its own dedicated hardware.

Start Data Server

Copy the Druid distribution and your edited configurations to your Data servers.

From the distribution root, run the following command to start the Data server:

bin/start-cluster-data-server

You can add more Data servers as needed.

info

For clusters with complex resource allocation needs, you can break apart Historicals and MiddleManagers and scale the components individually. This also allows you take advantage of Druid’s built-in MiddleManager autoscaling facility.

Start Query Server

Copy the Druid distribution and your edited configurations to your Query servers.

From the distribution root, run the following command to start the Query server:

bin/start-cluster-query-server

You can add more Query servers as needed based on query load. If you increase the number of Query servers, be sure to adjust the connection pools on your Historicals and Tasks as described in the basic cluster tuning guide.

Loading data

Congratulations, you now have a Druid cluster! The next step is to learn about recommended ways to load data into Druid based on your use case. Read more about loading data.