Diff for "Running/LXC"

Not logged in - Log In / Register
Differences between revisions 46 and 103 (spanning 57 versions)
⇤ ← Revision 46 as of 2011-07-15 05:30:52 →
Size: 4621
Editor: wgrant
Comment: /etc/hosts change is no longer required; rabbitfixture is more robust.
← Revision 103 as of 2014-02-21 15:47:48 → ⇥
Size: 5534
Editor: cjwatson
Comment: LP is on precise now
Deletions are marked like this. Additions are marked like this.
Line 1: Line 1:
## page was copied from Running/VirtualMachine
This page explains how to set up and run Launchpad (for development) inside a LXC.		 This page explains how to set up and run Launchpad (for development) inside an LXC container. LXC is the recommended environment for doing Launchpad development in. We are currently transitioning to using LXC for our Continuous Integration setup.
Line 6: Line 5:
Launchpad development setup makes significant changes to your machine; its nice to be unaffected by those except when you are actually doing such development. Launchpad development setup makes significant changes to your machine; it's nice to be unaffected by those when you're not doing such development. Also, multiple containers can be used to work around Launchpad's limitations regarding concurrent test runs on a single machine.
Line 8: Line 7:
Also, launchpad has some limitations on concurrent testing per-machine and so forth - multiple container's can be used to work around this. These instructions should work on Ubuntu 12.04 LTS and later. Older versions of LXC are significantly less reliable and polished, so if you've used a version of LXC older than 12.04 LTS's final release on your development machine, you'll want to remove `/var/cache/lxc` first to ensure that you don't have a broken cache.
Line 10: Line 9:
= Make a LXC = = Create an LXC container =
Line 12: Line 11:
 1. Install lxc  1. Install LXC's userspace tools.
Line 17: Line 16:
 1. Work around Bug:800456 and Bug:801002  1. Create a container. You might want to use an HTTP proxy or alternate Ubuntu mirror; you can do this by specifying an http_proxy or MIRROR environment variable after `sudo`.
Line 19: Line 18:
sudo apt-get install cgroup-bin libvirt-bin sudo lxc-create -t ubuntu -n lpdev -- -r precise -a i386 -b $USER
Line 22: Line 21:
 1. Work around Bug:784093  1. Start the container. You'll probably see a few early warnings about boot processes dying -- they're normal and can be ignored as long as you end up at a login prompt.
Line 24: Line 23:
sudo dd of=/etc/cgconfig.conf << EOF
mount {
 cpu = /sys/fs//cgroup/cpu;
 cpuacct = /sys/fs/cgroup/cpu;
 devices = /sys/fs/cgroup/cpu;
 memory = /sys/fs/cgroup/cpu;
}
EOF
sudo service cgconfig restart		 sudo lxc-start -n lpdev
}}}
    
 1. '''[Inside the container]''' Log in with your normal username and password. You'll have full sudo powers.

 1. '''[Inside the container]''' Install various packages needed to be able to connect easily (avahi-daemon) and run `rocketfuel-setup` [[#postgresql-locale-breakage|successfully]].
 {{{
 sudo apt-get install avahi-daemon bzr language-pack-en
}}}

 1. '''[Inside the container]''' Shut down by running `sudo poweroff` inside the container, and you should eventually be dumped back out to your host system. If it looks like it's hanging, force it to stop with `sudo lxc-stop -n lpdev` from outside the container.
 
 1. Start it up again, headless this time (`-d`). The same IP address will be used, so you don't need console access.
 {{{
sudo lxc-start -n lpdev -d
Line 35: Line 40:
 1. Work around Bug:798476 (optional if you run i386 or have a -tonne- of memory and don't care about 64-bit footprint.
    Grab the patch from the bug and apply it to /usr/lib/lxc/templates/lxc-lucid. If you're running i386 already or want a 64-bit lxc then do not pass arch= on the lxc-create command line.		  1. `ssh -A lpdev.local` to connect to the container. If your SSH key is in your local `authorized_keys` file you shouldn't be prompted for a password, as your home directory (including public and private keys) is bind mounted into the container. The `-A` permits you to access Launchpad code hosting from within the container without needing to reenter passphrases.
Line 38: Line 42:
 1. Create a config for your containers
 {{{
sudo dd of=/etc/lxc/local.conf << EOF
lxc.network.type=veth
lxc.network.link=virbr0
lxc.network.flags=up
#fuse (workaround for Bug:800886)
lxc.cgroup.devices.allow = c 10:229 rwm
# part of the Bug:798476 workaround -
# remove if you are running a 64 bit lxc or
# 32 bit on 32-bit base os
lxc.arch = i686
EOF		  1. '''[Inside the container]''' You can now follow the normal [[Getting|LP installation instructions]]. Be warned that changes in your home directory will also be seen outside the container and vice versa. If your home directory already has a Launchpad work area set up you'll want to run `rocketfuel-setup --no-workspace` to avoid trying to recreate it, but all subsequent steps are still required.

