This page shows the answers to the common questions.
Please refer to the troubleshooting page when you encounter operation problems.
What is the ingress filter?
The ingress filter is the user-defined component that connects to the
For example, you can convert object data (e.g., image binary, text, etc.) to the vector using your ML models as pre-process of each request.
What is the egress filter?
The egress filter is the user-defined component that connects to the vald-filter-gateway component. You can use it for filtering search results, e.g., when you’d like to get a white T-shirt list, use it to remove other colors of T-shirts from search results that the vald-lb-gateway component returns.
Is Vald Index Manager recommended using?
We recommend using it when you’d like to operate as a cluster. It helps you to manage indexing timing for each Vald Agent.
What are the pluggable options?
Vald has three pluggable options:
- Backup with the external storage for Vald Agent
- You can connect the external storage, like S3, GCS, etc., to the Vald Agent Sidecar component for backup.
- Algorithm of the core engine for Vald Agent
- We’re going to add another algorithm in the near future.
- Filtering with filter gateway
- You can filter the search results by your own defined filter component by connecting to the filter gateway before returning the search result.
- You can convert object data to vector by own defined ingress filter component by connection to filter gateway before inserting.
How to deploy the Vald cluster?
We recommend using Helm for deploying the Vald cluster. You can deploy by following the steps.
- Install Helm(v3~) and prepare the Kubernetes cluster
Helm chartas your demand
- Deploy by Helm command
Is there any support for bulk inserts?
StreamInsert for bulk insert.
Please refer to the insert API documentation for more detail.
Vald also provides
StreamXXX as bulk operations for each service.
For more detail, please refer to the API document overview.
Can Vald handle multi-embedding vectors?
Unfortunately, the current Vald cannot directly handle multi-embedding spaces with a single Vald cluster. You have to choose one of two options to use the multi-embedding vectors in Vald.
- Deploy multiple Vald cluster
- Covert vector to new vector in the specific space by some methods
How to backup index data?
There are three ways to backup index data:
- Using external storage (S3, GCS)
- Using Persistent Volume
- Using the external storage and Persistent Volume
Please refer to the sample configurations.