Skip to content

Conversation

@dantasse
Copy link
Contributor

Adds docs from geneva internal docs about error handling. These looked pretty good already; I only tweaked a couple things to make them sound a little bit less AI, and adjusted the hierarchy to make it clear that there are 3 levels of complexity you can use to customize your error handling.

Touched up the main UDF page too.

`checkpoint_size` controls how many rows are processed before checkpointing, and therefore reporting and saving progress.

In LanceDB, the default number of rows per fragment is 1024 * 1024 rows. Conside a captioning UDF that takes 30s per row. It could take a year (!) before any results show up! (e.g. 30s/row * 1024*1024 rows/fragment => 30M s/fragment => 8.3k hours/fragment -> 347 days/fragment). To enable these to be parallelized, we provide a `batch_size` setting so the work can be split between workers and so that that partial results are checkpointed more frequently to enable finer-grained progress and job recovery.
By default, Geneva uses adaptive checkpoint sizing, so you don't need to set this parameter. This is because UDFs can be quite varied -- some can be simple operations where thousands of calls can be completed per second, while others may be slow and require 30s per row. So a simple default like "write every 1000 rows" might write once a second or once every 8 hours! Geneva will avoid this by adapting its checkpointing rate to your task. But if you want to see writes more or less frequently, you can set `checkpoint_size` manually.
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I've been having some issues with this feature, we might want to mark the adaptive checkpoitn sizing as experiemental for now.

@dantasse dantasse merged commit 5043ae2 into main Jan 20, 2026
2 checks passed
@dantasse dantasse deleted the dantasse/error_handling_docs branch January 20, 2026 19:50
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants