Schema Deletion Guidelines¶
The Schema Registry API supports deleting a specific schema version or all versions of a subject. On a soft delete, the API only deletes the version and the underlying schema ID would still be available for any lookup.
Soft delete a schema¶
# Deletes all schema versions registered under the subject "Kafka-value" curl -X DELETE http://localhost:8081/subjects/Kafka-value  # Deletes version 1 of the schema registered under subject "Kafka-value" curl -X DELETE http://localhost:8081/subjects/Kafka-value/versions/1 1 # Deletes the most recently registered schema under subject "Kafka-value" curl -X DELETE http://localhost:8081/subjects/Kafka-value/versions/latest 1
There may be a variance in what is reported on the Confluent Cloud web UI and CLI for number of deleted schemas. See also, Hard delete a schema and Looking under the hood at schema deletion, versioning, and compatibility.
Hard delete a schema¶
Both Confluent Platform and Confluent Cloud now support hard delete of a schema with the use of the query string,
?permanent=true. A hard delete removes all metadata, including schema IDs. This can be useful for freeing up space for new schemas on Confluent Cloud, which sets limits on the number of schema versions supported in the registry.
You can perform a hard delete on all schema versions registered under a subject or on a specific version of a subject.
To accomplish a hard delete of a schema (all versions or a specific version), use this two-step process. You must first soft delete the schema, then hard-delete it.
Perform a soft delete of all versions of the schema.
curl -X DELETE -u <schema-registry-api-key>:<schema-registry-api-secret> <schema-registry-url>/subjects/my-existing-subject
Perform a hard delete of all versions of the schema by appending
?permanent=trueto the command.
curl -X DELETE -u <schema-registry-api-key>:<schema-registry-api-secret> <schema-registry-url>/subjects/<my-existing-subject>/versions/<version-id>?permanent=true
The following commands hard-delete all versions of the schema registered under the subject “Kafka-value”:
curl -X DELETE <schema-registry-api-key>:<schema-registry-api-secret> <schema-registry-url>/subjects/Kafka-value curl -X DELETE <schema-registry-api-key>:<schema-registry-api-secret> <schema-registry-url>/subjects/Kafka-value?permanent=true
To hard-delete version 1 of the schema registered under the subject Kafka-value:
curl -X DELETE <schema-registry-api-key>:<schema-registry-api-secret> <schema-registry-url>/subjects/Kafka-value/versions/1 curl -X DELETE <schema-registry-api-key>:<schema-registry-api-secret> <schema-registry-url>/subjects/Kafka-value/versions/1?permanent=true
For the version-specific delete, a hard delete requires a version number as input; not just
version:latest, which results in a soft delete only.
Guidelines for using these APIs¶
The above APIs are primarily intended to be used be in a development environment where it’s common to go through iterations before finalizing a schema. While it’s not recommended to be used in a production environment, there are few scenarios where these APIs can be used in production, but with utmost care.
- A new schema to be registered has compatibility issues with one of the existing schema versions
- An old version of the schema needs to be registered again for the same subject
- The schema’s are used only in real-time streaming systems and the older version(s) are absolutely no longer required
- A topic needs to be recycled
- You want to free up space for new schemas on Confluent Cloud (with a hard delete of schemas).
It is also important to note that any registered compatibility settings for the subject would also be deleted while using Delete Subject or when you delete the only available schema version.
Viewing available schemas on Confluent Cloud¶
You can view available and used schemas, and infer the number of soft deleted schemas, from the Confluent Cloud web UI or from the CLI.
On the Confluent Cloud CLI, you can use
ccloud schema-registry cluster describe to view
the number of available schemas for the currently selected environment. For example:
$ ccloud schema-registry cluster describe +----------------------+----------------------------------------------------+ | Name | account schema-registry | | Cluster ID | lsrc-g2p81 | | Endpoint URL | https://psrc-0xx5p.us-central1.gcp.confluent.cloud | | Used Schemas | 7 | | Available Schemas | 993 | | Global Compatibility | <Requires API Key> | | Mode | <Requires API Key> | | Service Provider | | +----------------------+----------------------------------------------------+
“Used Schemas” reports the total of active and soft deleted schemas. If you know the number of active schemas you have, you can infer the number of soft deleted schemas. If a schema is permanently deleted, the number of used schemas will decrease and available schemas will increase accordingly.
Looking under the hood at schema deletion, versioning, and compatibility¶
In reality, deleting a schema removes only the versioned instance(s) of the schema. The actual schema (with the hashed ID) does not go away. The canonical MD5 hash of the schema still exists in the system. This is why you can perform a lookup on a deleted schema in the rest API.
If, on a subject that has three versions, you delete the third one but register again the same schema it will be registered under version 3 and the associated schema will have the same ID as before, because the hash matched the existing one.
When the Avro serializer serializes Avro data, it stores the ID (not the version, which is specific to the subject). If the version (in a subject) is deleted, the schema and its ID remain in Schema Registry so that the data and can still be deserialized using Schema Registry
This has implications for versioning and compatibility. Suppose, for a given subject, you have two schemas:
- Schema ID: 23, version 1 with fields (a,b)
- Schema ID: 45, version 2 with fields (a,b,c)
If you delete version 2, version 1 becomes the latest version.
If the compatibility level is BACKWARD, the next schema that is registered to the subject need only be backward compatible with version 1, not version 2.