Vector Search Index Sync fails in Initializing

[tiho]

Vector Search Index Sync fails in Initializing. This index table was already up and running, and when I tried to sync it, it failed in Initializing. See the attached. 

[Kaniz_Fatma]

Hi @tiho, I understand that you’re encountering issues with the Vector Search Index in Azure Databricks.

Let’s troubleshoot this together.

1. Temporary Issue: It appears that you’re not alone in experiencing this problem. Other users have reported similar issues, and in some cases, it resolved itself after a period of time. For instance, one user mentioned that the issue was resolved within 48 hours without any changes mad...¹. So, it might be a temporary glitch with Databricks.

2. Region and Provisioning: You mentioned that you’re using Azure Databricks in the West Europe region. Sometimes, the provisioning process can take longer due to various factors. However, if this persists, consider the following steps:

   • Check Logs: Inspect the logs or error messages to identify any specific issues during the provisioning process.
   • Retry: Retry the provisioning process after some time. If it still fails, proceed to the next step.
   • Resource Constraints: Ensure that there are no resource constraints (such as memory or CPU) affecting the provisioning process.
   • Network Connectivity: Verify that your Databricks cluster has proper network connectivity to the required services.

3. Delta Sync Index: When creating a Vector Search Index, there are two types: Delta Sync Index and Non-Delta Sync Index. The Delta Sync Index automatically syncs with a source Delta Table, incrementally updating the index...². Make sure you’re using the appropriate type for your use case.

4. Python SDK and REST API: You mentioned using both the UI and the Python VectorSearchClient. If the UI is not working as expected, try creating the index programmatically using the Python SDK or the REST API. This might provide additional insights into any issues.

[amitpphatak]

Were you able to resolve this issue? I am facing the same issue right now - I am on AWS.

[jnkthms]

We are still running in the same issue. On Azure in Westeurope Region. Any updates?

[casgiedus]

I'm having a similar issue - it's stuck at "Waiting for initial sync...".

Nothing is happening, I'm not getting any requests on the embedding endpoint.

[jnkthms]

The issue for us was most likely that we used CPU compute for the deployed embedding model, switching to GPU (small) solved the issue.