« Monitoring settings in ElasticsearchNetworking »
Elastic Docs›Elasticsearch Guide [8.15]›Set up Elasticsearch›Configuring Elasticsearch
Any time that you start an instance of Elasticsearch, you are starting a node. Acollection of connected nodes is called a cluster. If youare running a single node of Elasticsearch, then you have a cluster of one node.
Every node in the cluster can handle HTTP and transporttraffic by default. The transport layer is used exclusively for communicationbetween nodes; the HTTP layer is used by REST clients.
All nodes know about all the other nodes in the cluster and can forward clientrequests to the appropriate node.
The performance of an Elasticsearch node is often limited by the performance of the underlying storage.Review our recommendations for optimizing your storage for indexing andsearch.
You define a node’s roles by setting node.roles
in elasticsearch.yml
. If youset node.roles
, the node is only assigned the roles you specify. If you don’tset node.roles
, the node is assigned the following roles:
master
data
data_content
data_hot
data_warm
data_cold
data_frozen
ingest
ml
remote_cluster_client
transform
If you set node.roles
, ensure you specify every node role your cluster needs.Every cluster requires the following node roles:
master
data_content
anddata_hot
ORdata
Some Elastic Stack features also require specific node roles:
- Cross-cluster search and cross-cluster replication require the
remote_cluster_client
role. - Stack Monitoring and ingest pipelines require the
ingest
role. - Fleet, the Elastic Security app, and transforms require the
transform
role.Theremote_cluster_client
role is also required to use cross-cluster search with thesefeatures. - Machine learning features, such as anomaly detection, require the
ml
role.
As the cluster grows and in particular if you have large machine learning jobs orcontinuous transforms, consider separating dedicated master-eligible nodes fromdedicated data nodes, machine learning nodes, and transform nodes.
- Master-eligible node
- A node that has the
master
role, which makes it eligible to beelected as the master node, which controls the cluster. - Data node
- A node that has one of several data roles. Data nodes hold data and perform datarelated operations such as CRUD, search, and aggregations. A node with a generic
data
role can fill any of the specialized data node roles. - Ingest node
- A node that has the
ingest
role. Ingest nodes are able to apply aningest pipeline to a document in order to transform and enrich thedocument before indexing. With a heavy ingest load, it makes sense to usededicated ingest nodes and to not include theingest
role from nodes that havethemaster
ordata
roles. - Remote-eligible node
- A node that has the
remote_cluster_client
role, which makes it eligible to actas a remote client. - Machine learning node
- A node that has the
ml
role. If you want to use machine learning features, there must beat least one machine learning node in your cluster. For more information, seeMachine learning settings and Machine learning in the Elastic Stack. - Transform node
- A node that has the
transform
role. If you want to use transforms, theremust be at least one transform node in your cluster. For more information, seeTransforms settings and Transforming data.
Coordinating node
Requests like search requests or bulk-indexing requests may involve data heldon different data nodes. A search request, for example, is executed in twophases which are coordinated by the node which receives the client request — the coordinating node.
In the scatter phase, the coordinating node forwards the request to the datanodes which hold the data. Each data node executes the request locally andreturns its results to the coordinating node. In the gather phase, thecoordinating node reduces each data node’s results into a single globalresult set.
Every node is implicitly a coordinating node. This means that a node that hasan explicit empty list of roles via node.roles
will only act as a coordinatingnode, which cannot be disabled. As a result, such a node needs to have enoughmemory and CPU in order to deal with the gather phase.
The master node is responsible for lightweight cluster-wide actions such ascreating or deleting an index, tracking which nodes are part of the cluster,and deciding which shards to allocate to which nodes. It is important forcluster health to have a stable master node.
Any master-eligible node that is not a voting-only node maybe elected to become the master node by the master electionprocess.
Master nodes must have a path.data
directory whose contentspersist across restarts, just like data nodes, because this is where thecluster metadata is stored. The cluster metadata describes how to read the datastored on the data nodes, so if it is lost then the data stored on the datanodes cannot be read.
