In Part 1, we created a working BIND container with local data storage.  We can make changes on the local system that will get picked up in the running container.  In this part, we’ll explore how we can manage the service from the host with systemctl.

In Part 1, we installed two OCI hooks, oci-register-machine and oci-systemd-hook.  OCI hooks are executed either before the container process is executed (prestart) or after shut down (poststop).  These two hooks provide the integration points to systemd on the host.  We used machinectl in the previous post to copy files out of our test container, but we didn’t look at what else this integration can do for us.

Using machinectl list, we see what containers systemd knows about.  The MACHINE field holds the machine-id created by oci-register-machine from the container ID.  We can use that ID with both machinectl and systemctl.

With machinectl status we can see some interesting things about our container, like IP address, OS release info, and the systemd hierarchy running.  Since this is from the host view, you’ll see the host PID for /sbin/init not PID 1.

[root@rhel7-host bind]# machinectl status 47d2c93035cafb1174dedd924cfa4308 

47d2c93035cafb1174dedd924cfa4308(47d2c93035cafb1174dedd924cfa4308)
           Since: Mon 2017-05-15 12:01:19 EDT; 3h 58min ago
          Leader: 2734 (systemd)
         Service: docker; class container
            Root: /var/lib/docker/devicemapper/mnt/5d897b599c6b003df74968f73ad56
         Address: 172.17.0.2
                  fe80::42:acff:fe11:2
              OS: Red Hat Enterprise Linux Server 7.3 (Maipo)
            Unit: docker-47d2c93035cafb1174dedd924cfa4308ebb8b924bc0b0b661f60ffe
                  ├─2734 /sbin/init
                  └─system.slice
                    ├─named.service
                    │ └─2788 /usr/sbin/named -u named
                   ├─dbus.service
                    │ └─2777 /bin/dbus-daemon --system --address=systemd: --nofo
                    └─systemd-journald.service
                      └─2766 /usr/lib/systemd/systemd-journald

CONTAINER SYSTEMD

We can also get status in the container via systemctl on the host by specifying the -M option.  You can query the overall systemd status, the process status, and interact with the process as if it were on the host.

To see the internal container state and check for failed or queued jobs:

[root@rhel7-host bind]# systemctl -M 47d2c93035cafb1174dedd924cfa4308 status

To see the status of the named service in container:

[root@rhel7-host bind]# systemctl -M 47d2c93035cafb1174dedd924cfa4308 status named

To see what actions are available to us via the unit file, you can use show service.  Let’s grep for Exec to see what’s available.  

[root@rhel7-host bind]# systemctl -M 47d2c93035cafb1174dedd924cfa4308 show named | grep Exec

ExecStart={ path=/usr/sbin/named ; argv[]=/usr/sbin/named -u named $OPTIONS ; ignore_errors=no ; start_time=[Mon 2017-05-15 12:01:20 EDT] ; stop_time=[Mon 2017-05-15 12:01:20 EDT] ; pid=26 ; code=exited ; status=0 }

ExecReload={ path=/bin/sh ; argv[]=/bin/sh -c /usr/sbin/rndc reload > /dev/null 2>&1 || /bin/kill -HUP $MAINPID ; ignore_errors=no ; start_time=[n/a] ; stop_time=[n/a] ; pid=0 ; code=(null) ; status=0/0 }

ExecStop={ path=/bin/sh ; argv[]=/bin/sh -c /usr/sbin/rndc stop > /dev/null 2>&1 || /bin/kill -TERM $MAINPID ; ignore_errors=no ; start_time=[n/a] ; stop_time=[n/a] ; pid=0 ; code=(null) ; status=0/0 }

 

The named unit file exposes a start, stop, and reload action.  We also can use reload-or-restart and let systemd determine what to do instead of searching the show output.

[root@rhel7-host bind]# systemctl -M 47d2c93035cafb1174dedd924cfa4308 reload-or-restart named

[root@rhel7-host bind]# systemctl -M 47d2c93035cafb1174dedd924cfa4308 status named
 named.service - Berkeley Internet Name Domain (DNS)
   Loaded: loaded (/usr/lib/systemd/system/named.service; enabled; vendor preset: disabled)
   Active: active (running) since Mon 2017-05-15 12:01:20 EDT; 4h 32min ago
  Process: 54 ExecReload=/bin/sh -c /usr/sbin/rndc reload > /dev/null 2>&1 || /bin/kill -HUP $MAINPID (code=exited, status=0/SUCCESS)
  Process: 26 ExecStart=/usr/sbin/named -u named $OPTIONS (code=exited, status=0/SUCCESS)
  Process: 24 ExecStartPre=/bin/bash -c if [ ! "$DISABLE_ZONE_CHECKING" == "yes" ]; then /usr/sbin/named-checkconf -z /etc/named.conf; else echo "Checking of zone files is disabled"; fi (code=exited, status=0/SUCCESS)
 Main PID: 27 (khugepaged)
   CGroup: /system.slice/docker-47d2c93035cafb1174dedd924cfa4308ebb8b924bc0b0b661f60ffee3e4bf715.scope/system.slice/named.service
           └─2788 /usr/sbin/named -u named
           ‣ 27 [khugepaged]

Be sure to note, if you stop the service running in the container with systemctl, that doesn’t stop the container.  We’re directly manipulating the service in these commands not the container.  You could use systemctl -M halt to stop the container.

