Transparently convert pandas df's when passed as UDF args
When a pandas dataframe is a UDF argument, it is automagically converted to a TileDB array and passed as a URI. Inside the UDF it is re-rendered as a pandas dataframe.
Support arbitrary/custom packages for UDFs
The idea is to design a method to allow serverless UDFs (both generic and arrays) to support a user specified the list of dependency needed for a function.
In addition to GitHub, Google, and username/password, it would be useful to support Okta authentication with single sign-on (SSO). Okta provides two primary methods of implementing authentication: embedded and redirect. Redirect is the simplest to implement and provides implicit support for SSO. See https://developer.okta.com/docs/concepts/redirect-vs-embedded/ .
Fine-grained access policies
Currently, access policies are defined at the array level. TileDB Cloud should support up to cell-level policies.
Better support for frequent writes
Hi, I just filled out the survey on TileDB Cloud. The main reason why we're not using TileDB Cloud has to do with poor performance on S3/MinIO for small files. So, either please also support other filesystems such as SeaweedFS or add a mechanism to limit the number of fragments for use-cases that demand high-frequency writes (order of 1/s - 1/10min).