# Atlas Search node scaling per shard > Source-anchored research for Atlas Search node scaling per shard. Run status: COMPLETED. Parent: [Independent shard scaling and per-shard tiering](https://llms-explorer.com/tree/independent-shard-scaling-and-per-shard-tiering/) · 4 facets · 10 facts · page: https://llms-explorer.com/tree/atlas-search-node-scaling-per-shard/ ## How it works - For a uniform dedicated Search deployment, the selected Search Node count applies to each shard: the official example configures two nodes with three shards and provisions six nodes. Treat the node count as a per-shard multiplier when estimating capacity. (high confidence) (verified as of 2026-10-27) — [source](https://www.mongodb.com/docs/search/deployment/deployment-options.md) - A customer request dated February 8, 2025 asks for independent Search Node allocation because only selected shards would need search capacity. Its proposed configurations include no Search Nodes on two shards and S40 on a third, or smaller tiers on two shards and S40 on a third. This is evidence of a requested cost optimization, not proof of current service-wide support or non-support. (low confidence) (verified as of 2025-02-08) — [source](https://feedback.mongodb.com/ideas/CUSTOMER-I-6138) - The Search deployment guide explains that an index working set larger than available Search Node RAM causes paging and disk I/O; selecting a larger-memory tier can reduce that pressure. It does not promise that adding identical Search Nodes reduces the index working set each node must serve. (low confidence) (verified as of 2026-10-27) — [source](https://www.mongodb.com/docs/search/deployment/deployment-options.md) - Datadog documents Atlas Search gauges for index size, required vector-index memory, JVM memory, and search-process resident memory, plus query counters and replication-lag metrics. These can inform capacity investigation before increasing Search Node count or tier; they do not establish that throughput scales linearly with purchased nodes. (low confidence) (verified as of 2026-10-14) — [source](https://docs.datadoghq.com/integrations/mongodb-atlas/) ## How-to and procedures - Use observed search index disk size and search process memory when assessing capacity pressure: Atlas exposes Search Index Size and Search Process Memory charts, and Datadog exposes corresponding Atlas search index-size and resident-memory metrics. These signals provide measurements for sizing; they do not themselves establish a throughput scaling ratio. (medium confidence) (verified as of 2026-10-14) — [source](https://www.mongodb.com/docs/atlas/review-atlas-search-metrics/) ## Problems, failure modes and limitations - The Create Search Nodes API documentation describes tier autoscaling within deployment-wide bounds, evaluated independently for each shard and region. It exposes shardId in effectiveSpecs and baselineSpecs responses. The same page says automatic tier scale-down is not yet supported and enabling it returns an error; this is documentation evidence, not a live deployment test. (low confidence) (verified as of 2026-10-30) — [source](https://www.mongodb.com/docs/api/doc/atlas-admin-api-v2/2024-11-13/operation/operation-creategroupclustersearchdeployment) ## Facts and statements - Derived cost model for one region and a uniform tier: Search Node compute cost equals shard count × Search Nodes per shard × hourly price per node × billed hours. This excludes data-transfer charges; billing documentation meters each Search Node hourly and includes traffic between Search Nodes and database nodes. (high confidence) (verified as of 2026-10-14) — [source](https://www.mongodb.com/docs/search/deployment/deployment-options.md) - MongoDB documents horizontal scaling as changing the number of Search Nodes, with 2–32 nodes per region per shard, and vertical scaling as changing the search tier and its CPU, RAM and storage configuration. (medium confidence) (verified as of 2026-10-27) — [source](https://www.mongodb.com/docs/search/deployment/deployment-options.md) - The Terraform mongodbatlas_search_deployment resource allows one search deployment per cluster and currently limits specs to one element containing instance_size and node_count. The documented resource does not expose a per-shard Search Node sizing map. (medium confidence) (verified as of 2026-10-30) — [source](https://raw.githubusercontent.com/mongodb/terraform-provider-mongodbatlas/master/docs/resources/search_deployment.md) - The documented horizontal scaling ceiling is 32 Search Nodes per region per shard. Increasing Search Node count distributes query load over more Search Nodes; selecting a different search tier changes per-node resources instead. (high confidence) (verified as of 2026-10-27) — [source](https://www.mongodb.com/docs/search/deployment/deployment-options.md)