It is important for the health of the cluster that the elected master node hasthe resources it needs to fulfill its responsibilities. If the elected masternode is overloaded with other tasks then the cluster will not operate well. Themost reliable way to avoid overloading the master with other tasks is toconfigure all the master-eligible nodes to be dedicated master-eligible nodeswhich only have the master
role, allowing them to focus on managing thecluster. Master-eligible nodes will still also behave ascoordinating nodes that route requests from clients tothe other nodes in the cluster, but you should not use dedicated master nodesfor this purpose.
A small or lightly-loaded cluster may operate well if its master-eligible nodeshave other roles and responsibilities, but once your cluster comprises morethan a handful of nodes it usually makes sense to use dedicated master-eligiblenodes.
To create a dedicated master-eligible node, set:
node.roles: [ master ]
A voting-only master-eligible node is a node that participates inmaster elections but which will not act as the cluster’selected master node. In particular, a voting-only node can serve as a tiebreakerin elections.
It may seem confusing to use the term "master-eligible" to describe avoting-only node since such a node is not actually eligible to become the masterat all. This terminology is an unfortunate consequence of history:master-eligible nodes are those nodes that participate in elections and performcertain tasks during cluster state publications, and voting-only nodes have thesame responsibilities even if they can never become the elected master.
To configure a master-eligible node as a voting-only node, include master
andvoting_only
in the list of roles. For example to create a voting-only datanode:
node.roles: [ data, master, voting_only ]
Only nodes with the master
role can be marked as having thevoting_only
role.
High availability (HA) clusters require at least three master-eligible nodes, atleast two of which are not voting-only nodes. Such a cluster will be able toelect a master node even if one of the nodes fails.
Voting-only master-eligible nodes may also fill other roles in your cluster.For instance, a node may be both a data node and a voting-only master-eligiblenode. A dedicated voting-only master-eligible nodes is a voting-onlymaster-eligible node that fills no other roles in the cluster. To create adedicated voting-only master-eligible node, set:
node.roles: [ master, voting_only ]
Since dedicated voting-only nodes never act as the cluster’s elected master,they may require less heap and a less powerful CPU than the true master nodes.However all master-eligible nodes, including voting-only nodes, are on thecritical path for publishing cluster stateupdates. Cluster state updates are usually independent ofperformance-critical workloads such as indexing or searches, but they areinvolved in management activities such as index creation and rollover, mappingupdates, and recovery after a failure. The performance characteristics of theseactivities are a function of the speed of the storage on each master-eligiblenode, as well as the reliability and latency of the network interconnectionsbetween the elected master node and the other nodes in the cluster. You musttherefore ensure that the storage and networking available to the nodes in yourcluster are good enough to meet your performance goals.
Data nodes hold the shards that contain the documents you have indexed. Datanodes handle data related operations like CRUD, search, and aggregations.These operations are I/O-, memory-, and CPU-intensive. It is important tomonitor these resources and to add more data nodes if they are overloaded.
The main benefit of having dedicated data nodes is the separation of the masterand data roles.
In a multi-tier deployment architecture, you use specialized data roles toassign data nodes to specific tiers: data_content
,data_hot
, data_warm
,data_cold
, or data_frozen
. A node can belong to multiple tiers.
If you want to include a node in all tiers, or if your cluster does not use multiple tiers, then you can use the generic data
role.
Cluster shard limits prevent creation of more than1000 non-frozen shards per node, and 3000 frozen shards per dedicated frozennode. Make sure you have enough nodes of each type in your cluster to handlethe number of shards you need.
If you assign a node to a specific tier using a specialized data role, then you shouldn’t also assign it the generic data
role. The generic data
role takes precedence over specialized data roles.
Generic data nodes are included in all content tiers.
To create a dedicated generic data node, set:
node.roles: [ data ]
Content data nodes are part of the content tier.Data stored in the content tier is generally a collection of items such as a product catalog or article archive.Unlike time series data, the value of the content remains relatively constant over time,so it doesn’t make sense to move it to a tier with different performance characteristics as it ages.Content data typically has long data retention requirements, and you want to be able to retrieveitems quickly regardless of how old they are.
