Read one level up: https://skills.qdrant.tech/md/documentation/cloud

> Explore Qdrant's agent skills catalog at https://skills.qdrant.tech/
> Search the documentation at https://search.qdrant.tech/md/documentation?q=your+query+here
> Use this file for an index of the main pages: https://qdrant.tech/llms.txt
> Content inside `<only-applies-to>` tags applies only to the named products, or only as of the named version. All other content applies to every Qdrant product. If you know which product the user is on, ignore tagged content for other products.


# Scaling Qdrant Cloud Clusters

The amount of data is always growing and at some point you might need to change the capacity of your cluster. You can easily scale your Qdrant cluster up or down from the Cluster detail page in the Qdrant Cloud console.

![Cluster Scaling](/md/documentation/cloud/cluster-scaling.png)

## Vertical Scaling

Vertical scaling is the process of increasing the capacity of a cluster by adding or removing CPU, storage and memory resources on each database node.

You can start with a minimal cluster configuration and scale it up over time to accommodate the growing amount of data in your application. If your cluster consists of several nodes each node will need to be scaled to the same size. 

Changing the CPU or memory restarts each node, whether you scale up or down. If the collections in your cluster are not replicated, this causes a short downtime. Increasing only the disk size does not restart the nodes. See [Restarts and Downtime](https://qdrant.tech/documentation/cloud/configure-cluster/index.md#restarts-and-downtime).

When you scale down, the new memory size must still fit all the data in the cluster. Otherwise, the cluster can run out of memory and become unavailable until you scale it back up. The Qdrant Cloud Console and the `qcloud` CLI warn you when your cluster's recent peak memory usage, plus a buffer, is more than the new configuration provides, or when current usage can't be determined. The warning doesn't block the change. Other tools, such as the [Qdrant Cloud API](https://qdrant.tech/documentation/cloud-api/index.md), Terraform, or Pulumi, apply the change without a warning. Before you scale down with them, check your cluster's memory usage. See [Monitor Clusters](https://qdrant.tech/documentation/cloud/cluster-monitoring/index.md).

### Scaling Down Disk Space

You can reduce the disk size of your cluster nodes as well as increase it. Cloud providers don't support shrinking a disk in place, so Qdrant Cloud replaces the nodes instead. For each node, Qdrant Cloud:

1. Creates a new node with the smaller disk.
2. Replicates all data from the old node to the new node.
3. Shuts down the old node with the larger disk once replication is complete.

Qdrant Cloud repeats this one node at a time until every node in the cluster has the new disk size. The process doesn't cause downtime, but keep the following in mind:

- The new disk size must still fit all the data on each node, with room to grow. Check your cluster's current disk usage before you scale down. See [Monitor Clusters](https://qdrant.tech/documentation/cloud/cluster-monitoring/index.md).
- Replacing nodes can take a while, depending on how much data your cluster holds.
- Replicating data to the new nodes increases the disk I/O load on your cluster until the process completes.

### Changing Disk Speed

On AWS, you can change the IOPS and throughput of your cluster's disks by picking a different disk speed tier. Each tier has more IOPS and throughput than the one before it:

- **Cost-Optimized** (the default): the lowest IOPS and throughput. For test clusters and small production clusters.
- **Balanced**: several times the IOPS and throughput of Cost-Optimized. For medium-sized clusters.
- **Performance**: the most IOPS and throughput. For large clusters with a high write load.

Each tier requires a minimum disk size.

The more your workload reads from and writes to disk, the more it gains from a faster tier. This applies to write-heavy workloads, bulk loads, and configurations with a low RAM cache ratio, such as vectors in the `cold` [memory tier](https://qdrant.tech/documentation/ops-configuration/memory-tiers/index.md). If disk I/O isn't the bottleneck, Cost-Optimized is enough.