CREATING SYSTEMD UNIT

Now that we’ve looked at manually manipulating the service from systemctl, the only thing left is to start the container when the host starts, so we’re never without DNS.  We’ll create a simple unit file for the service, and then enable it on the host.  

[root@rhel7-host ~]# vi named-container.service
[Unit]
Description=Containerized BIND service
Requires=docker.service
After=docker.service

[Service]
Restart=on-failure
RestartSec=10
ExecStart=/usr/bin/docker start -a %p
ExecStop=-/usr/bin/docker stop -t 2 %p

[Install]
WantedBy=multi-user.target

 

Since the name of the unit matches the name of the container we built, we use the %p specifier to pass the right argument.  In this case, that’s named-container.  If you wanted different names for the unit and container, you could replace the %p with the name of the container.  Copy the unit file to /etc/systemd/system and we can manipulate the container like a service on the host.

Let’s use restart so we know the container is managed by systemd and not already running from a previous step, then we can check the status of BIND inside the container.

[root@rhel7-host bind]# cp named-container.service /etc/systemd/system/
[root@rhel7-host bind]# systemctl enable named-container
Created symlink from /etc/systemd/system/multi-user.target.wants/named-container.service to /etc/systemd/system/named-container.service.

[root@rhel7-host bind]# systemctl restart named-container
[root@rhel7-host bind]# systemctl status named-container

 named-container.service - Containerized BIND service
   Loaded: loaded (/etc/systemd/system/named-container.service; enabled; vendor preset: disabled)
   Active: active (running) since Wed 2017-05-10 16:45:44 EDT; 26s ago
 Main PID: 23455 (docker-current)
   Memory: 5.9M
   CGroup: /system.slice/named-container.service
           └─23455 /usr/bin/docker-current start -a named-container
May 10 16:45:44 rhel7-host systemd[1]: Started Containerized BIND service.
May 10 16:45:44 rhel7-host systemd[1]: Starting Containerized BIND service...

[root@rhel7-host bind]# systemctl -M 47d2c93035cafb1174dedd924cfa4308 status named
 named.service - Berkeley Internet Name Domain (DNS)
   Loaded: loaded (/usr/lib/systemd/system/named.service; enabled; vendor preset: disabled)
   Active: active (running) since Wed 2017-05-10 16:44:52 EDT; 1min 49s ago
  Process: 26 ExecStart=/usr/sbin/named -u named $OPTIONS (code=exited, status=0/SUCCESS)
  Process: 24 ExecStartPre=/bin/bash -c if [ ! "$DISABLE_ZONE_CHECKING" == "yes" ]; then /usr/sbin/named-checkconf -z /etc/named.conf; else echo "Checking of zone files is disabled"; fi (code=exited, status=0/SUCCESS)
 Main PID: 27 (khugepaged)
   CGroup: /system.slice/docker-47d2c93035cafb1174dedd924cfa4308ebb8b924bc0b0b661f60ffee3e4bf715.scope/system.slice/named.service
           └─23438 /usr/sbin/named -u named
           ‣ 27 [khugepaged]

There we have it, a containerized BIND service that can be updated independently of any other services of the system, yet easily managed and updated from the host.

LOOKING AHEAD

This concept of a system service container isn’t quite the same as what some others are calling “system containers”.  System containers are simply stand alone containers that provide services.  The distinction isn’t between apps and OS containers, but if the application in the container benefits from distribution and orchestration.  Typically, BIND doesn’t need to scale up and down based on usage, so it wouldn’t benefit from a full orchestration configuration.  

System containers are something we think can be very useful.  In the Container Catalog you'll find some supported examples, like etcd.  This container uses the atomic command found on the Red Hat Enterprise Linux Atomic Host to install and configure the service.  These system containers are built somewhat differently than we did here, using runc to launch the container.  There are definite benefits to the approach and you can read more about the install and run process for etcd in the official documentation.

If you'd like to see how the upstream work is progressing on creating and managing these sorts of system containers, you can take a look at the Project Atomic repository.

 


Additional Resources:

GitHub repository for accompanying files: https://github.com/nzwulfin/named-container