Content tier nodes are usually optimized for query performance—they prioritize processing power over IO throughputso they can process complex searches and aggregations and return results quickly.While they are also responsible for indexing, content data is generally not ingested at as high a rateas time series data such as logs and metrics. From a resiliency perspective the indices in thistier should be configured to use one or more replicas.
The content tier is required. System indices and other indices that aren’t partof a data stream are automatically allocated to the content tier.
To create a dedicated content node, set:
node.roles: [ data_content ]
Hot data nodes are part of the hot tier.The hot tier is the Elasticsearch entry point for time series data and holds your most-recent,most-frequently-searched time series data.Nodes in the hot tier need to be fast for both reads and writes,which requires more hardware resources and faster storage (SSDs).For resiliency, indices in the hot tier should be configured to use one or more replicas.
The hot tier is required. New indices that are part of a data stream are automatically allocated to the hot tier.
To create a dedicated hot node, set:
node.roles: [ data_hot ]
Warm data nodes are part of the warm tier.Time series data can move to the warm tier once it is being queried less frequentlythan the recently-indexed data in the hot tier.The warm tier typically holds data from recent weeks.Updates are still allowed, but likely infrequent.Nodes in the warm tier generally don’t need to be as fast as those in the hot tier.For resiliency, indices in the warm tier should be configured to use one or more replicas.
To create a dedicated warm node, set:
node.roles: [ data_warm ]
Cold data nodes are part of the cold tier.When you no longer need to search time series data regularly, it can move fromthe warm tier to the cold tier. While still searchable, this tier is typicallyoptimized for lower storage costs rather than search speed.
For better storage savings, you can keep fully mounted indicesof searchable snapshots on the cold tier. Unlike regularindices, these fully mounted indices don’t require replicas for reliability. Inthe event of a failure, they can recover data from the underlying snapshotinstead. This potentially halves the local storage needed for the data. Asnapshot repository is required to use fully mounted indices in the cold tier.Fully mounted indices are read-only.
Alternatively, you can use the cold tier to store regular indices with replicas insteadof using searchable snapshots. This lets you store older data on less expensive hardwarebut doesn’t reduce required disk space compared to the warm tier.
To create a dedicated cold node, set:
node.roles: [ data_cold ]
Frozen data nodes are part of the frozen tier.Once data is no longer being queried, or being queried rarely, it may move fromthe cold tier to the frozen tier where it stays for the rest of its life.
The frozen tier requires a snapshot repository.The frozen tier uses partially mounted indices to storeand load data from a snapshot repository. This reduces local storage andoperating costs while still letting you search frozen data. Because Elasticsearch mustsometimes fetch frozen data from the snapshot repository, searches on the frozentier are typically slower than on the cold tier.
To create a dedicated frozen node, set:
node.roles: [ data_frozen ]
Ingest nodes can execute pre-processing pipelines, composed of one or moreingest processors. Depending on the type of operations performed by the ingestprocessors and the required resources, it may make sense to have dedicatedingest nodes, that will only perform this specific task.
To create a dedicated ingest node, set:
node.roles: [ ingest ]
If you take away the ability to be able to handle master duties, to hold data,and pre-process documents, then you are left with a coordinating node thatcan only route requests, handle the search reduce phase, and distribute bulkindexing. Essentially, coordinating only nodes behave as smart load balancers.
Coordinating only nodes can benefit large clusters by offloading thecoordinating node role from data and master-eligible nodes. They join thecluster and receive the full cluster state, like every othernode, and they use the cluster state to route requests directly to theappropriate place(s).
Adding too many coordinating only nodes to a cluster can increase theburden on the entire cluster because the elected master node must awaitacknowledgement of cluster state updates from every node! The benefit ofcoordinating only nodes should not be overstated — data nodes can happilyserve the same purpose.
To create a dedicated coordinating node, set:
node.roles: [ ]
A remote-eligible node acts as a cross-cluster client and connects toremote clusters. Once connected, you can searchremote clusters using cross-cluster search. You can also syncdata between clusters using cross-cluster replication.
node.roles: [ remote_cluster_client ]
Machine learning nodes run jobs and handle machine learning API requests. For more information, seeMachine learning settings.
