Plans and limits

Compute plans

A plan sets each server's CPU and memory, the size of each server's volume, and the number of replicas. Memory is reserved in full: a server's memory request equals its limit. These are the default plans; your installation can add its own.

Plan	CPU (request–limit)	Memory	Storage per server	Replicas
nano	0.1–0.25	512 MiB	5 GiB	1
small	0.25–0.5	1 GiB	10 GiB	1
medium	0.5–1	2 GiB	25 GiB	1
large	1–2	4 GiB	50 GiB	3
xlarge	2–4	8 GiB	100 GiB	3
2xlarge	4–8	16 GiB	200 GiB	3

medium and larger are marked as meant for production. Three replicas give three servers; whether a collection keeps more than one copy of its data is its own setting — see Replicas and sharding.

A boundary can be limited to some plans by a policy, and a plan larger than your installation's nodes can host is refused.

Names

• 3 to 50 characters in the portal: lowercase letters, digits and hyphens, not starting or ending with a hyphen. The API accepts up to 63.
• Unique among the vector stores in the resource group.
• Fixed after creation.

Replicas and storage

• 1 to 5 replicas.
• Each replica has its own volume. Its size comes from the plan when the vector store is created and does not change with the plan later.

Versions

• The version is a tag of the qdrant/qdrant image and must start with a version number, such as v1.13.2.
• The portal creates every vector store with v1.13.2; change it afterwards in Runtime.

Ports

Port	Protocol	Where
6333	REST (HTTP)	Inside the Kubernetes cluster, and on the load balancer
6334	gRPC	Inside the Kubernetes cluster, and on the load balancer
443	REST over HTTPS, and the dashboard	On the HTTPS ingress

What is not offered

• Backups, snapshots and restore by the platform — see Back up a collection.
• TLS on the internal endpoints and on the load balancer.
• Access rules per collection: the API key opens the whole vector store.