Elasticsearch memory requirements
The Elasticsearch process is very memory intensive. Elasticsearch uses a JVM (Java Virtual Machine), and best practices recommend allocating close to 50% of the memory available on a node to the JVM. The JVM relies on memory so that the Lucene process can access index values very fast. The other 50% is left to the file system cache which keeps data that is regularly accessed in memory.
For a full explanation of JVM memory management, read this blog. It is also common to receive warnings from the different types of circuit breakers.
jvm.mem
The most important memory section is the JVM heap.
GET _nodes/stats/jvmThe output looks like this:
"jvm" : {
"timestamp" : 1603351829573,
"uptime_in_millis" : 150932107,
"mem" : {
"heap_used_in_bytes" : 258714272,
"heap_used_percent" : 24,
"heap_committed_in_bytes" : 1073741824,
"heap_max_in_bytes" : 1073741824,
"non_heap_used_in_bytes" : 192365488,
"non_heap_committed_in_bytes" : 209186816,Note that the heap_used_in_bytes in a healthy JVM will follow a saw tooth pattern due to the garbage collection process, increasing steadily up to around 70%, then reducing sharply to 30% when garbage collection occurs.
The JVM heap_max_in_bytes will depend on the value set in jvm.options file, and you should set it to be 50% of the RAM available for your container or server.
For more information on JVM heap issues, please see: Heap size usage and JVM garbage collection in Elasticsearch.
Explaining the different types of memory statistics
When looking at the memory statistics, we need to be aware that many Elasticsearch applications are running inside containers on much larger machines. This is typical if you are using a hosted service such as AWS Elasticsearch service or Elastic cloud, or if you are running Elasticsearch on Docker or Kubernetes. In such cases, it’s important to be careful to interpret the memory statistics available to us.
There are various memory statistics available from the Elasticsearch monitoring APIs, as explained below:
GET _nodes/stats/os
"os" : {
"timestamp" : 1603350306857,
"cpu" : {
"percent" : 13,
"load_average" : {
"1m" : 3.37,
"5m" : 3.18,
"15m" : 2.8
}
},
"mem" : {
"total_in_bytes" : 16703369216,
"free_in_bytes" : 361205760,
"used_in_bytes" : 16342163456,
"free_percent" : 2,
"used_percent" : 98
},
"swap" : {
"total_in_bytes" : 1023406080,
"free_in_bytes" : 1302528,
"used_in_bytes" : 1022103552
},
"cgroup" : {
"cpuacct" : {
"control_group" : "/",
"usage_nanos" : 2669636420088
},
"cpu" : {
"control_group" : "/",
"cfs_period_micros" : 100000,
"cfs_quota_micros" : -1,
"stat" : {
"number_of_elapsed_periods" : 0,
"number_of_times_throttled" : 0,
"time_throttled_nanos" : 0
}
},
"memory" : {
"control_group" : "/",
"limit_in_bytes" : "9223372036854771712",
"usage_in_bytes" : "4525641728"
}
}
}The above statistics are for a development node running on Docker. Let’s interpret the sections we received here:
os.mem
The first section “os.mem” refers to the host server where the machine is running. In this case we are running on Docker, so 16GB refers to the memory of the host machine where the container is running. Note that it is quite normal for a machine to be using close to 100% of its memory, and this does not indicate a problem.
os.swap
The “os.swap” section again refers to the host machine. In this case, we can see that the host machine allows swapping. This is quite normal when we are running a container inside a host with swapping still enabled. We can double check that inside the Docker container no swap is permitted by running:
GET _nodes?filter_path=**.mlockallos.cgroup
Finally, we can see the “os.cgroup” section which refers to the container itself. In this case, the container is using 4GB of memory.
process.mem
We also have access to a statistic virtual memory:
GET _nodes/stats/process
"process" : {
"timestamp" : 1603352751181,
"open_file_descriptors" : 436,
"max_file_descriptors" : 1048576,
"cpu" : {
"percent" : 0,
"total_in_millis" : 1964850
},
"mem" : {
"total_virtual_in_bytes" : 5035376640
}Note that “process.total_virtual_in_bytes” includes the entire memory available to the process, including virtual memory available via mmap.
Ready to try this out on your own? Start a free trial.
Want to get Elastic certified? Find out when the next Elasticsearch Engineer training is running!
Related content
September 3, 2025
Vector search filtering: Keep it relevant
Performing vector search to find the most similar results to a query is not enough. Filtering is often needed to narrow down search results. This article explains how filtering works for vector search in Elasticsearch and Apache Lucene.
August 26, 2025
Lighter by default: Excluding vectors from source
Elasticsearch now excludes vectors from source by default, saving space and improving performance while keeping vectors accessible when needed.
August 13, 2025
Failure store: see what didn’t make it
Learn about failure store, a new feature in the Elastic Stack that captures and indexes previously lost events.
July 29, 2025
Introducing a more powerful, resilient, and observable ES|QL in Elasticsearch 8.19 & 9.1
Exploring ES|QL enhancements in Elasticsearch 8.19 & 9.1, including built-in resilience to failures, new monitoring and observability capabilities, and more.
July 29, 2025
Unify your data: Cross-cluster search with ES|QL is now generally available!
Cross-Cluster search with ES|QL is now GA! Query data across multiple clusters with a single, elegant query. Learn about its performance, resilience, and syntax.