Containers created using osctl are actually LXC containers. osctl provides an abstracted interface to control these containers, while in fact every container can be run under a different user and within a different resource group. It can be useful to know what is osctld actually doing for you and how you can manipulate the LXC containers directly. The first step is understanding how osctld uses ZFS pools.
Let's create a pool and install it into osctld:
# Install a pool dd if=/dev/zero of=/tank.zpool bs=1M count=4096 zpool create tank /tank.zpool osctl pool install tank # List pool layout zfs list NAME USED AVAIL REFER MOUNTPOINT tank 5.50M 874M 41.5K /tank tank/conf 30.5K 874M 30.5K /tank/conf tank/ct 24K 874M 24K /tank/ct tank/log 30K 874M 30K /tank/log tank/repository 5.03M 874M 5.03M /tank/repository tank/user 55.5K 874M 24K /tank/user
tank/conf is used to store configuration files for users, groups,
containers, repositories and migrations in respective subdirectories.
The configuration files actually define those entities, i.e. when there is no
configuration file for a container, the container does not exist.
tank/ct contains subdatasets, one for every container, named by container ids.
These datasets contain the containers' rootfs and have their user/group ids
shifted into the user namespace.
tank/log is used to store log files generated by
lxc-start, one for each
tank/repository contains local caches for remote template repositories.
Whenever you create a container using a template from a remote repository,
the downloaded template is cached for later use.
Users representing user namespaces are stored in dataset
user has a subdataset bearing the user's name. We're calling these subdatasets
user dataset or user dir (they are not the user's home directory). User
directories contain subdirectories for every user/group combination that any
containers are using. These directories are what LXC calls LXC home or
LXC path, which by default in LXC is
/var/lib/lxc. In vpsAdminOS, there is
one LXC path for every user/group combination used by containers.
Inside LXC path are directories representing containers, containing config
files read by LXC.
Let's create a container and see it in action:
# Create a one user namespace and a container osctl user new --ugid 5000 --map 0:666000:65536 myuser01 osctl ct new --user myuser01 --distribution ubuntu --version 16.04 myct01 # Review pool layout zfs list -oname,uidmap,gidmap,mountpoint NAME UIDMAP GIDMAP MOUNTPOINT tank none none /tank tank/conf none none /tank/conf tank/ct none none /tank/ct tank/ct/myct01 0:666000:65536 0:666000:65536 /tank/ct/myct01 # Container rootfs tank/log none none /tank/log tank/repository none none /tank/repository tank/user none none /tank/user tank/user/myuser01 none none /tank/user/myuser01 # User dataset/directory
Because every user, group and container is defined by multiple datasets, files or directories, osctl provides a way to see what those entities are and what is their state. First, review the user namespace:
osctl user assets myuser01 TYPE PATH VALID PURPOSE dataset tank/user/myuser01 true User's home dataset directory /tank/user/myuser01 true User directory directory /tank/user/myuser01/.home true Home directory file /tank/conf/user/myuser01.yml true osctld's user config entry /etc/passwd true System user entry /etc/group true System group
VALID is not
true for some asset, it means that either it does not exist
or is invalid, such as wrong file owner, group or access mode. Use option
to show these errors.
Because we haven't selected a group when creating the container, osctld put it into the default group. Let's see its assets:
group assets /default TYPE PATH VALID PURPOSE file /tank/conf/group/default/config.yml true osctld's group config directory /tank/user/myuser01/group.default/cts true LXC path for myuser01:/default
As you can see, there is one LXC path for user
myuser01 and group
where our container resides. Let's review the container:
ct assets myct01 TYPE PATH VALID PURPOSE dataset tank/ct/myct01 true Container's rootfs dataset directory /tank/ct/myct01/private true Container's rootfs directory /tank/user/myuser01/group.default/cts/myct01 true LXC configuration file /tank/user/myuser01/group.default/cts/myct01/config true LXC base config file /tank/user/myuser01/group.default/cts/myct01/network true LXC network config file /tank/user/myuser01/group.default/cts/myct01/prlimits true LXC resource limits file /tank/user/myuser01/group.default/cts/myct01/mounts true LXC mounts file /tank/user/myuser01/group.default/cts/myct01/.bashrc true Shell configuration file for osctl ct su file /tank/conf/ct/myct01.yml true Container config for osctld file /tank/log/ct/myct01.log true LXC log file
Whenever you need to manage a container from the host, you might want to
into the container's directory. osctl has helpers for that:
osctl ct cd <id>for the container's rootfs
osctl ct cd -l | --lxc <id>for the container's LXC configuration directory
osctl ct cd -r | --runtime <id>for the mounted rootfs, available only when the container is running.
