How does the Python import system work?

Question presented to candidate: "Walk me through what actually happens when Python runs import foo.bar. Where does it look, what objects are involved, why does a module's code run only once, and why do circular imports sometimes fail with 'partially initialized module'?"

What a strong answer should cover:

• import is a run-time statement (IMPORT_NAME in bytecode), not a textual include; the first stop is the sys.modules cache, which is why a module body executes only once per process
• On a cache miss, each finder on sys.meta_path is asked find_spec(name, path); the defaults are BuiltinImporter, FrozenImporter and PathFinder, which walks sys.path (or the parent package __path__) using path-entry finders from sys.path_hooks, cached in sys.path_importer_cache
• The finder returns a ModuleSpec (name, loader, origin, submodule_search_locations); the loader creates the module, it is put in sys.modules before the body runs, then exec_module runs the code; on error the entry is removed
• Dotted imports import each parent first (foo, then foo.bar using foo.__path__); import foo.bar binds foo, from foo import bar binds bar; relative imports are resolved against __package__
• Circular imports work only when the half-initialised module already has the names being asked for; fixes are importing the module instead of names, moving the import, or restructuring
• Source files are compiled to __pycache__/*.cpython-3XX.pyc, revalidated by source mtime and size by default (hash-based pycs since 3.7, PEP 552); importlib.reload re-executes into the same object

Clarifying questions expected:

• Are we talking about CPython's default machinery, or a custom setup (zip apps, frozen apps, a custom finder)?
• Is the interest in how to debug a specific import error (wrong module picked up, circular import)?

Code / implementation expected: Optional — a custom MetaPathFinder / Loader, or a circular-import repro and its fix.