Validate select fields before model loading to prevent cryptic crashes

[arjunrajlab]

Summary

Adds validation for select interface fields (particularly model selections) before they are used in model loading or checkpoint path construction. This prevents cryptic crashes deep inside model loaders when a saved tool config holds null or a stale model name.

Problem

A production sam_fewshot_segmentation job crashed with FileNotFoundError: [Errno 2] No such file or directory: '/None.pth' because the saved tool config held "Model": null. The null value was silently used to build the checkpoint path, and the job died deep inside SAM's model loader with no indication that the model selection was the problem.

Additionally, configs saved against older worker images can name model checkpoints that no longer exist in the current image, leading to similar downstream failures.

Solution

1. New utility function annotation_tools.get_required_select() validates that a select field value is:

   • Not None or empty/whitespace
   • A string (not a list, dict, or number)
   • In the allowed values list (if provided)

   Raises ValueError with a clear message directing users to re-select the field.

2. Early validation in compute() functions across all affected workers:

   • Validates the model selection before any heavy imports or GPU setup
   • Checks that the checkpoint file exists on disk
   • Calls sendError() with a clear message on validation failure
   • Returns early to prevent downstream crashes

3. Affected workers:

   • sam_fewshot_segmentation: validates against MODELS list
   • sam2_fewshot_segmentation, sam2_automatic_mask_generator, sam2_propagate, sam2_video, sam2_refine: validate against MODEL_TO_CFG mapping
   • stardist: validates against MODELS list
   • cellpose, cellposesam, piscis (predict/train), cellpose_train, cellposesam_train: validate model selections (some without static allowed values for custom models)

Key Changes

• annotation_utilities/annotation_tools.py: Added get_required_select() function with comprehensive validation and user-friendly error messages
• annotation_utilities/tests/test_required_select.py: New test file with 14 regression tests covering valid values, missing values, wrong types, and stale values
• Worker entrypoints: Each affected worker now:
  • Extracts model selection early via get_required_select()
  • Validates against a static MODELS or MODEL_TO_CFG constant
  • Checks checkpoint file existence before loading
  • Calls sendError() on any validation failure
• CLAUDE.md and .claude/skills/nimbus-worker-hardening/SKILL.md: Updated with guidance on the pitfall and the fix pattern
• Test coverage: Added TestComputeModelValidation classes to sam_fewshot_segmentation and sam2_fewshot_segmentation test suites

Implementation Details

• Validation happens before lazy imports of torch/model libraries, so failures are fast
• Error messages include the field name and available options (when applicable)
• The fix rejects null/stale values rather than silently substituting defaults, since the saved value is what the user believes the tool will run with
• Workers with custom models from Girder (cellpose family, piscis) validate shape only—no static allowed_values list exists for those

https://claude.ai/code/session_01FFwLaqzmDWDnPuZvF6EL6m

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 325145ae90

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".