Containers are a lightweight alternative to fully virtualized VMs. Instead of emulating a complete Operating System (OS), containers simply use the OS of the host they run on. This implies that all containers use the same kernel, and that they can access resources from the host directly.
This is great because containers do not waste CPU power nor memory due to kernel emulation. Container run-time costs are close to zero and usually negligible. But there are also some drawbacks you need to consider:
-
You can only run Linux based OS inside containers, i.e. it is not possible to run FreeBSD or MS Windows inside.
-
For security reasons, access to host resources needs to be restricted. This is done with AppArmor, SecComp filters and other kernel features. Be prepared that some syscalls are not allowed inside containers.
{pve} uses LXC as underlying container
technology. We consider LXC as low-level library, which provides
countless options. It would be too difficult to use those tools
directly. Instead, we provide a small wrapper called pct, the
"Proxmox Container Toolkit".
The toolkit is tightly coupled with {pve}. That means that it is aware of the cluster setup, and it can use the same network and storage resources as fully virtualized VMs. You can even use the {pve} firewall, or manage containers using the HA framework.
Our primary goal is to offer an environment as one would get from a VM, but without the additional overhead. We call this "System Containers".
|
Note
|
If you want to run micro-containers (with docker, rkt, …), it is best to run them inside a VM. |
-
Integrated into {pve} graphical user interface (GUI)
-
Easy to use command line tool
pct -
Access via {pve} REST API
-
lxcfs to provide containerized /proc file system
-
AppArmor/Seccomp to improve security
-
CRIU: for live migration (planned)
-
Runs on modern Linux kernels
-
Image based deployment (templates)
-
Use {pve} storage library
-
Container setup from host (network, DNS, storage, …)
Containers use the same kernel as the host, so there is a big attack surface for malicious users. You should consider this fact if you provide containers to totally untrusted people. In general, fully virtualized VMs provide better isolation.
The good news is that LXC uses many kernel security features like AppArmor, CGroups and PID and user namespaces, which makes containers usage quite secure.
We normally try to detect the operating system type inside the container, and then modify some files inside the container to make them work as expected. Here is a short list of things we do at container startup:
- set /etc/hostname
-
to set the container name
- modify /etc/hosts
-
to allow lookup of the local hostname
- network setup
-
pass the complete network setup to the container
- configure DNS
-
pass information about DNS servers
- adapt the init system
-
for example, fix the number of spawned getty processes
- set the root password
-
when creating a new container
- rewrite ssh_host_keys
-
so that each container has unique keys
- randomize crontab
-
so that cron does not start at the same time on all containers
Changes made by {PVE} are enclosed by comment markers:
# --- BEGIN PVE --- <data> # --- END PVE ---
Those markers will be inserted at a reasonable location in the file. If such a section already exists, it will be updated in place and will not be moved.
Modification of a file can be prevented by adding a .pve-ignore.
file for it. For instance, if the file /etc/.pve-ignore.hosts
exists then the /etc/hosts file will not be touched. This can be a
simple empty file created via:
# touch /etc/.pve-ignore.hosts
Most modifications are OS dependent, so they differ between different
distributions and versions. You can completely disable modifications
by manually setting the ostype to unmanaged.
OS type detection is done by testing for certain files inside the container:
- Ubuntu
-
inspect /etc/lsb-release (
DISTRIB_ID=Ubuntu) - Debian
-
test /etc/debian_version
- Fedora
-
test /etc/fedora-release
- RedHat or CentOS
-
test /etc/redhat-release
- ArchLinux
-
test /etc/arch-release
- Alpine
-
test /etc/alpine-release
- Gentoo
-
test /etc/gentoo-release
|
Note
|
Container start fails if the configured ostype differs from the auto
detected type.
|
Container images, sometimes also referred to as “templates” or
“appliances”, are tar archives which contain everything to run a
container. You can think of it as a tidy container backup. Like most
modern container toolkits, pct uses those images when you create a
new container, for example:
pct create 999 local:vztmpl/debian-8.0-standard_8.0-1_amd64.tar.gz
{pve} itself ships a set of basic templates for most common
operating systems, and you can download them using the pveam (short
for {pve} Appliance Manager) command line utility. You can also
download TurnKey Linux containers using
that tool (or the graphical user interface).
