<!-- llms-explorer concept facts · https://llms-explorer.com/tree/one-way-tier-gates-sharding-global-writes-edition-changes/ · pack 2026-10-31 · ~1575 tokens -->

# One-way tier gates: sharding, Global Writes, edition changes

> Source-anchored research for One-way tier gates: sharding, Global Writes, edition changes. Run status: COMPLETED.

Parent: [Cluster Tier Limits and Feature Gates as Cost Drivers (Connections, IOPS Caps, Minimum Tiers)](https://llms-explorer.com/tree/cluster-tier-limits-and-feature-gates-as-cost-drivers-connections-iops-caps-minimum-tiers/) · 3 facets · 16 facts · page: https://llms-explorer.com/tree/one-way-tier-gates-sharding-global-writes-edition-changes/

## How it works

- Alibaba DTS warns against shardCollection, reshardCollection, unshardCollection, moveCollection and movePrimary on the source during a sharded-cluster migration because these operations can cause data inconsistency. A migration alternative therefore needs its own operational plan. (low confidence) (verified as of 2026-09-30) — [source](https://www.alibabacloud.com/help/en/dts/user-guide/precautions-and-limits-for-migrating-data-from-a-mongodb-database)
- The MongoDB 8.0 unsharding procedure says an affected Atlas Search index becomes unavailable when unsharding finishes and must be manually rebuilt; capacity planning should therefore account for a rebuild when this operation is chosen. (low confidence) (verified as of 2026-10-30) — [source](https://www.mongodb.com/docs/v8.0/tutorial/unshard-collection/)

## Problems, failure modes and limitations

- Global Writes requires an M30-or-larger sharded cluster. Once deployed, Global Writes cannot be disabled and the Global Cluster cannot be converted to a standard sharded cluster. (high confidence) (verified as of 2026-10-30) — [source](https://www.mongodb.com/docs/atlas/create-database-deployment.md)
- The database edition of an existing cluster cannot be changed in either direction between Atlas Core and Atlas Infinite. (high confidence) (verified as of 2026-10-30) — [source](https://www.mongodb.com/docs/atlas/scale-cluster.md)

## Facts and statements

- Atlas Core replica-set conversion to a sharded cluster requires M30 or larger; after conversion Atlas does not support converting that existing cluster back into a replica set. (high confidence) (verified as of 2026-10-30) — [source](https://www.mongodb.com/docs/atlas/scale-cluster.md)
- Global Clusters with Global Writes enabled do not support dedicated Search Nodes. This compatibility gate must be checked before treating dedicated Search capacity as an available remedy for a workload. (high confidence) (verified as of 2026-10-30) — [source](https://www.mongodb.com/docs/atlas/create-database-deployment.md)
- The Global Cluster creation workflow fixes the choice between Atlas-Managed Sharding and Self-Managed Sharding after deployment. (low confidence) (verified as of 2026-10-30) — [source](https://www.mongodb.com/docs/atlas/tutorial/create-global-cluster.md)
- Atlas Infinite public preview excludes sharded and Global Clusters, Atlas Search and Vector Search, and cross-edition live migration or backup/restore between Core and Infinite. Protocol compatibility therefore does not establish a supported edition-conversion path. (low confidence) (verified as of 2026-10-30) — [source](https://www.mongodb.com/docs/atlas/infinite/atlas-infinite-landing.md)
- The in-place Atlas conversion restriction is narrower than an absolute prohibition on moving data to a replica set: Percona ClusterSync documents sharded-source to replica-set-target replication, while Alibaba DTS documents migration from a MongoDB sharded cluster to a replica set. (high confidence) (verified as of 2026-10-30) — [source](https://docs.percona.com/percona-clustersync-for-mongodb/deployment.html)
- Percona ClusterSync 1.0.0 supports sharded-source to replica-set-target replication and creates source-sharded collections as ordinary target collections. Its documented Atlas-source target is Percona Server for MongoDB; this is not evidence of an Atlas-to-Atlas in-place reversal. (low confidence) (verified as of 2026-10-13) — [source](https://docs.percona.com/percona-clustersync-for-mongodb/deployment.html)
- Atlas Core supports sharded clusters at M30 or larger; sharding therefore imposes a database-tier floor distinct from choosing Search capacity. (medium confidence) (verified as of 2026-10-30) — [source](https://www.mongodb.com/docs/atlas/scale-cluster.md)
- After an Atlas replica set is converted into a sharded cluster, Atlas does not support converting that same deployment back into a replica set. Reducing the shard count is a separate supported operation and does not establish a topology reversal. (low confidence) (verified as of 2026-10-30) — [source](https://www.mongodb.com/docs/atlas/scale-cluster.md)
- During public preview, Atlas Infinite does not support sharded clusters or Global Clusters/Global Writes, and does not support Atlas Search or Vector Search. (low confidence) (verified as of 2026-10-30) — [source](https://www.mongodb.com/docs/atlas/infinite/atlas-infinite-landing.md)
- Atlas Infinite public preview does not support cross-edition live migration or backup restore between Core and Infinite. Its overview separately lists mongodump, mongorestore and mongoimport as supported data-loading tools, but forbids mongorestore oplog replay; the edition lock is not evidence that every logical data transfer is impossible. (low confidence) (verified as of 2026-10-30) — [source](https://www.mongodb.com/docs/atlas/infinite/atlas-infinite-landing.md)
- MongoDB 8.0 documentation supports unshardCollection on eligible Atlas deployments: it removes a collection's shard key and places its data on one shard within the sharded cluster. That collection operation does not provide an Atlas cluster-to-replica-set conversion. (medium confidence) (verified as of 2026-10-30) — [source](https://www.mongodb.com/docs/v8.0/tutorial/unshard-collection/)
- The Atlas in-place reversal restriction must not be generalized to all MongoDB deployments. Percona describes reversal possibilities in self-managed MongoDB, while Alibaba DTS explicitly documents migration from a sharded source to a replica-set destination. Neither establishes an Atlas in-place conversion or Core/Infinite migration entitlement. (medium confidence) (verified as of 2026-10-30) — [source](https://www.mongodb.com/docs/atlas/scale-cluster.md)