To create a dedicated machine learning node, set:
node.roles: [ ml, remote_cluster_client]
The remote_cluster_client
role is optional but strongly recommended.Otherwise, cross-cluster search fails when used in machine learning jobs or datafeeds. If you use cross-cluster search inyour anomaly detection jobs, the remote_cluster_client
role is also required on allmaster-eligible nodes. Otherwise, the datafeed cannot start. See Remote-eligible node.
Transform nodes run transforms and handle transform API requests. Formore information, see Transforms settings.
To create a dedicated transform node, set:
node.roles: [ transform, remote_cluster_client ]
The remote_cluster_client
role is optional but strongly recommended.Otherwise, cross-cluster search fails when used in transforms. See Remote-eligible node.
Each data node maintains the following data on disk:
- the shard data for every shard allocated to that node,
- the index metadata corresponding with every shard allocated to that node, and
- the cluster-wide metadata, such as settings and index templates.
Similarly, each master-eligible node maintains the following data on disk:
- the index metadata for every index in the cluster, and
- the cluster-wide metadata, such as settings and index templates.
Each node checks the contents of its data path at startup. If it discoversunexpected data then it will refuse to start. This is to avoid importingunwanted dangling indices which can leadto a red cluster health. To be more precise, nodes without the data
role willrefuse to start if they find any shard data on disk at startup, and nodeswithout both the master
and data
roles will refuse to start if they have anyindex metadata on disk at startup.
It is possible to change the roles of a node by adjusting itselasticsearch.yml
file and restarting it. This is known as repurposing anode. In order to satisfy the checks for unexpected data described above, youmust perform some extra steps to prepare a node for repurposing when startingthe node without the data
or master
roles.
- If you want to repurpose a data node by removing the
data
role then youshould first use an allocation filter to safelymigrate all the shard data onto other nodes in the cluster. - If you want to repurpose a node to have neither the
data
normaster
rolesthen it is simplest to start a brand-new node with an empty data path and thedesired roles. You may find it safest to use anallocation filter to migrate the shard data elsewherein the cluster first.
If it is not possible to follow these extra steps then you may be able to usethe elasticsearch-node repurpose tool to delete anyexcess data that prevents a node from starting.
Every data and master-eligible node requires access to a data directory whereshards and index and cluster metadata will be stored. The path.data
defaultsto $ES_HOME/data
but can be configured in the elasticsearch.yml
configfile an absolute path or a path relative to $ES_HOME
as follows:
path.data: /var/elasticsearch/data
Like all node settings, it can also be specified on the command line as:
./bin/elasticsearch -Epath.data=/var/elasticsearch/data
The contents of the path.data
directory must persist across restarts, becausethis is where your data is stored. Elasticsearch requires the filesystem to act as if itwere backed by a local disk, but this means that it will work correctly onproperly-configured remote block devices (e.g. a SAN) and remote filesystems(e.g. NFS) as long as the remote storage behaves no differently from localstorage. You can run multiple Elasticsearch nodes on the same filesystem, but each Elasticsearchnode must have its own data path.
When using the .zip
or .tar.gz
distributions, the path.data
settingshould be configured to locate the data directory outside the Elasticsearch homedirectory, so that the home directory can be deleted without deleting your data!The RPM and Debian distributions do this for you already.
Don’t modify anything within the data directory or run processes thatmight interfere with its contents. If something other than Elasticsearch modifies thecontents of the data directory, then Elasticsearch may fail, reporting corruption orother data inconsistencies, or may appear to work correctly having silentlylost some of your data. Don’t attempt to take filesystem backups of the datadirectory; there is no supported way to restore such a backup. Instead, useSnapshot and restore to take backups safely. Don’t run virus scanners on thedata directory. A virus scanner can prevent Elasticsearch from working correctly and maymodify the contents of the data directory. The data directory contains noexecutables so a virus scan will only find false positives.
More node settings can be found in Configuring Elasticsearch and Important Elasticsearch configuration,including:
- cluster.name
- node.name
- network settings
« Monitoring settings in ElasticsearchNetworking »
Most Popular
Video
Get Started with ElasticsearchVideo
Intro to KibanaVideo