Docs Menu

Global Cluster Sharding Reference

On this page

  • Sharding Collections for Global Writes
  • Error Handling
  • Hashed Sharding Options
  • Global Cluster Write Operations
  • Global Cluster Read Operations
  • Sharding Collections without Global Writes
  • Unsharded Collections in Global Write Clusters

The following sections describe sharding behavior and how to enable sharding for Global Clusters.

Unsharded collections must meet the following compatibility requirements prior to sharding to utilize Global Writes when sharded:

  • Every document in the collection must include a location field.
  • The value of the location field must be either an ISO-3166-1 alpha 2 country code ("US", "DE", "IN") or a supported ISO-3166-2 subdivision code ("US-DC", "DE-BE", "IN-DL"). Documents that do not match this criteria can't be routed to any shard in the cluster. To view the complete list of currently supported country or subdivision codes, visit

For collections that meet the stated requirements, you must shard the collection using the following pattern:

{ "location" : 1, "<secondary_field>" : 1 }

A shard key on the location field alone may result in bottlenecks, especially for workloads where a subset of countries or subdivisions receive the majority of write operations. Atlas Global Writes requires a compound shard key to facilitate efficient distribution of sharded data across the cluster. Atlas Global Cluster shard keys share the same restrictions as MongoDB shard keys. For example, the secondary shard key field cannot be an array.

See also:

To learn more about:

Changing the Shard Key

After sharding, what you can modify depends upon the version of MongoDB that you run:

MongoDB Version
Modify Shard Key Keys
Modify Shard Key Values
MongoDB 4.0
MongoDB 4.2
MongoDB 4.4
Yes, add fields to Key using Data Explorer only
MongoDB 5.0
Yes, add fields to Key using Data Explorer only

The Atlas Perform CRUD Operations in Atlas supports creating sharded collections with specific validations for Global Writes.

You can also use mongosh to execute the sh.shardCollection(). After sharding the collection, you must use the Atlas Perform CRUD Operations in Atlas to enable Global Writes for that collection.

See also:

To learn more about sharding collections via the Data Explorer, see Shard a Global Collection for Global Writes in Data Explorer.

If Atlas encounters an error while sharding a collection for global writes, a message appears in the banner at the top of the screen.

Screenshot of an Atlas sharding error alert
  1. Click See Details to learn about the error and the namespace where the error occured. A modal window appears with the complete error message and a Fix Now button:

    Screenshot of an Atlas sharding error modal window
  2. Click the Close and navigate to the collection in Data Explorer. You can also click the Fix Now button to go to Data Explorer for that Atlas cluster.
  3. Click the Global Writes tab for the collection mentioned in the error message. You will see an error similar to the following:

    Screenshot of an Atlas sharding error in Data Explorer
  4. Click Unmanage Collection to cancel the global writes sharding operation. You must have the Project Data Access Admin role to cancel the sharding operation.

After making any necessary changes to the collection as indicated by the error message, you can start the sharding process again.

Possible errors include:

An index already exists on the custom shard key.
If the field chosen as the second part of the compound shard key is already indexed, the sharding operation may fail.
The shard key field is not present.
All documents in the collection must contain both the shard key fields. This error occurs only in versions earlier than MongoDB 4.4.
The collection is already sharded.
If the collection has already been manually sharded, the operation fails.
The collection has a custom default collation.
A custom default collation on the collection may cause a sharding error.

Shard keys use hashed sharding and pre-split data for even distribution. This is only available on Atlas clusters running MongoDB 4.4 and above.

If you enable the use of hashed index shard keys by selecting Use hashed index as the shard key in the User Interface or by setting isCustomShardKeyHashed through the API, Atlas distributes the sharded data evenly by hashing the second field of the shard key.

You can optionally select Pre-split data for even distribution in the Atlas User Interface or set presplitHashedZone using the API to specify whether to perform initial chunk creation and distribution for an empty or non-existing collection based on the defined zones and zone ranges for the collection.

When creating a sharded collection using a compound hashed shard key for Global Clusters, Atlas creates at least 1 chunk per location code and attempts to distribute chunks evenly across shards in the cluster. By default, Atlas creates one chunk per location code.

You can also specify the minimum number of chunks to create initially when sharding an empty collection with a hashed shard key using the Atlas Data Explorer User Interface or by setting numInitialChunks parameter through the API.


If you specify the number of chunks per shard, Atlas creates at least the minimum number of chunks that you specified, with the same number of chunks per location code. If you specify the minimum number of chunks, Atlas sets up zoned sharding quickly, especially if you already know how to geographically distribute your data before sharding.

Each write to a sharded collection must include the shard key for the operation to succeed. For each document in a write operation, MongoDB uses the location field of the shard key to determine the zone to which to route the data. MongoDB selects a shard associated to that zone as the target for writing the document, facilitating geographically isolated and segmented data storage.

MongoDB can only guarantee this behavior for inserted documents that meet the criteria defined in Sharding Collections for Global Writes. Specifically, MongoDB can route a document whose location field doesn't conform to ISO-3166-1 alpha 2 or ISO-3166-2 to any shard in the cluster.

MongoDB query routing depends on whether the read operation includes the full shard key and that the location value corresponds to a supported ISO-3166-1 alpha 2 country code ("US", "DE", "IN") or a supported ISO-3166-2 subdivision code ("US-DC", "DE-BE", "IN-DL").

  • For queries that do include the full shard key and whose location value meets the requirements for Global Writes, MongoDB targets the read operation to the zone which maps to the location value or values specified in the query.
  • For read operations that don't include the location value , or if the location value doesn't correspond to a supported ISO-3166-1 alpha 2 country code or ISO-3166-2 subdivision code, MongoDB must broadcast the read operation to every zone in the cluster.
  • For Global Writes zones which have Read-only nodes in geographically distant regions, clients in those regions can query the local Read-only node for that zone by specifying the full shard key as part of the query and issuing the read operation with a Read Preference of nearest.

Secondary reads may return stale data depending on the level of replication lag between the secondary node and the primary.

See also:

To learn more about:

Global Writes clusters support the same Ranged and Hashed sharding strategies as a standard Atlas sharded cluster. For sharded collections whose shard keys and document schema do not support Global Writes, MongoDB distributes the sharded data evenly across the available shards in the cluster with respect to the chosen shard key. Consider using a separate sharded cluster for data that can't take advantage of Global Writes.

You can't modify a collection to support Global Writes after sharding. We recommend that you choose a shard key that will allow you to use Global Writes for a collection the future.

See also:

To learn more about Global Writes sharding requirements, see Sharding Collections for Global Writes.

Global Clusters provide the same support for unsharded collections as a standard Atlas sharded cluster. For each database in the cluster, MongoDB stores its unsharded collections on a primary shard. Use sh.status() from the mongosh to determine the primary shard for the database.

Give Feedback
MongoDB logo
© 2021 MongoDB, Inc.


  • Careers
  • Legal Notices
  • Privacy Notices
  • Security Information
  • Trust Center
© 2021 MongoDB, Inc.