Milvus Query Node Out of Memory (OOM)

How to fix it

Try these in order — the most common fix is first.

1. 1. Reduce what is loaded

   Copy steps
   1. Use load_fields to load only needed fields (beta, not recommended for production per docs)
   2. Set skip_load_dynamic_field if you do not use $meta keys
   3. Release collections you are not using
   Did this fix it? Yes No

2. 2. Use a lighter index

   Copy steps
   1. Move to IVF_SQ8 (third-party guide says about 4x less memory), IVF_PQ, or DISKANN for data larger than RAM
   2. Rebuild the index and reload
   Did this fix it? Yes No

3. 3. Size Kubernetes resources

   Copy steps
   1. Set resources.requests and limits for queryNode in the Helm values (docs example: 16Gi limit, 4Gi request)
   2. helm upgrade ... --reuse-values -f resources.yaml
   3. Add query nodes
   Did this fix it? Yes No

Symptoms

Query node killed, load fails with OOM, loaded collections show 0 after a kill.

Why it happens

• Collection larger than node memory
• Float vectors with HNSW use lots of RAM
• Index builds peaking alongside searches
• No Kubernetes resource limits or limits too low

This guide also covers

• milvus out of memory
• milvus querynode oom
• milvus load segment failed oom

The same problem in other apps

All slow, laggy or resource-hungry guides →

• Weaviate Out of Memory (OOM Killed) (Weaviate)
• Qdrant Out of Memory (OOMKilled / High RAM Usage) (Qdrant)

Other Milvus errors

• Milvus "collection not loaded" Error
• Milvus "field dimension mismatch" Error
• Milvus Connection Refused / Timeout (19530)
• Milvus etcd and MinIO Dependency Errors
• Why Is Milvus Search Slow?
• Milvus "index not exist" Error
• Milvus Search Returns Empty Results After Insert
• How to Cancel Zilliz Cloud (Milvus) Subscription