 1. Follow [[Running/RemoteAccess]] to set up access from the host's applications to the container's Launchpad instance.


= Troubleshooting =

<<Anchor(postgresql-locale-breakage)>>
== launchpad-database-setup fails ==

PostgreSQL will fail to create a cluster during installation if your
locale is configured to something non-C but not supported by the
container, so you need to install the relevant language pack.

You will know you need to do this if bzr or apt commands have been
spewing locale warnings.

For instance, if your computer has a localised English locale, use this:

{{{
apt-get install language-pack-en
Line 53: Line 65:
 1. Create a container If you didn't install the language pack before running rocketfuel-setup,
you'll need to run `sudo pg_createcluster 8.4 main` afterwards to fix
the damage.

== rabbitmq does not start up ==

rabbitmq may fail to start up. If that happens it appears to be a [[http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/2010-April/007024.html|mnesia glitch]] best sorted by zapping mnesia.
Line 55: Line 73:
sudo arch=i386 lxc-create -n lucid-test-lp -t lucid -f /etc/lxc/local.conf
}}}
    If you want to use a proxy
 {{{
sudo arch=i386 http_proxy=http://host:port/ lxc-create -n lucid-test-lp -t lucid -f /etc/lxc/local.conf
}}}
    And if you want to set a custom mirror, similar to http_proxy, but set MIRROR= instead.

 1. (Outside the container) grab your user id and username so you can setup a bind mount outside the container:
 {{{
id -u
id -nu
}}}

 1. Start the container
 {{{
sudo lxc-start -n lucid-test-lp
}}}
    Ignore the warning about openssh crashing - it restarts on a later event.
    The initial credentials are root:root.

 1. Grab the ip address (handed out via libvirt's dhcp server) - you may wish to ssh in rather than using the console (seems to have better termcap experience).
 {{{
ip addr show dev eth0 | grep 'inet'
}}}

 1. The new container won't have your proxy / mirror settings preserved. Customise it at this point before going further if you care about this.

 1. Enable multiverse (rocketfuel-setup wants it, don't ask me why).

 1. Install some additional packages we'll need to run rocketfuel-setup etc.
 {{{
apt-get install python-software-properties
apt-add-repository ppa:ubuntu-virt
apt-get update
apt-get install bzr less sudo lxcguest
# select I for 'install' when prompted about console.conf
}}}

 1. Inside the container add the user:
 {{{
adduser --uid $id $username
adduser $username sudo
}}}

 1. To stop it now run 'poweroff -n' (normally you would use lxc-stop, but see Bug:784093).

 1. Setup a bind mount so you can access your home dir (and thus your LP source code) from within the lxc container:
    * edit /var/lib/lxc/lucid-test-lp/fstab
    * Add a line:
 {{{
/home/$username /var/lib/lxc/lucid-test-lp/rootfs/home/$username none bind 0 0
}}}
 
 1. Start it up again - headless now, we have the ip address from before.
 {{{
sudo lxc-start -n lucid-test-lp -d
}}}

 1. ssh <vm IP address> to connect to the VM. Your ssh key is already present because of the bind mount to your home dir.

 1. You can now follow the [[Getting|getting-started]] on LP instructions. Be warned that changes in ~ will affect you outside the container. You will want to run rocketfuel-setup with --no-workspace if your home already has a workarea. You may need to run utilities/launchpad-database-setup separately too.

 1. You probably want to follow [[Running/RemoteAccess]] has a discussion for how you can configure things so your non-container browser can access web pages from within the container.

 1. rabbitmq may fail to start up. If that happens it appears to be a [[http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/2010-April/007024.html|mnesia glitch]] best sorted by zapping mnesia.
{{{
Line 125: Line 76:

== lxc-start hangs ==

[[http://paste.ubuntu.com/772517/|The symptom looks like this]]. It hangs after that.

No fix or workaround identified yet, other than making a new lxc container.

To debug, try '''{{{lxc-start -n $containername -l debug -o outout}}}''' and look at output.

== DNS fails inside the container ==

After restarting in daemon mode and logging in as a regular user, DNS was not working.
Ensure there is a nameserver in the container's /etc/resolv.conf, which is created at startup by resolverconf. Stopping and starting the container solved the problem.

== Random flakiness ==

Using lxc via juju I ran into all sorts of problems with DNS, version mismatches, etc. Since it was via juju I wasn't able to muck around with /etc/resolv.conf (the damage was done before I got the chance to ssh to the guest.) I found {{{sudo rm -rf /var/cache/lxc}}} solved the problem. It is rather brutal but worked. Of course the next run took a long time as all of that previously cached stuff had to be refetched.

