Jupyter 'The kernel appears to have died. It will restart automatically' or kernel keeps crashing/restarting: how to find the cause

Cause: The kernel is a separate process, and the notebook server only notices that it stopped responding. It dies for reasons the notebook cannot show: the operating system killing it for using too much memory, a native extension crashing (segmentation fault, often from a mismatched or broken compiled library such as numpy, torch, or a GUI backend), or a kernelspec that points at an interpreter that was removed or has the wrong packages. Large arrays or datasets are the most common trigger. Fix (best first): 1. Reproduce outside the notebook: run the same code as a script (python script.py). A crash message such as "Segmentation fault" or "Killed" tells you which case you have; "Killed" generally means the system ran out of memory (from memory of common behaviour, not from the Jupyter docs). 2. Turn on Python's fault handler to get a traceback for native crashes: put import faulthandler; faulthandler.enable() in the first cell, or start with PYTHONFAULTHANDLER=1 or python -X faulthandler. It handles SIGSEGV, SIGFPE, SIGABRT, SIGBUS and SIGILL. 3. If memory is the cause: load data in chunks, free large objects (del x), use smaller dtypes, restart the kernel to clear state, and watch usage with your system monitor. Containers and cloud notebooks often have lower limits than the machine. 4. Check the kernel is using the right environment: jupyter kernelspec list shows kernel specs, each with a kernel.json whose argv names the Python executable. Inside the environment you want, run python -m pip install ipykernel and python -m ipykernel install --user --name myenv --display-name "Python (myenv)", then pick that kernel. Remove stale specs that point to deleted interpreters. 5. Reinstall the suspect native package in that same environment (pip install --force-reinstall for numpy, torch, etc.), avoiding mixing conda and pip builds of the same library. 6. Start the server in debug mode (jupyter lab --debug) and read the terminal output for the kernel's exit message. Verify: the same cell completes without the restart banner and the kernel stays alive across repeated runs. Sources: https://jupyter-client.readthedocs.io/en/stable/kernels.html https://ipython.readthedocs.io/en/stable/install/kernel_install.html https://docs.python.org/3/library/faulthandler.html