How to Run Elasticsearch in Docker (With Persistent Storage)

A single-node Elasticsearch in Docker is the standard dev setup: one container, dev-mode discovery, modest JVM heap, persistent volume. The image is from docker.elastic.co, not Docker Hub, and there are three things that catch first-time users — a Linux kernel setting, the JVM heap default, and the security defaults that turned on in 8.0.

How do I run Elasticsearch in Docker?

docker run -d --name es \
  -e discovery.type=single-node \
  -e xpack.security.enabled=false \
  -e ES_JAVA_OPTS="-Xms512m -Xmx512m" \
  -p 9200:9200 \
  -v es-data:/usr/share/elasticsearch/data \
  docker.elastic.co/elasticsearch/elasticsearch:8.16.0

Single-node mode, security disabled (dev only), 512 MB JVM heap, port 9200 published, data volume mounted. Test it:

curl http://127.0.0.1:9200

The cluster info JSON comes back.

On Linux: fix vm.max_map_count first

Elasticsearch uses memory-mapped files heavily and needs a higher vm.max_map_count than the Linux default. Without this, the container starts but logs an error and refuses to serve requests.

sudo sysctl -w vm.max_map_count=262144

Make it persistent across reboots:

echo 'vm.max_map_count=262144' | sudo tee /etc/sysctl.d/99-elasticsearch.conf

On Docker Desktop (Mac and Windows), this is handled inside Docker's VM — you don't need to set it on the host.

The basic run command

docker run -d --name :container_name 
-e discovery.type=single-node 
-e xpack.security.enabled=false 
-e ES_JAVA_OPTS="-Xms:heap -Xmx:heap" 
-p :host_port:9200 
-v :volume:/usr/share/elasticsearch/data 
docker.elastic.co/elasticsearch/elasticsearch::es_version

Three env vars to know:

• discovery.type=single-node — tells ES not to look for cluster peers. Required for one-container setups.
• xpack.security.enabled=false — disables the auth/TLS features turned on by default in 8.0+. For dev only. With it enabled, the first start prints a generated elastic user password and you need it to connect; convenient for production-shaped local setups, friction for quick experiments.
• ES_JAVA_OPTS="-Xms512m -Xmx512m" — sets the JVM heap (initial = max = 512 MB). The image's default heap can easily eat 1-2 GB on a host that didn't ask for it; pin the size.

Why mount /usr/share/elasticsearch/data

ES stores all indices, shard data, and the cluster state under /usr/share/elasticsearch/data inside the container. Same pattern as the other databases — without a volume, that data is in the container's writable layer and dies with docker rm.

docker stop :container_name && docker rm :container_namedocker run -d --name :container_name 
-e discovery.type=single-node 
-e xpack.security.enabled=false 
-e ES_JAVA_OPTS="-Xms:heap -Xmx:heap" 
-p :host_port:9200 
-v :volume:/usr/share/elasticsearch/data 
docker.elastic.co/elasticsearch/elasticsearch::es_versionIndices are still there.

Practical usage: connecting and indexing

Health check:

curl http://127.0.0.1::host_port/_cluster/health?pretty

Create an index, post a document, search:

curl -X POST http://127.0.0.1::host_port/articles/_doc -H 'Content-Type: application/json' -d '{
"title": "Hello Elasticsearch",
"body": "First document"
}'

curl 'http://127.0.0.1::host_port/articles/_search?pretty&q=Hello'

Pair with Kibana for a UI:

services:
es:
image: docker.elastic.co/elasticsearch/elasticsearch::es_version
environment:
discovery.type: single-node
xpack.security.enabled: "false"
ES_JAVA_OPTS: "-Xms:heap -Xmx:heap"
volumes:
:volume:/usr/share/elasticsearch/data
ports:
"::host_port:9200"
kibana:
image: docker.elastic.co/kibana/kibana::es_version
depends_on:
es
ports:
"5601:5601"
volumes:
:volume:

Kibana on http://127.0.0.1:5601 after both containers start.

Common pitfalls

• vm.max_map_count too low — set it (see Linux section above). Docker Desktop handles this; bare-metal Linux does not.
• OOM kills. Set ES_JAVA_OPTS="-Xms512m -Xmx512m" (or larger for real workloads). Without the cap, the JVM can grab gigabytes.
• Security blocking connections in 8.x. Either set xpack.security.enabled=false for dev, or accept the generated password from the first-start logs.
• Publishing 9200 to the whole internet. -p 9200:9200 binds to 0.0.0.0, every interface, and Docker can punch that past your host firewall. On a box with a public IP, an unauthenticated node is then reachable from the internet and gets found and wiped by automated bots within hours. Bind to 127.0.0.1:9200:9200 for anything that is not purely local. I learned this the expensive way: how an open Elasticsearch port got my database ransomed.
• Pulling from Docker Hub instead of docker.elastic.co. The Hub still has an old elasticsearch mirror; the current ES is only on docker.elastic.co. Same with Kibana.
• Apple Silicon and older ES. The arm64 image arrived in 7.17+; pre-7.17 ES on Apple Silicon needs --platform linux/amd64 and emulation.

What to do next

• How to Use ElasticPress with WP_Query — pair this with WordPress.
• Elasticsearch Cheat Sheet — once it's running, the query-side reference.
• How to reindex Elasticsearch with zero downtime: when you need to change a mapping, the alias-and-reindex dance that avoids dropping and recreating the index.
• Elasticsearch vs OpenSearch for ElasticPress — if you were considering OpenSearch instead.
• Elasticsearch ransomware: how an open port wiped my database — before you publish 9200, read what happens when it is reachable from the internet.

FAQ