tags: linux, container, cgroup
I kernel document about this is very clear.
We can use cgroup freezer to freeze a group of processes.
It has 3 state: THAWED, FREEZING, FROZEN. You can read state from freezer.state.
And you can write state to that file to change state. But only THAWED and FROZEN
can be written. FREEZING is a internal state that processes will go to the FREEZING
at first and then FROZEN.
If a cgroup state is frozen, then all processes of it and its descendant cgroups are
frozen. So a descendant cgroup may be frozen even if state of itself is THAWED.
Then kernel provides freezer.self_freezing and freezer.parent_freezing to indicate
that whether a cgroup itself is frozen or its parent is frozen.
For cgroup v1, the freezer may put the process into unkillable state when the process is in uninterruptible state. Then you cannot kill it. The cgroup v2 changes this to make it more like the SIGSTOP, it will freeze when the process is in killable state.