Changes

5,105 bytes removed ,  19:05, 19 September 2014
no edit summary
Line 1: Line 1: −
This article describes how to perform checkpoint-restore for an LXC container. There are two ways to do this, the first part of the article describes how to dump a container with LXC directly, and the second part describes how to do it using tools under development in LXC upstream.
+
== Requirements ==
   −
== Preparing a Linux Container ==
+
You should have built and installed a recent (>= 1.3.1) version of CRIU.
   −
=== Requirements ===
+
== Checkpointing and restoring a container ==
 
  −
* A console should be disabled (<code>lxc.console = none</code>)
  −
* udev should not run inside containers (<code>mv /sbin/udevd{,.bcp}</code>)
  −
 
  −
=== Preparing a host environment ===
  −
 
  −
* Mount cgroupfs
  −
$ mount -t cgroup c /cgroup
  −
 
  −
* Create a network bridge
  −
# cat /etc/sysconfig/network-scripts/ifcfg-br0
  −
DEVICE=br0
  −
TYPE=Bridge
  −
BOOTPROTO=dhcp
  −
ONBOOT=yes
  −
DELAY=5
  −
NM_CONTROLLED=n
  −
$ cat /etc/sysconfig/network-scripts/ifcfg-eth0
  −
DEVICE="eth0"
  −
NM_CONTROLLED="no"
  −
ONBOOT="yes"
  −
BRIDGE=br0
  −
 
  −
=== Create and start a container ===
  −
* Download an OpenVZ template and extract it.
  −
<pre><nowiki>curl http://download.openvz.org/template/precreated/centos-6-x86_64.tar.gz | tar -xz -C test-lxc
  −
</nowiki></pre>
  −
* Create config files
  −
$ cat ~/test-lxc.conf
  −
lxc.console=none
  −
lxc.utsname = test-lxc
  −
lxc.network.type = veth
  −
lxc.network.flags = up
  −
lxc.network.link = br0
  −
lxc.network.name = eth0
  −
lxc.mount = /root/test-lxc/etc/fstab
  −
lxc.rootfs = /root/test-lxc-root/
  −
lxc.console = none
  −
lxc.tty = 0
  −
 
  −
$ cat /root/test-lxc/etc/fstab
  −
none /root/test-lxc-root/dev/pts devpts defaults 0 0
  −
none /root/test-lxc-root/proc    proc  defaults 0 0
  −
none /root/test-lxc-root/sys    sysfs  defaults 0 0
  −
none /root/test-lxc-root/dev/shm tmpfs  defaults 0 0
  −
 
  −
* Register the container
  −
$ lxc-create -n test-lxc -f test-lxc.conf
  −
 
  −
* Start the container
  −
$ mount --bind test-lxc test-lxc-root/
  −
$ lxc-start -n test-lxc
  −
 
  −
== Checkpoint and restore an LXC Container ==
  −
 
  −
=== Preparations ===
  −
 
  −
You not only need to [[Installation | install]] the criu, but also check that the iproute2 utility (<code>ip</code>) is v3.6.0 or higher.
  −
 
  −
You can clone the [http://git.kernel.org/?p=linux/kernel/git/shemminger/iproute2.git;a=summary git repo] or download the [http://kernel.org/pub/linux/utils/net/iproute2/ tarball] to compile it manually. In order to tell to criu where the proper ip tool is set the <code>CR_IP_TOOL</code> environment variable.
  −
 
  −
=== Dump and restore ===
  −
Dumping and restoring an LXC contianer means -- dumping a subtree of processes starting from container init plus all kinds of namespaces.
  −
Restoring is symmetrical. The way LXC container works imposes some more requirements on criu usage.
  −
 
  −
* In order to properly isolate container from unwanted networking communication during checkpoint/restore you should provide a script for locking/unlocking the container network (see below)
  −
* When restoring a container with veth device you may specify a name for the host-side veth device
  −
* In order to checkpoint and restore alive TCP connections you should use the <code>--tcp-established</code> option
  −
 