A new shell with a modified prompt is spawned. Simply exit the shell to return to your previous session.
Administrators can use
osctl ct attach to enter containers and get root shell,
without the need of knowing password for SSH or
osctl ct console. osctl
is by default trying to unify the shell interface from the administrator's
point of view, so that whatever container you enter, you always get the same
prompt and behaviour.
ct attach myct01 [CT myct01] root@myct01:/# cat /etc/alpine-release 3.7.0 ct attach myct02 [CT myct02] root@myct02:/# lsb_release -a No LSB modules are available. Distributor ID: Ubuntu Description: Ubuntu 18.04 LTS Release: 18.04 Codename: bionic
This is done by ignoring shell configuration from the container and configuring
the shell from the host. If you wish to open the shell and load configuration
from the container, you can use option
Identifying container processes from the host
When you find a misbehaving process on the host, you need to identify to which
container the process belongs to, so that you may notify its owner or set
a limit. You could do this manually by looking at
identifying the container by the cgroup path. This is what
osctl ct pid does.
You give it process IDs as seen on the host and it will tell you what containers
they belong to.
osctl ct pid 22690 23292 PID CONTAINER CTPID NAME 22690 tank:myct01 226 agetty 23292 tank:myct01 228 agetty
osctl can also read process IDs from standard input:
osctl ct pid - PID CONTAINER CTPID NAME > 22690 22690 tank:myct01 226 agetty > 23292 23292 tank:myct01 228 agetty
Container resource monitor
osctl ct top is a top-like TUI application that monitors real-time resource
usage. Instead of processes it monitors containers.
It can work in two modes: real-time and cumulative. Real-time shows resource
consumption since the last update, i.e. 1 second by default. Cumulative shows
resource usage since the program was started. Mode can be switched by pressing
Left and right arrows can be used to change the sort column,
r to reverse
sort order. Up and down arrows move selection cursor up and down, spacebar will
highlight selected container for easier spotting. Enter/return key will open
htop focused on the selected container, see below. Press
? to show all key
htop in vpsAdminOS is patched to be able to identify to which containers
processes belong. It has three extra columns:
NSUID shows the process user ID as it is seen inside the
container, i.e. its user namespace.
Our htop can filter processes by container ID. You can use CLI option
--container to select which container to filter at start, e.g.:
htop -c tank:10796. Filters can also be changed at runtime, press
and select a container, the host, or all processes.
You don't need to remember the path to a container's log file, because osctl can either dump it for you or just print the path:
osctl ct log cat <id> osctl ct log path <id>
When debugging a container that's failing to start, you can enable LXC debug
osctl ct start --debug.
Should you want to switch to a container's user and use LXC utilities directly,
you have to use
osctl ct su <id>. Using
/bin/su is insufficient, because standard
su does not manage cgroups.
osctl ct su will open a shell with the same
environment as osctld itself is using to start the container.
osctl ct su myct01 Opened shell for: User: myuser01 Group: /default CT: myct01 Available LXC utilities: lxc-attach lxc-cgroup lxc-console lxc-device lxc-execute lxc-info lxc-ls lxc-monitor lxc-stop lxc-top lxc-wait Implicit arguments: -P /tank/user/myuser01/default -n myct01 Do not use this shell to manipulate any other container than myct01. [unsmyuser01@vpsadminos:/tank/user/myuser01/group.default/cts]$
The shell uses
/tank/user/myuser01/group.default/cts/myct01/.bashrc instead of
~/.bashrc, see user assets above. The shell can be used to manipulate only
the chosen container, because every container has a specific cgroup path,
so running other containers would put them to a wrong group.
The LXC utilities are unmodified except for the implicit arguments, which let you forget that there is a non-default LXC path and some container id.