# Config shard versus dedicated config server > Source-anchored research for Config shard versus dedicated config server. Run status: COMPLETED. Parent: [Shard Key Design and Resharding as Cost Levers](https://llms-explorer.com/tree/shard-key-design-and-resharding-as-cost-levers/) · 4 facets · 20 facts · page: https://llms-explorer.com/tree/config-shard-versus-dedicated-config-server/ ## How it works - In the Atlas pages retrieved on October 30, 2026, Atlas-managed configuration uses embedded config servers at ten or fewer shards and dedicated config servers above ten; shard-count changes can trigger reselection. The pages list feature exceptions, so shard count alone does not guarantee conversion. (medium confidence) (verified as of 2026-10-30) — [source](https://www.mongodb.com/docs/atlas/cluster-additional-settings/) - Atlas's additional-settings page restricts snapshot restore destinations to the same config-server type as the source: dedicated to dedicated and embedded to embedded. Include restore compatibility when evaluating a topology change. (low confidence) (verified as of 2026-10-30) — [source](https://www.mongodb.com/docs/atlas/cluster-additional-settings/) - Atlas HTML guidance retrieved on 2026-10-30 selects embedded config servers at ten or fewer shards and dedicated config servers above ten when Atlas-managed config servers are enabled and the cluster is eligible. A 7.0-to-8.0 upgrade enables management without itself changing the existing topology. (medium confidence) (verified as of 2026-10-30) — [source](https://www.mongodb.com/docs/atlas/cluster-additional-settings/) ## Problems, failure modes and limitations - The server architecture README uses Config Shard as a synonym for CSRS, while the operator manual uses config shard specifically for storing application data with metadata. Code terminology alone therefore cannot establish that a deployment uses embedded application-data storage. (medium confidence) — [source](https://github.com/mongodb/mongo/blob/master/src/mongo/db/sharding_environment/README.md) ## Comparisons and alternatives - SERVER-73865 records the internal decision to use the shard id config instead of catalogShard and remove special cases as config servers gained shard components. This historical identity change does not itself prove application data is present on a config server. (low confidence) — [source](https://jira.mongodb.org/browse/SERVER-73865) - An Atlas embedded-to-dedicated transition drains application data from the config shard while retaining metadata there. Atlas adds a replacement data shard to preserve the target shard count; the operation therefore changes the resource mix rather than simply relabeling a shard. (low confidence) (verified as of 2026-10-30) — [source](https://www.mongodb.com/docs/atlas/transition-to-dedicated-config-servers/) ## Facts and statements - A config shard stores application data alongside cluster metadata; a dedicated config server stores metadata separately from application shards. Both provide the required config-server role. (high confidence) — [source](https://www.mongodb.com/docs/manual/core/sharded-cluster-config-servers.md) - Embedding the config-server role in an application shard avoids an additional replica set and reduces infrastructure cost. Dedicated config servers add a separate replica set; these sources provide no universal dollar or percentage saving. (high confidence) — [source](https://www.mongodb.com/docs/atlas/cluster-additional-settings/) - MongoDB's manual says config shards have no measurable performance impact at low shard counts and dedicated config servers isolate metadata from application data. The retrieved passage supplies neither a numeric low-count boundary nor a benchmark methodology, so it does not establish performance equivalence for a particular workload. (low confidence) — [source](https://www.mongodb.com/docs/manual/core/sharded-cluster-config-servers.md) - Atlas's embedded-to-dedicated conversion drains user data from the config shard while retaining its metadata. Atlas adds a replacement application shard to preserve the target shard count and scales the dedicated config replica set to M30 in its documented example. (low confidence) (verified as of 2026-10-30) — [source](https://www.mongodb.com/docs/atlas/transition-to-dedicated-config-servers/) - Atlas describes the conversion as online but warns that receiving shards need CPU, memory and I/O headroom, large datasets can require hours to days, and the running transition blocks other long-running plans such as tier or storage changes. Capacity must therefore be prepared before starting the conversion. (low confidence) (verified as of 2026-10-30) — [source](https://www.mongodb.com/docs/atlas/transition-to-dedicated-config-servers/) - Atlas documents automatic-transition restrictions for Search/Vector Search, unsharded time-series collections and unsharded Queryable Encryption collections. Global Clusters always use dedicated config servers. Feature compatibility can therefore outweigh the smaller embedded infrastructure footprint. (medium confidence) (verified as of 2026-10-30) — [source](https://www.mongodb.com/docs/atlas/cluster-additional-settings/) - Percona's July 30, 2026 announcement describes its MongoDB 8.3 release as a technical preview for testing, and reports granular embedded-to-dedicated transition commands plus a replicaSetConfigShardMaintenanceMode path that avoids first building a separate dedicated config-server replica set. This is release-specific evidence, not a production-support guarantee. (low confidence) (verified as of 2026-07-30) — [source](https://www.percona.com/blog/percona-server-for-mongodb-8-3-technical-preview-is-now-available/) - Historical bug SERVER-88172 allowed explicitly naming config in enableSharding or moveRange on clusters with dedicated config servers, leading to invalid metadata or crashes. The ticket records fixes in 7.3.3 and 7.0.12; this is evidence to check affected patch levels, not evidence that embedding is generally unsafe. (low confidence) — [source](https://jira.mongodb.org/browse/SERVER-88172) - Starting in MongoDB 8.0, a config shard can hold application data alongside cluster metadata; dedicated config servers keep metadata separate from application data. (medium confidence) — [source](https://www.mongodb.com/docs/manual/core/sharded-cluster-config-servers.md) - Percona identifies smaller clusters as the main cost-saving use case and recommends separate roles for large clusters, where the relative cost benefit becomes small. MongoDB scopes its claim of no measurable performance impact to low shard counts. (medium confidence) — [source](https://www.percona.com/blog/scaling-smarter-what-you-have-missed-in-mongodb-8-0/) - The self-managed commands transitionFromDedicatedConfigServer and transitionToDedicatedConfigServer change from dedicated to embedded mode and from embedded to dedicated mode, respectively. (medium confidence) — [source](https://www.mongodb.com/docs/manual/core/sharded-cluster-config-servers.md) - Atlas Search/Vector Search can pin the config-server type; unsharded queryable-encryption collections can block automatic transition. Global Clusters require dedicated config servers. Check feature eligibility before assuming Atlas will automatically switch topology. (medium confidence) (verified as of 2026-10-30) — [source](https://www.mongodb.com/docs/atlas/cluster-additional-settings/) - Dedicated describes the config-server replica-set role, not necessarily exclusive physical hardware: Xillio's MongoDB 8 deployment guide runs a separate config replica set on the application/router host while keeping documents on two other shard hosts. Its one-member replica sets explicitly trade redundancy for a performance-focused setup. (low confidence) (verified as of 2026-10-16) — [source](https://docs.xill.io/xill4/advanced/mongodb-horizontal-scaling/) - Inspect listShards through mongos for an entry whose _id is config to confirm config-shard mode. Atlas cautions that replica-set names containing config or shard do not reliably identify which data they store. (medium confidence) — [source](https://www.mongodb.com/docs/manual/core/sharded-cluster-config-servers.md)