Search & Vector Search Resource Planning and Sizing

Choose a Workload Class

The CPU-to-memory ratio you allocate depends on your search workload profile.

For most general use cases, a small or medium High-CPU configuration is a balanced starting point.

Pick a Starting Size

Size your mongot pods by expected vector data volume (Low-CPU) or by queries-per-second (High-CPU).

For example, if you expect to handle 100 queries per second for a full-text search application, start with a Medium High-CPU configuration.

Configure CPU and Memory

Set the CPU and memory for mongot pods in spec.clusters[].resourceRequirements. The requests field reserves capacity on the node; the limits field caps how much the pod can consume.

spec:
  clusters:
    - resourceRequirements:
        requests:
          cpu: "2"
          memory: 4Gi
        limits:
          cpu: "3"
          memory: 5Gi

If you omit spec.clusters[].resourceRequirements, the Kubernetes Operator uses the following defaults:

• requests.cpu: 2

• requests.memory: 4Gi

• No limits — the pod can consume the node's full available resources.

Set limits to match your workload. A pod without limits can saturate the node and impact other workloads.

Configure the JVM Heap

When you don't specify -Xms or -Xmx in spec.clusters[].jvmFlags, the Kubernetes Operator auto-calculates the JVM heap by setting both flags to half of spec.clusters[].resourceRequirements.requests.memory.

To override the heap explicitly, set -Xms and -Xmx in spec.clusters[].jvmFlags. The Kubernetes Operator does not modify any flags you provide and appends them after operator-computed flags.

spec:
  clusters:
    - jvmFlags:
        - -Xms2g
        - -Xmx2g

Estimate Index Size

A collection's size and the resulting search index's size are not always correlated. Index size depends on the fields you map and the features (such as autocomplete) that you enable on the index. To estimate the index size for your workload:

Size the Persistent Volume to be Deployed

Allocate double the disk space your index requires. The extra space allows mongot to rebuild indexes when needed. mongot becomes read-only when disk usage reaches 90%.

Set the volume size in spec.clusters[].persistence.single.storage using binary quantity suffixes:

spec:
  clusters:
    - persistence:
        single:
          storage: 60Gi
          storageClass: local-nvme

If you omit spec.clusters[].persistence, the Kubernetes Operator provisions a default volume of 16 GB. For a 30 GB index, set spec.clusters[].persistence.single.storage to 60Gi to leave room for index rebuilds.

Choose a Storage Class

Reference the StorageClass for the volume in spec.clusters[].persistence.single.storageClass. Pick a class that meets the following guidance:

• Disk type: Use general-purpose SSD-backed storage. Both read and write IOPS are important for mongot performance. Replication involves disk writes for new index segments and disk reads when mongot merges old segments into larger ones.

• Read-only threshold: When a volume reaches 90% usage, mongot enters read-only mode and stops write operations. To resume writes, delete index data to bring usage below 85%.