Migrate from Pinecone to Zilliz Cloud

Pinecone is a vector database that allows for similarity searches. Migrating data from Pinecone to Zilliz Cloud can enhance capabilities for managing both dense and sparse vectors while taking advantage of Zilliz Cloud’s high-performance search and analytics.

📘Notes

This migration only supports Pinecone serverless indexes.

The migration process is structured into these steps:

1. Connect to data source: Enter your Pinecone API key to establish a connection.

2. Select source and target:

   • Choose one or more Pinecone indexes for migration.

   • Select an existing Zilliz Cloud cluster as the target. Each selected Pinecone index will become a new collection in Zilliz Cloud.

3. Configure schema: Verify that field types are correctly mapped between Pinecone and Zilliz Cloud. For detailed mapping rules, refer to Mapping rules.

Mapping rules

The following table summarizes how field types in Pinecone are mapped to Zilliz Cloud field types, along with details on any customization options.

Pinecone Field Type	Zilliz Cloud Field Type	Description
Primary key	Primary key	Pinecone's ID field is automatically mapped as the primary key in Zilliz Cloud. When migrating data, you can enable Auto ID. However, if you do so, the original primary key values from your source index will be discarded.
Dense vector	FLOAT_VECTOR	Dense vector fields are transferred as FLOAT_VECTOR with no modifications required.
Sparse vector	SPARSE_FLOAT_VECTOR	If the sparse vector field in a sample row is non-empty, it is mapped by default; otherwise, it remains unselected in schema mapping.
Metadata	Dynamic field	By default, Pinecone's metadata is mapped as a dynamic schema in Zilliz Cloud. For more information, refer to Dynamic Field. When migrating data, consider converting dynamic fields into fixed fields when their patterns have stabilized and you want to enforce strict data types and optimized index configurations for these fields.
String	VARCHAR	If a metadata field is of type string and you convert it to a fixed field, it becomes a VARCHAR type. Note: The maximum length for this field is fixed at 65,535 bytes and cannot be modified. The capacity calculation is determined by the actual field length.
Number (integer or floating point)	DOUBLE	If a metadata field is of type number and you convert it to a fixed field, it becomes a DOUBLE type.
Boolean	BOOL	If a metadata field is of type boolean and you convert it to a fixed field, it becomes a BOOL type.
List of strings	ARRAY<VARCHAR>	If a metadata field is a list of strings and you convert it to a fixed field, it becomes an array of VARCHAR.

Before you start

• The source Pinecone index is accessible from the public internet.

• If you have an allowlist configured in your network environment, ensure that Zilliz Cloud IP addresses are added to it. For more information, refer to Zilliz Cloud IPs.

• You have obtained the API key to access the target Pinecone project.

• You have been granted the Organization Owner or Project Admin role on Zilliz Cloud. If you do not have the necessary permissions, contact your Zilliz Cloud administrator.

Migrate from Pinecone to Zilliz Cloud

1. Log in to the Zilliz Cloud console.

2. Go to the target project page and select Migrations > Pinecone.

   

3. In the Connect to Data Source step, enter the API key that can be used to access the target Pinecone project. Then, click Next.

   📘Notes

   Authentication can guide you in obtaining the required connection information.

   

4. In the Select Source and Target step, configure settings for the source Pinecone index and target database in your Zilliz Cloud cluster. Then, click Next.

   

5. In the Configure Schema step, set up field mappings between Zilliz Cloud and Pinecone:

   

   1. Confirm field mappings:

      • Zilliz Cloud automatically detects and displays your Pinecone fields alongside their corresponding target fields. For details on how these fields are mapped, refer to Mapping rules.

      • Verify that each Pinecone field is correctly paired with its corresponding target field. You can rename fields as needed, but note that the data type cannot be changed.

   2. Handle metadata fields:

      • Your Pinecone metadata fields appear in the Metadata section and are set as dynamic fields by default.

        📘Notes

        Dynamic fields store metadata in a JSON format, enabling more flexible and evolving data structures. For details, refer to Dynamic Field.

        Fixed fields are explicitly defined in your schema with a predetermined structure. They allow you to enforce specific data types and index configurations.

      • To convert a metadata field into a fixed field, select the field and click the Convert to Fixed Field icon. Note that Zilliz Cloud samples only 100 rows to extract fields from metadata. To add more fields, click the Settings icon.

      • For metadata fields converted to fixed fields, configure the following attributes:

        • Nullable: Decide whether a field can accept null values. This feature is enabled by default. For details, refer to Nullable & Default.

        • Default Value: Specify a default value for a field. For details, refer to Nullable & Default.

   3. Set namespace mapping:

      In the Namespace Mapping section, the Pinecone namespace is configured as the partition key by default. We recommend you retaining this setting for enhanced performance.

      📘Notes

      A partition is a subset of a collection containing part of the data, while a partition key is a scalar field that automatically distributes entities into partitions based on their hash values to optimize search performance. For more information, refer to Use Partition Key and Manage Partitions.

   4. (Optional) Adjust shards:

      • Click Advanced Settings to configure the number of shards for your target collection.

      • For datasets of around 100 million rows, a single shard is typically sufficient.

      • If your dataset exceeds 1 billion rows, contact us to discuss optimal shard configuration for your use case.

6. Click Migrate.

Monitor the migration process

Once you click Migrate, a migration job will be generated. You can check the migration progress on the Jobs page. When the job status switches from In Progress to Successful, the migration is complete.

Post-migration

After the migration job is completed, note the following:

• Index Creation: The migration process does not automatically create indexes for vector fields when migrating from external data sources. You must manually create the index for each vector field. For details, refer to Index Vector Fields.

• Manual Loading Required: After creating the necessary indexes, manually load the collections to make them available for search and query operations. For details, refer to Load & Release.

📘Notes

Once you have completed indexing and loading, verify that the number of collections and entities in the target cluster matches the data source. If discrepancies are found, delete the collections with missing entities and re-migrate them.

Cancel migration job

If the migration process encounters any issues, you can take the following steps to troubleshoot and resume the migration:

1. On the Jobs page, identify the failed migration job and cancel it.

2. Click View Details in the Actions column to access the error log.