#cgroups-v2 #cli #grpc #tokio

nightly app runkins

Runkins makes it easy to manage command execution

1 unstable release

Uses new Rust 2021

0.1.0 Nov 15, 2021

#101 in #grpc

MIT license

930 lines


Process executor with cgroup support and gRPC interface. Executes processes, stores stdout, stderr in memory and streams them to the CLI.

Possible use cases

  • Usage in scripts to manage long running processes that might outlive the script
  • Debugging where a full blown container runtime would get in the way but a better interface than plain linux is desired
  • Light-weight CI: steps in a workflow can be triggered by exit codes, stdout content, timeouts etc.

Folder structure

  • runkins - binary package containing the runkins CLI and the runkins-server gRPC server
  • lib - library used by the server, examples/slow binary for testing
  • proto - library and *.proto definition
  • systemd - sample .service file



  • Rust 2021 edition

For cgroup functionality (required when running a process with limits set):

  • cgroup v2 enabled
  • systemd --version >= 244 for cgroup v2 controller delegation to non-root users. This can be worked around e.g. by running as root.

More information can be found in the cgroup section.


cargo install runkins

Running the server

To see the error logs in the terminal, use

export RUST_LOG=runkins_server=debug,runkins_lib=debug,info

To run the server, execute:


This will start the gRPC server on (currently hardcoded) localhost:50051. The CLI has the same hardcoded address.

Alternatively, use the systemd service example.

Running the CLI

To start a process, run

runkins start -- ls -la

The start subcommand outputs the RUNKINS_EID to stdout.

All other subcommands (status, stop, logs, rm) expect to receive it either as the first parameter or as RUNKINS_EID environment variable.

Example workflow:

RUNKINS_EID=$(runkins start -- ls -la)
runkins status
runkins logs
runkins stop
runkins rm
rm .env

To get help with commands, use --help flag.


cargo build

To create a debug build of all components run

cargo build

Alternatively run with --release to get the release build.


Manual testing

Start the server, then execute the binary slow created just for testing:

RUNKINS_EID=$(runkins start -- cargo run --example slow 10)

Test that the system works as expected. The logs subcommand should write lines to stdout and stderr. After the program writes 10 lines, the clients should disconnect and the status should be Exited with code 0.

To test different exit codes, use slow 1 3, which will end with status Exited with code 3.

Killing the slow process using killall slow or stop RPC should end with status Exited with signal.

Automated testing


cargo test

To execute tests that are dependent on cgroup v2 and systemd-run, enable test_systemd_run configuration either via RUST_FLAGS or .cargo/config.

cgroup support

Following environment variables need to be set, otherwise cgroup support will not be enabled.

  • PARENT_CGROUP - full path to a cgroup directory that can be written by the current user
  • CGROUP_BLOCK_DEVICE_ID - in form of MAJOR:MINOR, see lsblk
  • CGROUP_MOVE_CURRENT_PID_TO_SUBFOLDER_ENABLED - required for systemd service support. If set (to anything), creates PARENT_CGROUP/service cgroup if needed and moves current pid there. This setting also adds the required controllres to $PARENT_CGROUP/cgroup.subtree_control .

Starting new user slice using systemd-run

If the parent cgroup is created using systemd-run, make sure the shell is still open. Example:

$ systemd-run --user -p Delegate=yes --slice=my.slice --shell
# depending on the user, path might be:
# export PARENT_CGROUP=/sys/fs/cgroup/user.slice/user-1000.slice/user@1000.service/my.slice


TODO: this should be a healthcheck running on the server startup.

Verify that the process runs in its own cgroup:

# note the --limits flag - if not set, cgroup will not be created
$ RUNKINS_EID=$(runkins start --limits -- cat /proc/self/cgroup)
$ runkins logs $RUNKINS_EID
$ runkins rm $RUNKINS_EID

Verify that all required controllers cpu io memory are available:

$ RUNKINS_EID=$(runkins start --limits -- \
 sh -c 'cat /sys/fs/cgroup/$(cat /proc/self/cgroup | cut -d ':' -f 3)/cgroup.controllers')
$ runkins logs $RUNKINS_EID
cpu io memory pids
$ runkins rm $RUNKINS_EID

Setting cgroup limits via CLI

Switches for limits can be discovered using help.

$ runkins start --help
runkins-start 0.1.0

    runkins start [FLAGS] [OPTIONS] <path> [args]...

    -h, --help       Prints help information
    -l, --limits     Enable cgroup limits
    -V, --version    Prints version information

        --cpu-max-period-micros <cpu-max-period-micros>    Set cpu.max, period part, both parts must be set together
        --cpu-max-quota-micros <cpu-max-quota-micros>      Set cpu.max, quota part, both parts must be set together
        --io-max-rbps <io-max-rbps>                        Set io.max, rbps value
        --io-max-riops <io-max-riops>                      Set io.max, riops value
        --io-max-wbps <io-max-wbps>                        Set io.max, wbps value
        --io-max-wiops <io-max-wiops>                      Set io.max, wiops value
        --memory-max <memory-max>                          Set memory.max in bytes
        --memory-swap-max <memory-swap-max>                Set memory.swap.max in bytes


Note that --limits flag must be set, otherwise the CLI will complain.


# this should succeed:
$ RUNKINS_EID=$(runkins start \
 --limits --memory-max 1000000 --memory-swap-max 0 -- \
  bash --help)
# this should exit with signal:
$ RUNKINS_EID=$(runkins start \
 --limits --memory-max 100000 --memory-swap-max 0 -- \
  bash --help)


Currently not implemented.


~230K SLoC