Understanding Cgroup Migration on SystemD

systemd.md

#Make the following files:

Note: depending on your system you may need to modify the following variables SYSTEMD_LIB_DIR and GROUP_DIR

bad-cgroup-demo.service

[Unit]
Description=Systemd bad cgroup demo

[Service]
Type=forking
ExecStart=/tmp/systemd-demo-unit
StandardOutput=journal
Delegate=false

[Install]
WantedBy=multi-user.target
Alias=bad-cgroup-demo.service

good-cgroup-demo.service

[Unit]
Description=Systemd good cgroup demo

[Service]
Type=forking
ExecStart=/tmp/systemd-demo-unit
StandardOutput=journal
Delegate=true

[Install]
WantedBy=multi-user.target
Alias=good-cgroup-demo.service

systemd-demo-unit.cc

#include <fstream>
#include <unordered_set>

#include <sys/stat.h>

#include <unistd.h>

using std::string;
using std::ofstream;
using std::ifstream;
using std::unordered_set;

const std::string GROUP_DIR = "/sys/fs/cgroup/";
const std::string SUBSYSTEM = "memory";
const std::string PROC_NAME = "systemd-demo-unit";
const std::string PROC_PATH = GROUP_DIR + SUBSYSTEM + "/" + PROC_NAME;
const std::string TASKS_PATH = PROC_PATH + "/tasks";

void force_proc_dir()
{
  struct stat st;
  if (stat(PROC_PATH.c_str(), &st) == 0) {
    if(!(st.st_mode & S_IFDIR)) { exit(EXIT_FAILURE); }
    // If we get here the directory already exists.
  } else {
    if (mkdir(PROC_PATH.c_str(), S_IRWXU | S_IRWXG | S_IROTH | S_IXOTH) != 0) { exit(EXIT_FAILURE); }
  }
}

void add_task(pid_t pid)
{
  ofstream out(TASKS_PATH.c_str());
  if (!out.is_open()) { exit(EXIT_FAILURE); }
  out << pid;
  if (out.fail()) { exit(EXIT_FAILURE); }
  out.close();
}

unordered_set<pid_t> get_tasks()
{
  unordered_set<pid_t> result;
  ifstream in(TASKS_PATH.c_str());

  while (in.is_open() && !in.bad() && !in.eof() && !in.fail()) {
    pid_t pid;
    in >> pid;
    result.emplace(pid);
  }

  in.close();
  return result;
}

int main()
{
  pid_t pid = fork();

  if (pid < 0) {
    // Bail on failed fork.
    exit(EXIT_FAILURE);
  }

  if (pid == 0) {
    // Child.
    sleep(3);

    auto tasks = get_tasks();
    for (const pid_t pid : tasks) {
      printf("Child task view = [%d]\n", pid);
    }

    printf("Reloading daemon\n");
    system("systemctl daemon-reload");
    printf("Restarting syslog\n");
    system("systemctl restart rsyslog.service");

    sleep(3);

    tasks = get_tasks();
    for (const pid_t pid : tasks) {
      printf("Child task view = [%d]\n", pid);
    }

    return 0;
  } else {
    // Parent.

    // Make sure the cgroup dir exists.
    force_proc_dir();

    sleep(1);

    // Write the child pid to the cgroup tasks.
    printf("Adding child task [%d] to tasks\n", pid);
    add_task(pid);

    auto tasks = get_tasks();
    for (const pid_t pid : tasks) {
      printf("Parent task view = [%d]\n", pid);
    }

    printf("Parent exiting\n");
  }

  return 0;
}

cgroup-demo.sh

#!/usr/bin/env bash

SYSTEMD_LIB_DIR=/lib/systemd/system/

if [ "$EUID" -ne 0 ]
  then echo "Please run as root"
  exit
fi

cp good-cgroup-demo.service $SYSTEMD_LIB_DIR
cp bad-cgroup-demo.service $SYSTEMD_LIB_DIR

systemctl daemon-reload

echo "Running Bad Cgroup Demo:"

systemctl start bad-cgroup-demo.service
sleep 10
journalctl -u bad-cgroup-demo.service -n9

echo "Running Good Cgroup Demo:";

systemctl start good-cgroup-demo.service
sleep 10
journalctl -u good-cgroup-demo.service -n9

echo "Cleaning up..."

rm "$SYSTEMD_LIB_DIR""good-cgroup-demo.service"
rm "$SYSTEMD_LIB_DIR""bad-cgroup-demo.service"

echo "Finished"

#Run the following commands:

// Allow the script to be executed.
chmod +x cgroup-demo.sh
// Compile the demo binary.
clang++-3.6 -std=c++11 systemd-demo-unit.cc -o systemd-demo-unit
// Copy the binary to /tmp
cp systemd-demo-unit /tmp/
// Run the script as ROOT
./cgroup-demo.sh

###You should see output similar to this:

Running Bad Cgroup Demo:
-- Logs begin at ***, end at ***. --
DDD 12:51:53 XXX systemd[1]: Starting Systemd bad cgroup demo...
DDD 12:51:54 XXX systemd-demo-unit[13824]: Adding child task [13825] to tasks
DDD 12:51:54 XXX systemd-demo-unit[13824]: Parent task view = [13825]
DDD 12:51:54 XXX systemd-demo-unit[13824]: Parent exiting
DDD 12:51:54 XXX systemd[1]: Started Systemd bad cgroup demo.
DDD 12:51:59 XXX systemd-demo-unit[13824]: Child task view = [13825]
DDD 12:51:59 XXX systemd-demo-unit[13824]: Reloading daemon
DDD 12:51:59 XXX systemd-demo-unit[13824]: Restarting syslog
DDD 12:51:59 XXX systemd-demo-unit[13824]: Child task view = [0]
Running Good Cgroup Demo:
-- Logs begin at *** , end at ***. --
DDD 12:52:04 XXX systemd[1]: Starting Systemd good cgroup demo...
DDD 12:52:05 XXX systemd-demo-unit[13858]: Adding child task [13859] to tasks
DDD 12:52:05 XXX systemd-demo-unit[13858]: Parent task view = [13859]
DDD 12:52:05 XXX systemd-demo-unit[13858]: Parent exiting
DDD 12:52:05 XXX systemd[1]: Started Systemd good cgroup demo.
DDD 12:52:10 XXX systemd-demo-unit[13858]: Child task view = [13859]
DDD 12:52:10 XXX systemd-demo-unit[13858]: Reloading daemon
DDD 12:52:10 XXX systemd-demo-unit[13858]: Restarting syslog
DDD 12:52:10 XXX systemd-demo-unit[13858]: Child task view = [13859]
Cleaning up...
Finished

Notice that when the Delegate is set to false, the child task gets removed from the cgroup.