Release Vector Index

The vector indexes currently Hippo supports are in-memory indexes. Hippo supports releasing index to reduce memory consumption. Released vector index does not occupy memory anymore, however it cannot be used for vector search indeed.

curl -u shiva:shiva -XPOST 'localhost:8902/hippo/v1/{table}/_release_embedding_index?database_name={database_name}&pretty' -H 'Content-Type: application/json' -d'{
  "index_name" : "ivf_flat_index",
  "wait_for_completion" : true,
  "timeout" : "2m"
}';

Result:

{
  "job_id" : "3d585076c8994ef881eab2f16daaf096",
  "job_status" : "SHIVA_JOB_SUCCESS",
  "embedding_number" : 100,
  "task_results" : [
    {
      "id" : "0c4502bcefd4436a9efdddf78902c7a7",
      "status" : "TASK_SUCCESS",
      "server" : "172.29.40.26:27861",
      "embedding_number" : 100,
      "execute_time" : 0.014
    }
  ]
}

Parameter description:

The parameters used in index release are similar to the ones used in index activation. After release is done, the number of vectors released will be returned and the whole job will not have negative impact on write operations.