Client-server architecture separates a client that requests services from server processes that perform the work. Docker uses a Docker client, Docker daemon, containerd, and an OCI runtime; the container becomes a child of containerd rather than the client. section 1.3.3
Figure 1.8 Docker client-server architecture. The container is a direct descendant of containerd, not the Docker client. The kernel sees no relationship between the client program and the container.
The architecture makes the process boundary visible: the kernel sees the container as a descendant of containerd, not of the Docker client. section 1.3.3
This indirection adds daemon failure points and complicates diagnosis. It also weakens user attribution: containers launched through a daemon inherit the daemon's loginuid, whereas fork/exec execution preserves the launching user's identity for kernel auditing. section 1.3.3 section 11.1.2
Inspecting process attribution
A Podman container inspected with `podman inspect -l --format '{{ .State.Pid }}'` retains the user's loginuid, while an equivalent Docker container reports `4294967295`, the kernel's representation of an unset loginuid. section 11.1.2







