Elasticsearch container OOMKilled / heap too small in CI

The container was killed by the kernel out-of-memory killer because the JVM tried to reserve more heap than the container was allowed. Elasticsearch sizes heap from available memory by default, and a constrained CI container hits the limit before the node is even ready.

What this error means

The Elasticsearch service exits with status 137, the runner shows "OOMKilled", or the node dies mid-startup after logging heap allocation. Health checks never pass.

elasticsearch exited with code 137
State:  Terminated
Reason: OOMKilled

Common causes

Default heap exceeds the container memory limit

Elasticsearch auto-sizes the JVM heap from the memory it sees, which can be larger than the cgroup limit imposed on the service container.

No explicit heap sizing for the CI environment

Without ES_JAVA_OPTS pinning -Xms/-Xmx, the node reserves more than the constrained runner can back, and the kernel kills it.

How to fix it

Pin the heap with ES_JAVA_OPTS

1. Set a fixed, modest heap such as 512m for tests.
2. Keep -Xms and -Xmx equal so the JVM does not grow the heap at runtime.
3. Ensure the container memory limit is at least double the heap for off-heap needs.

services:
  elasticsearch:
    image: docker.elastic.co/elasticsearch/elasticsearch:8.13.4
    env:
      discovery.type: single-node
      ES_JAVA_OPTS: -Xms512m -Xmx512m

Give the container enough memory

On OpenSearch use OPENSEARCH_JAVA_OPTS and make sure the runner has headroom above the heap for the JVM, page cache, and OS.

env:
  OPENSEARCH_JAVA_OPTS: -Xms512m -Xmx512m

How to prevent it

• Always set an explicit -Xms/-Xmx heap for CI service containers.
• Provision container memory to at least twice the configured heap.
• Use the smallest heap that still passes your integration tests.

Related guides