  −
Typically a container dump command will look like
  −
<pre>
  −
criu dump
  −
    --tcp-established                # allow for TCP connections dump
  −
    --action-script "net-script.sh"  # use net-script.sh to lock/unlock networking
  −
    -D dump/ -o dump.log            # set images dir to dump/ and put logs into dump.log file
  −
    -t ${init-pid}                  # start dumping from task ${init-pid}. It should be container's init
  −
</pre>
  −
and restore command like
  −
<pre>
  −
criu restore
  −
  --tcp-established
  −
  --action-script "net-script.sh"
  −
  --veth-pair eth0=${veth-name}    # when restoring a veth link use ${veth-name} for host-side device end
  −
  --root ${path}                    # path to container root. It should be a root of a (bind)mount
  −
  -D data/ -o restore.log
  −
  -t ${init-pid}
  −
</pre>
  −
 
  −
We also find it useful to use the <code>--restore-detached</code> option for restore to make contianer reparent to init rather than hanging on a criu process launched from shell. Another useful option is the <code>--pidfile</code> one -- you will be able to find out the host-side pid of a container init after restore.
  −
 
  −
Also note, that there's a BUG in how LXC prepares the /dev filesystem for a container which sometimes makes it impossible to dump and container. The <code>--evasive-devices</code> option can help.
  −
 
  −
More details on the option mentioned can be found in [[Usage]] and [[Advanced usage]] pages.
  −
 
  −
=== Example ===
  −
We have [http://git.criu.org/?p=crtools.git;a=tree;f=test/app-emu/lxc;hb=HEAD an application test] for dumping/restoring an LXC Container. You may look at it for better understanding how to dump and restore your container with criu.
  −
 
  −
This test contains two scripts:
  −
;[http://git.criu.org/?p=crtools.git;a=blob;f=test/app-emu/lxc/run.sh;hb=HEAD run.sh]
  −
:This is the main script, which executes ''criu'' two times for dumping and restoring CT. It contains a working commands for dumping and restoring a container.
  −
 
  −
;[http://git.criu.org/?p=crtools.git;a=blob;f=test/app-emu/lxc/network-script.sh;hb=HEAD network-script.sh]
  −
: This one is used to lock and unlock CT's network as described above.
  −
 
  −
===FAQ===
  −
* CRIU supports restricted number of file systems: proc, sysfs, devtmpfs, tmpfs, binfmt_misc. All unsupported file systems must be umounted or handled by plugins.
  −
Error (mount.c:737): FS mnt /sys/fs/pstore dev 0x18 root / unsupported
  −
 
  −
* /dev/console isn't supported yet. You can try to remove the "lxc.cgroup.devices.allow = c 5:1 rwm" from a CT config.
  −
Error (tty.c:203): tty: Can't obtain ptmx index: Inappropriate ioctl for device
  −
 
  −
* Netlink sockets with pending messages are not supported. You can try again later.
  −
Error (sk-netlink.c:77): The socket has data to read
  −
 
  −
* Quite often CRIU can complain about [[external bind mounts]] like this
  −
  ... doesn't have a proper root mount
  −
consider using the <code>--ext-mount-map</code> option
  −
 
  −
== Using LXC Upstream tools ==
      
LXC upstream has begun to integrate checkpoint/restore support through the lxc-checkpoint tool. Although this functionality is not in any released version of LXC yet, you can check out the development version on Ubuntu by doing:
 
LXC upstream has begun to integrate checkpoint/restore support through the lxc-checkpoint tool. Although this functionality is not in any released version of LXC yet, you can check out the development version on Ubuntu by doing:
Line 168: Line 49:  
ssh ubuntu@$(sudo lxc-info -i -H -n u1)
 
ssh ubuntu@$(sudo lxc-info -i -H -n u1)
 
</pre>
 
</pre>
 +
 +
===Troubleshooting===
 +
* <pre>Error (sk-netlink.c:77): The socket has data to read</pre>
 +
Netlink sockets with pending messages are not supported. Usually if you run the dump again it will succeed.
 +
 +
* <pre>Error (mount.c:805): fusectl isn't empty: 8388625.</pre>
 +
Dumping of fuse filesystems is currently not supported. Empty the container's /sys/fs/fuse/connections and try again.
 +
 +
* <pre>Error (mount.c:517): Mount 58 (master_id: 12 shared_id: 0) has unreachable sharing</pre>
 +
criu doesn't yet support shared mountpoints as lxc does; make sure your rootfs is on a non-shared mount.
    
[[Category: HOWTO]]
 
[[Category: HOWTO]]