Our image repositories contain a list of available images, and there is a cron job run each day to download that list. You can trigger that update manually with:
pveam update
After that you can view the list of available images using:
pveam available
You can restrict this large list by specifying the section you are
interested in, for example basic system images:
# pveam available --section system system archlinux-base_2015-24-29-1_x86_64.tar.gz system centos-7-default_20160205_amd64.tar.xz system debian-6.0-standard_6.0-7_amd64.tar.gz system debian-7.0-standard_7.0-3_amd64.tar.gz system debian-8.0-standard_8.0-1_amd64.tar.gz system ubuntu-12.04-standard_12.04-1_amd64.tar.gz system ubuntu-14.04-standard_14.04-1_amd64.tar.gz system ubuntu-15.04-standard_15.04-1_amd64.tar.gz system ubuntu-15.10-standard_15.10-1_amd64.tar.gz
Before you can use such a template, you need to download them into one
of your storages. You can simply use storage local for that
purpose. For clustered installations, it is preferred to use a shared
storage so that all nodes can access those images.
pveam download local debian-8.0-standard_8.0-1_amd64.tar.gz
You are now ready to create containers using that image, and you can
list all downloaded images on storage local with:
# pveam list local local:vztmpl/debian-8.0-standard_8.0-1_amd64.tar.gz 190.20MB
The above command shows you the full {pve} volume identifiers. They include the storage name, and most other {pve} commands can use them. For example you can delete that image later with:
pveam remove local:vztmpl/debian-8.0-standard_8.0-1_amd64.tar.gz
Traditional containers use a very simple storage model, only allowing
a single mount point, the root file system. This was further
restricted to specific file system types like ext4 and nfs.
Additional mounts are often done by user provided scripts. This turned
out to be complex and error prone, so we try to avoid that now.
Our new LXC based container model is more flexible regarding storage. First, you can have more than a single mount point. This allows you to choose a suitable storage for each application. For example, you can use a relatively slow (and thus cheap) storage for the container root file system. Then you can use a second mount point to mount a very fast, distributed storage for your database application. See section Mount Points for further details.
The second big improvement is that you can use any storage type
supported by the {pve} storage library. That means that you can store
your containers on local lvmthin or zfs, shared iSCSI storage,
or even on distributed storage systems like ceph. It also enables us
to use advanced storage features like snapshots and clones. vzdump
can also use the snapshot feature to provide consistent container
backups.
Last but not least, you can also mount local devices directly, or mount local directories using bind mounts. That way you can access local storage inside containers with zero overhead. Such bind mounts also provide an easy way to share data between different containers.
|
Warning
|
Because of existing issues in the Linux kernel’s freezer subsystem the usage of FUSE mounts inside a container is strongly advised against, as containers need to be frozen for suspend or snapshot mode backups. |
If FUSE mounts cannot be replaced by other mounting mechanisms or storage technologies, it is possible to establish the FUSE mount on the Proxmox host and use a bind mount point to make it accessible inside the container.
Quotas allow to set limits inside a container for the amount of disk space that each user can use. This only works on ext4 image based storage types and currently does not work with unprivileged containers.
Activating the quota option causes the following mount options to be
used for a mount point:
usrjquota=aquota.user,grpjquota=aquota.group,jqfmt=vfsv0
This allows quotas to be used like you would on any other system. You
can initialize the /aquota.user and /aquota.group files by running
quotacheck -cmug / quotaon /
and edit the quotas via the edquota command. Refer to the documentation
of the distribution running inside the container for details.
|
Note
|
You need to run the above commands for every mount point by passing
the mount point’s path instead of just /.
|
The standard Posix Access Control Lists are also available inside containers. ACLs allow you to set more detailed file ownership than the traditional user/ group/others model.
By default additional mount points besides the Root Disk mount point are not included in backups. You can reverse this default behavior by setting the Backup option on a mount point.
By default additional mount points are replicated when the Root Disk
is replicated. If you want the {pve} storage replication mechanism to skip a
mount point when starting a replication job, you can set the
Skip replication option on that mount point.
