CoreML crashes in MacOS 26

[SuzukiHonoka]

I have searched the existing issues, both open and closed, to make sure this is not a duplicate report.

• Yes

The bug

[12/10/25 14:54:15] INFO Booting worker with pid: 87476
[12/10/25 14:54:16] INFO Started server process [87476]
[12/10/25 14:54:16] INFO Waiting for application startup.
[12/10/25 14:54:16] INFO Created in-memory cache with unloading after 300s of inactivity.
[12/10/25 14:54:16] INFO Initialized request thread pool with 10 threads.
[12/10/25 14:54:16] INFO Application startup complete.
[12/10/25 14:54:24] INFO Loading visual model 'ViT-B-16-SigLIP2__webli' to memory
[12/10/25 14:54:24] INFO Setting execution providers to ['CoreMLExecutionProvider', 'CPUExecutionProvider'], in descending order of
preference
2025-12-10 14:54:24.584 python[87476:601288] 2025-12-10 14:54:24.584213 [W:onnxruntime:, coreml_execution_provider.cc:113 GetCapability] CoreMLExecutionProvider::GetCapability, number of partitions supported by CoreML: 3 number of nodes in the graph: 473 number of nodes supported by CoreML: 470
objc[87476]: +[NSPlaceholderString initialize] may have been in progress in another thread when fork() was called.
objc[87476]: +[NSPlaceholderString initialize] may have been in progress in another thread when fork() was called. We cannot safely call it or ignore it in the fork() child process. Crashing instead. Set a breakpoint on objc_initializeAfterForkError to debug.
[12/10/25 14:54:25] ERROR Worker (pid:87476) was sent SIGABRT!

The OS that Immich Server is running on

MacOS26

Version of Immich Server

latest

Version of Immich Mobile App

latest

Platform with the issue

• Server
• Web
• Mobile

Device make and model

No response

Your docker-compose.yml content

N/A

Your .env content

N/A

Reproduction steps

System Preparation (once)

xcode-select --install # compilers & headers
brew install git pyenv pipx uv # uv = fast dep solver
pyenv install 3.12.2 && pyenv global 3.12.2
Grab Immich’s ML Code

git clone --depth 1 https://github.com/immich-app/immich.git
cd immich/machine-learning
Force Core ML to be Tried First Edit immich_ml/models/constants.py and move "CoreMLExecutionProvider" to the top of SUPPORTED_PROVIDERS.

Clean Virtual Environment with Only Runtime Dependencies

uv venv .venv --python $(pyenv which python)
UV_HTTP_TIMEOUT=120 uv sync --extra cpu --no-dev
source .venv/bin/activate

Relevant log output

Additional information

No response

[iamgerwin]

I've created a fix for this issue in PR #24517.

Root Cause:
On macOS, CoreML and other Objective-C frameworks are not fork-safe. When gunicorn forks workers, the Objective-C runtime crashes because it detects that class initialization (like NSPlaceholderString) was in progress when fork() was called.

Solution:
The fix sets OBJC_DISABLE_INITIALIZE_FORK_SAFETY=YES environment variable on macOS platforms before gunicorn forks workers. This is a known workaround for macOS fork-safety issues with Objective-C frameworks.

The change only affects macOS (sys.platform == 'darwin') and uses setdefault() so users can override if needed.