worker started without compute kernels; compute falls back to the main thread (columnization still offloads)

Warnings reference › worker/ready-no-compute

Identifier

worker/ready-no-compute

Raised by

The grid (background worker)

Since

1.19.0

What happened

The worker booted and answered, but reported that it has no compute kernels loaded. The boundary is alive, so the work that needs no kernels - columnising incoming rows - still crosses it; sorting, filtering, grouping and aggregation are marked unusable and run on the main thread from here on. The two are deliberately decoupled, so this is a partial degradation, not a dead worker.

The fix

This is a build shape, not a runtime setting: the worker entry you serve must be the one that carries the kernels. If you named your own worker.workerUrl, re-copy it from the published package.

Reference

Covered in full at the API reference.