As of {pve} 5.0, replication requires a storage of type zfspool, so adding a
mount point to a different type of storage when the container has replication
configured requires to Skip replication for that mount point.
General settings of a container include
-
the Node : the physical server on which the container will run
-
the CT ID: a unique number in this {pve} installation used to identify your container
-
Hostname: the hostname of the container
-
Resource Pool: a logical group of containers and VMs
-
Password: the root password of the container
-
SSH Public Key: a public key for connecting to the root account over SSH
-
Unprivileged container: this option allows to choose at creation time if you want to create a privileged or unprivileged container.
Security is done by dropping capabilities, using mandatory access control (AppArmor), SecComp filters and namespaces. The LXC team considers this kind of container as unsafe, and they will not consider new container escape exploits to be security issues worthy of a CVE and quick fix. So you should use this kind of containers only inside a trusted environment, or when no untrusted task is running as root in the container.
This kind of containers use a new kernel feature called user namespaces. The root UID 0 inside the container is mapped to an unprivileged user outside the container. This means that most security issues (container escape, resource abuse, …) in those containers will affect a random unprivileged user, and so would be a generic kernel security bug rather than an LXC issue. The LXC team thinks unprivileged containers are safe by design.
|
Note
|
If the container uses systemd as an init system, please be aware the systemd version running inside the container should be equal or greater than 220. |
You can restrict the number of visible CPUs inside the container using
the cores option. This is implemented using the Linux cpuset
cgroup (control group). A special task inside pvestatd tries
to distribute running containers among available CPUs. You can view
the assigned CPUs using the following command:
# pct cpusets --------------------- 102: 6 7 105: 2 3 4 5 108: 0 1 ---------------------
Containers use the host kernel directly, so all task inside a container are handled by the host CPU scheduler. {pve} uses the Linux CFS (Completely Fair Scheduler) scheduler by default, which has additional bandwidth control options.
cpulimit:
|
You can use this option to further limit assigned CPU time. Please note that this is a floating point number, so it is perfectly valid to assign two cores to a container, but restrict overall CPU consumption to half a core. cores: 2 cpulimit: 0.5 |
cpuunits:
|
This is a relative weight passed to the kernel scheduler. The larger the number is, the more CPU time this container gets. Number is relative to the weights of all the other running containers. The default is 1024. You can use this setting to prioritize some containers. |
Container memory is controlled using the cgroup memory controller.
memory:
|
Limit overall memory usage. This corresponds
to the |
swap:
|
Allows the container to use additional swap memory from the
host swap space. This corresponds to the |
The root mount point is configured with the rootfs property, and you can
configure up to 10 additional mount points. The corresponding options
are called mp0 to mp9, and they can contain the following setting:
Currently there are basically three types of mount points: storage backed mount points, bind mounts and device mounts.
rootfs configurationrootfs: thin1:base-100-disk-1,size=8G
Storage backed mount points are managed by the {pve} storage subsystem and come in three different flavors:
-
Image based: these are raw images containing a single ext4 formatted file system.
-
ZFS subvolumes: these are technically bind mounts, but with managed storage, and thus allow resizing and snapshotting.
-
Directories: passing
size=0triggers a special case where instead of a raw image a directory is created.
|
Note
|
The special option syntax STORAGE_ID:SIZE_IN_GB for storage backed
mount point volumes will automatically allocate a volume of the specified size
on the specified storage. E.g., calling
pct set 100 -mp0 thin1:10,mp=/path/in/container will allocate a 10GB volume
on the storage thin1 and replace the volume ID place holder 10 with the
allocated volume ID.
|
Bind mounts allow you to access arbitrary directories from your Proxmox VE host inside a container. Some potential use cases are:
-
Accessing your home directory in the guest
-
Accessing an USB device directory in the guest
-
Accessing an NFS mount from the host in the guest
Bind mounts are considered to not be managed by the storage subsystem, so you cannot make snapshots or deal with quotas from inside the container. With unprivileged containers you might run into permission problems caused by the user mapping and cannot use ACLs.
|
Note
|
The contents of bind mount points are not backed up when using vzdump.
|
|
Warning
|
For security reasons, bind mounts should only be established
using source directories especially reserved for this purpose, e.g., a
directory hierarchy under /mnt/bindmounts. Never bind mount system
directories like /, /var or /etc into a container - this poses a
great security risk.
|
|
Note
|
The bind mount source path must not contain any symlinks. |
For example, to make the directory /mnt/bindmounts/shared accessible in the
container with ID 100 under the path /shared, use a configuration line like
mp0: /mnt/bindmounts/shared,mp=/shared in /etc/pve/lxc/100.conf.
