Before you begin reading this guide, we recommend you run Elasticsearch Error Check-Up which analyzes 2 JSON files to detect many errors.
To easily locate the root cause and resolve this issue try AutoOps for Elasticsearch & OpenSearch. It diagnoses problems by analyzing hundreds of metrics collected by a lightweight agent and offers guidance for resolving them. Take a self-guided product tour to see for yourself (no registration required).
This guide will help you check for common problems that cause the log ” Dropping pending state . more than pending states. ” to appear. To understand the issues related to this log, read the explanation below about the following Elasticsearch concepts: cluster, discovery and queue.
Overview
The process known as discovery occurs when an Elasticsearch node starts, restarts or loses contact with the master node for any reason. In those cases, the node needs to contact other nodes in the cluster to find any existing master node or initiate the election of a new master node.
How it works
Upon startup, each node looks for other nodes, firstly by contacting the IP addresses of eligible master nodes held in the previous cluster state. If they are not available, it will look for nodes based upon the seed host provider mechanisms available.
Seed host providers may be defined in 3 ways: list based, file based or plugin based. All of these methods provide a list of IP addresses or hostnames which the node should contact in order to obtain a list of master eligible nodes. The node will contact all of these addresses in turn, until either an active master is found, or failing that, until sufficient nodes can be found to elect a new master node.
Examples
The simplest form is to define a list of seed host providers in elasticsearch.yml:
discovery.seed_hosts: - 192.168.1.10:9300 - 192.168.1.11 - seeds.mydomain.com
An alternative way is to refer to a file using the following setting:
discovery.seed_providers: file
The file MUST be placed in the following filepath: $ES_PATH_CONF/unicast_hosts.txt
10.10.10.5 10.10.10.6:9305 10.10.10.5:10005 # an IPv6 address [2001:0db8:85a3:0000:0000:8a2e:0370:7334]:9301
Note that the use of a port is optional. If not used, then the default port range of 9300-9400 will be used.
If you use AWS or GCS then you can install and use a plugin to obtain a list of seed hosts from an API. A plugin also exists for Azure but is deprecated since version 5.
AWS plugin
A typical configuration could be as follows:
discovery.seed_providers: ec2 discovery.ec2.tag.role: master discovery.ec2.tag.environment: dev discovery.ec2.endpoint: ec2.us-east-1.amazonaws.com cloud.node.auto_attributes: true cluster.routing.allocation.awareness.attributes: aws_availability_zone
The above configuration would look for all nodes with a tag called “environment” set to “dev” and a tag called “role” set to “master”, in the AWS zone us-east-1. The last two lines set up cluster routing allocation awareness based upon aws availability zones. (Not necessary, but nice to have).
GCE plugin
A typical configuration could be as follows:
discovery.seed_providers: gce cloud.gce.project_id: <your-google-project-id> cloud.gce.zone: <your-zone> discovery.gce.tags: <my-tag-name>
The above configuration would look for all virtual machines inside your project, zone and with a tag set to the tag name you provide.
Notes and good things to know
Cluster formation depends on correct setup of the network.host settings in elasticsearch.yml. Make sure that the nodes can reach each other across the network using their IP addresses / hostname, and are not getting blocked due to firewall settings on the ports required.
Overview
The queue term in Elasticsearch is used in the context of thread pools. Each node of the Elasticsearch cluster holds various thread pools to manage the memory consumption on that node for different types of requests. The queues come up with initial default limits as per node size but can be modified dynamically using _settings REST endpoint.
What it is used for
Queues are used to hold the pending requests for the corresponding thread pool instead of requests being rejected. For example, if there are too many search requests coming on the node which can not be processed at the same time, the requests are sent to the search thread pool queue.
Examples
Monitoring the thread pools using _cat API:
GET /_cat/thread_pool?v
Get details about each thread pool, including current size:
GET /_nodes/thread_pool
Notes
- Thread pool queues are one of the most important stats to monitor in Elasticsearch as they have a direct impact on the cluster performance and may halt the indexing and search requests.
- The specific thread pool queue size can be changed using its type-specific parameters.
- It is possible to update thread pool queue size dynamically using cluster setting API in version 2.x.
- From Elasticsearch version 5.x onward, it is not possible to update the thread pool settings dynamically via the cluster setting API. Rather, it is a node level setting and it must be configured inside elasticsearch.yml on each node and a node restart is required after the updates.
Common problems
- The most common problem that arises in Elasticsearch related to queues is EsRejectedExecutionException that occurs when queues are full and Elasticsearch nodes cannot keep up with the speed of the requests. This may lead to nodes not responding as well. To deal with this issue, thread pools need continuous monitoring and based on thread pool queue utilization, you may need to review and control the indexing/search requests or increase the resources of the cluster.
- In case of bulk indexing queue rejection, increasing the size of the queue may cause the node to keep more data in memory, which may cause requests taking longer to complete and more heap space to be consumed. As a result you may face impact on cluster performance and stability.
Log Context
Log “Dropping pending state [{}]. more than [{}] pending states.” classname is PendingClusterStatesQueue.java.
We extracted the following from Elasticsearch source code for those seeking an in-depth context :
/** Add an incoming; not yet committed cluster state */
public synchronized void addPending(ClusterState state) {
pendingStates.add(new ClusterStateContext(state));
if (pendingStates.size() > maxQueueSize) {
ClusterStateContext context = pendingStates.remove(0);
logger.warn("dropping pending state [{}]. more than [{}] pending states."; context; maxQueueSize);
if (context.committed()) {
context.listener.onNewClusterStateFailed(new ElasticsearchException("too many pending states ([{}] pending)";
maxQueueSize));
}
}
Find & fix Elasticsearch problems