== Other problems ==

If other lxc users don't have an idea (known lxc users as of this writing include lifeless, wgrant, frankban and gary_poster) try asking hallyn or Spamaps on #ubuntu-server on freenode.

This page explains how to set up and run Launchpad (for development) inside an LXC container. LXC is the recommended environment for doing Launchpad development in. We are currently transitioning to using LXC for our Continuous Integration setup.

Why?

Launchpad development setup makes significant changes to your machine; it's nice to be unaffected by those when you're not doing such development. Also, multiple containers can be used to work around Launchpad's limitations regarding concurrent test runs on a single machine.

These instructions should work on Ubuntu 12.04 LTS and later. Older versions of LXC are significantly less reliable and polished, so if you've used a version of LXC older than 12.04 LTS's final release on your development machine, you'll want to remove /var/cache/lxc first to ensure that you don't have a broken cache.

Create an LXC container

  1. Install LXC's userspace tools. 
    sudo apt-get install lxc

  2. Create a container. You might want to use an HTTP proxy or alternate Ubuntu mirror; you can do this by specifying an http_proxy or MIRROR environment variable after sudo. 

    sudo lxc-create -t ubuntu -n lpdev -- -r precise -a i386 -b $USER
  3. Start the container. You'll probably see a few early warnings about boot processes dying -- they're normal and can be ignored as long as you end up at a login prompt. 
    sudo lxc-start -n lpdev

  4. [Inside the container] Log in with your normal username and password. You'll have full sudo powers.

  5. [Inside the container] Install various packages needed to be able to connect easily (avahi-daemon) and run rocketfuel-setup successfully. 

     sudo apt-get install avahi-daemon bzr language-pack-en

  6. [Inside the container] Shut down by running sudo poweroff inside the container, and you should eventually be dumped back out to your host system. If it looks like it's hanging, force it to stop with sudo lxc-stop -n lpdev from outside the container.

  7. Start it up again, headless this time (-d). The same IP address will be used, so you don't need console access. 

    sudo lxc-start -n lpdev -d

  8. ssh -A lpdev.local to connect to the container. If your SSH key is in your local authorized_keys file you shouldn't be prompted for a password, as your home directory (including public and private keys) is bind mounted into the container. The -A permits you to access Launchpad code hosting from within the container without needing to reenter passphrases.

  9. [Inside the container] You can now follow the normal LP installation instructions. Be warned that changes in your home directory will also be seen outside the container and vice versa. If your home directory already has a Launchpad work area set up you'll want to run rocketfuel-setup --no-workspace to avoid trying to recreate it, but all subsequent steps are still required.

  10. Follow Running/RemoteAccess to set up access from the host's applications to the container's Launchpad instance.

Troubleshooting

launchpad-database-setup fails

PostgreSQL will fail to create a cluster during installation if your locale is configured to something non-C but not supported by the container, so you need to install the relevant language pack.

You will know you need to do this if bzr or apt commands have been spewing locale warnings.

For instance, if your computer has a localised English locale, use this: 

apt-get install language-pack-en

If you didn't install the language pack before running rocketfuel-setup, you'll need to run sudo pg_createcluster 8.4 main afterwards to fix the damage.

rabbitmq does not start up

rabbitmq may fail to start up. If that happens it appears to be a mnesia glitch best sorted by zapping mnesia. 

  • sudo rm -rf /var/lib/rabbitmq/mnesia/rabbit/*
sudo service rabbit-mq start

lxc-start hangs

The symptom looks like this. It hangs after that.

No fix or workaround identified yet, other than making a new lxc container.

To debug, try lxc-start -n $containername -l debug -o outout and look at output.

DNS fails inside the container

After restarting in daemon mode and logging in as a regular user, DNS was not working. Ensure there is a nameserver in the container's /etc/resolv.conf, which is created at startup by resolverconf. Stopping and starting the container solved the problem.

Random flakiness

Using lxc via juju I ran into all sorts of problems with DNS, version mismatches, etc. Since it was via juju I wasn't able to muck around with /etc/resolv.conf (the damage was done before I got the chance to ssh to the guest.) I found sudo rm -rf /var/cache/lxc solved the problem. It is rather brutal but worked. Of course the next run took a long time as all of that previously cached stuff had to be refetched.

Other problems

If other lxc users don't have an idea (known lxc users as of this writing include lifeless, wgrant, frankban and gary_poster) try asking hallyn or Spamaps on #ubuntu-server on freenode.

References

Alternatively

You can also run in a chroot environment or a VM.