Pick the tier under **Disk Speed** in the Qdrant Cloud console, or pass `--disk-performance` with `balanced`, `cost-optimised`, or `performance` to `qcloud cluster scale` or `qcloud cluster update`. Changing the tier doesn't restart the nodes. See [Restarts and Downtime](https://qdrant.tech/documentation/cloud/configure-cluster/index.md#restarts-and-downtime).

To decide whether your workload needs faster disks, see [Storage-Focused Configuration](https://qdrant.tech/documentation/capacity-planning/index.md#storage-focused-configuration).

## Horizontal Scaling

Vertical scaling can be an effective way to improve the performance of a cluster and extend the capacity, but it has some limitations. The main disadvantage of vertical scaling is that there are limits to how much a cluster can be expanded. At some point, adding more resources to a cluster can become impractical or cost-prohibitive.

In such cases, horizontal scaling may be a more effective solution.

Horizontal scaling is the process of increasing the capacity of a cluster by adding more nodes and distributing the load and data among them. The horizontal scaling at Qdrant starts on the collection level. You have to choose the number of shards you want to distribute your collection around while creating the collection.  Please refer to the [sharding documentation](https://qdrant.tech/documentation/scaling/distributed_deployment/index.md#sharding) section for details.

When scaling up horizontally, the cloud platform will automatically rebalance all available shards across nodes to ensure that the data is evenly distributed. See [Configuring Clusters](https://qdrant.tech/documentation/cloud/configure-cluster/index.md#shard-rebalancing) for more details.

When scaling down horizontally, the cloud platform will automatically ensure that any shards that are present on the nodes to be deleted, are moved to the remaining nodes.

Important: One shard can not be split across nodes. So, if you configure 2 shards for a collection, but then scale your cluster from 1 to 3 nodes, your cluster nodes can't be fully utilized. The cloud platform will automatically rebalance your shards, so that two nodes will have one shard each, but the third node will not have any shards at all. You can use the [resharding feature](https://qdrant.tech/documentation/cloud/cluster-scaling/index.md#resharding) to change the number of shards in an existing collection. Once resharding is complete, the cloud platform will rebalance the shards across all nodes, ensuring that all nodes are utilized.

We will be glad to consult you on an optimal strategy for scaling.

[Let us know](https://qdrant.tech/documentation/support/index.md) your needs and decide together on a proper solution.

<only-applies-to products="Available as of v1.13.0.">

## Resharding

<aside role="status">Resharding is exclusively available on multi-node clusters across our <a href="/documentation/deploy-intro/">cloud</a> offering, including <a href="/documentation/hybrid-cloud/">Hybrid</a> and <a href="/documentation/private-cloud/">Private</a> Cloud.</aside>

When creating a collection, it has a specific number of shards. The ideal number of shards might change as your cluster evolves.

Resharding allows you to change the number of shards in your existing collections, both up and down, without having to recreate the collection from scratch.

Resharding is a transparent process, meaning that the collection is still available while resharding is going on without having downtime. This allows you to scale from one node to any number of nodes and back, keeping your data perfectly distributed without compromise.

To increase the number of shards (reshard up), use the [Update collection cluster setup API](https://api.qdrant.tech/master/api-reference/distributed/update-collection-cluster) to initiate the resharding process:

```http
POST /collections/{collection_name}/cluster
{
    "start_resharding": {
        "direction": "up",
        "shard_key": null
    }
}
```

To decrease the number of shards (reshard down), you may specify the `"down"` direction.

The current status of resharding is listed in the [collection cluster info](https://api.qdrant.tech/v-1-12-x/api-reference/distributed/collection-cluster-info) which can be fetched with:

```http
GET /collections/{collection_name}/cluster
```

We always recommend to run an ongoing resharding operation till the end. But, if at any point the resharding operation needs to be aborted, you can use:

```http
POST /collections/{collection_name}/cluster
{
    "abort_resharding": {}
}
```

A few things to be aware of with regards to resharding:

- during resharding, performance of your cluster may be slightly reduced
- during resharding, reported point counts will not be accurate
- resharding may be a long running operation on huge collections
- you can only run one resharding operation per collection at a time


</only-applies-to>