Skip to content

fix: reload changed models and validate runtime health reports - #6

Merged
Ayushdevo merged 1 commit into
mainfrom
fix/runtime-artifacts-20261109
Nov 9, 2026
Merged

Ayushdevo merged 1 commit into
mainfrom
fix/runtime-artifacts-20261109

Conversation

@Ayushdevo

Copy link
Copy Markdown
Owner

Predictions now reload a replaced model artifact without restarting the worker, while unchanged artifacts are deserialized only once. Deleted or corrupt replacements return 503 instead of serving the cached model. The cache includes path, inode, size and nanosecond timestamps; deploy models with atomic replacement.

Readiness and /health now share validation of the required top-level report fields, score bounds and status. Empty/incomplete reports and malformed types return 503; existing optional report fields are preserved.

Fixes #2
Fixes #3

Validation: 236 tests passed on Python 3.12; scoped Ruff and git diff --check passed. Four reload regressions and 32 health-schema cases failed before their corresponding changes. The health check validates the top-level contract, not every nested report value. AI-assisted implementation and validation.

Fixes #2 and #3. AI-assisted implementation and regression validation.
@Ayushdevo
Ayushdevo merged commit 3118050 into main Nov 9, 2026
7 checks passed
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.

Cached model is not refreshed after the model artifact is replaced Readiness endpoint can report ready for corrupt runtime artifacts

1 participant