Alternatively, use pct set 100 -mp0 /mnt/bindmounts/shared,mp=/shared to
achieve the same result.
Device mount points allow to mount block devices of the host directly into the
container. Similar to bind mounts, device mounts are not managed by {PVE}'s
storage subsystem, but the quota and acl options will be honored.
|
Note
|
Device mount points should only be used under special circumstances. In most cases a storage backed mount point offers the same performance and a lot more features. |
|
Note
|
The contents of device mount points are not backed up when using vzdump.
|
You can configure up to 10 network interfaces for a single
container. The corresponding options are called net0 to net9, and
they can contain the following setting:
After creating your containers, you probably want them to start automatically when the host system boots. For this you need to select the option Start at boot from the Options Tab of your container in the web interface, or set it with the following command:
pct set <ctid> -onboot 1
If you want to fine tune the boot order of your containers, you can use the following parameters :
-
Start/Shutdown order: Defines the start order priority. E.g. set it to 1 if you want the CT to be the first to be started. (We use the reverse startup order for shutdown, so a container with a start order of 1 would be the last to be shut down)
-
Startup delay: Defines the interval between this container start and subsequent containers starts . E.g. set it to 240 if you want to wait 240 seconds before starting other containers.
-
Shutdown timeout: Defines the duration in seconds {pve} should wait for the container to be offline after issuing a shutdown command. By default this value is set to 60, which means that {pve} will issue a shutdown request, wait 60s for the machine to be offline, and if after 60s the machine is still online will notify that the shutdown action failed.
Please note that containers without a Start/Shutdown order parameter will always start after those where the parameter is set, and this parameter only makes sense between the machines running locally on a host, and not cluster-wide.
You can add a hook script to CTs with the config property hookscript.
pct set 100 -hookscript local:snippets/hookscript.pl
It will be called during various phases of the guests lifetime.
For an example and documentation see the example script under
/usr/share/pve-docs/examples/guest-example-hookscript.pl.
It is possible to use the vzdump tool for container backup. Please
refer to the vzdump manual page for details.
Restoring container backups made with vzdump is possible using the
pct restore command. By default, pct restore will attempt to restore as much
of the backed up container configuration as possible. It is possible to override
the backed up configuration by manually setting container options on the command
line (see the pct manual page for details).
|
Note
|
pvesm extractconfig can be used to view the backed up configuration
contained in a vzdump archive.
|
There are two basic restore modes, only differing by their handling of mount points:
If neither the rootfs parameter nor any of the optional mpX parameters
are explicitly set, the mount point configuration from the backed up
configuration file is restored using the following steps:
-
Extract mount points and their options from backup
-
Create volumes for storage backed mount points (on storage provided with the
storageparameter, or default local storage if unset) -
Extract files from backup archive
-
Add bind and device mount points to restored configuration (limited to root user)
|
Note
|
Since bind and device mount points are never backed up, no files are restored in the last step, but only the configuration options. The assumption is that such mount points are either backed up with another mechanism (e.g., NFS space that is bind mounted into many containers), or not intended to be backed up at all. |
This simple mode is also used by the container restore operations in the web interface.
By setting the rootfs parameter (and optionally, any combination of mpX
parameters), the pct restore command is automatically switched into an
advanced mode. This advanced mode completely ignores the rootfs and mpX
configuration options contained in the backup archive, and instead only
uses the options explicitly provided as parameters.
This mode allows flexible configuration of mount point settings at restore time, for example:
-
Set target storages, volume sizes and other options for each mount point individually
-
Redistribute backed up files according to new mount point scheme
-
Restore to device and/or bind mount points (limited to root user)
pct is the tool to manage Linux Containers on {pve}. You can create
and destroy containers, and control execution (start, stop, migrate,
…). You can use pct to set parameters in the associated config file,
like network configuration or memory limits.
Create a container based on a Debian template (provided you have already downloaded the template via the web interface)
pct create 100 /var/lib/vz/template/cache/debian-8.0-standard_8.0-1_amd64.tar.gz
Start container 100
pct start 100
Start a login session via getty
pct console 100
Enter the LXC namespace and run a shell as root user
pct enter 100
Display the configuration
pct config 100
Add a network interface called eth0, bridged to the host bridge vmbr0,
set the address and gateway, while it’s running
pct set 100 -net0 name=eth0,bridge=vmbr0,ip=192.168.15.147/24,gw=192.168.15.1
Reduce the memory of the container to 512MB
pct set 100 -memory 512
In case pct start is unable to start a specific container, it might be
helpful to collect debugging output by running lxc-start (replace ID with
the container’s ID):
lxc-start -n ID -F -l DEBUG -o /tmp/lxc-ID.log
This command will attempt to start the container in foreground mode, to stop the container run pct shutdown ID or pct stop ID in a second terminal.
The collected debug log is written to /tmp/lxc-ID.log.
|
Note
|
If you have changed the container’s configuration since the last start
attempt with pct start, you need to run pct start at least once to also
update the configuration used by lxc-start.
|
If you have a cluster, you can migrate your Containers with
pct migrate <vmid> <target>
This works as long as your Container is offline. If it has local volumes or mountpoints defined, the migration will copy the content over the network to the target host if the same storage is defined there.
If you want to migrate online Containers, the only way is to use restart migration. This can be initiated with the -restart flag and the optional -timeout parameter.
A restart migration will shut down the Container and kill it after the specified timeout (the default is 180 seconds). Then it will migrate the Container like an offline migration and when finished, it starts the Container on the target node.
The /etc/pve/lxc/<CTID>.conf file stores container configuration,
where <CTID> is the numeric ID of the given container. Like all
other files stored inside /etc/pve/, they get automatically
replicated to all other cluster nodes.
|
Note
|
CTIDs < 100 are reserved for internal purposes, and CTIDs need to be unique cluster wide. |
ostype: debian arch: amd64 hostname: www memory: 512 swap: 512 net0: bridge=vmbr0,hwaddr=66:64:66:64:64:36,ip=dhcp,name=eth0,type=veth rootfs: local:107/vm-107-disk-1.raw,size=7G
Those configuration files are simple text files, and you can edit them
using a normal text editor (vi, nano, …). This is sometimes
useful to do small corrections, but keep in mind that you need to
restart the container to apply such changes.
For that reason, it is usually better to use the pct command to
generate and modify those files, or do the whole thing using the GUI.
Our toolkit is smart enough to instantaneously apply most changes to
running containers. This feature is called "hot plug", and there is no
need to restart the container in that case.
Container configuration files use a simple colon separated key/value format. Each line has the following format:
# this is a comment OPTION: value
Blank lines in those files are ignored, and lines starting with a #
character are treated as comments and are also ignored.
It is possible to add low-level, LXC style configuration directly, for example:
lxc.init_cmd: /sbin/my_own_init
or
lxc.init_cmd = /sbin/my_own_init
Those settings are directly passed to the LXC low-level tools.
When you create a snapshot, pct stores the configuration at snapshot
time into a separate snapshot section within the same configuration
file. For example, after creating a snapshot called “testsnapshot”,
your configuration file will look like this:
memory: 512 swap: 512 parent: testsnaphot ... [testsnaphot] memory: 512 swap: 512 snaptime: 1457170803 ...
There are a few snapshot related properties like parent and
snaptime. The parent property is used to store the parent/child
relationship between snapshots. snaptime is the snapshot creation
time stamp (Unix epoch).
Container migrations, snapshots and backups (vzdump) set a lock to
prevent incompatible concurrent actions on the affected container. Sometimes
you need to remove such a lock manually (e.g., after a power failure).
pct unlock <CTID>
|
Caution
|
Only do that if you are sure the action which set the lock is no longer running. |