High availability with LVS using LVSadmin

The Linux Virtual Server is a highly scalable and highly available server built on a cluster of real servers, with the load balancer running on the Linux operating system. The architecture of the server cluster is fully transparent to end users, and the users interact as if it were a single high-performance virtual server.

We use LVS extensively at work to provide a scalable and highly available website which gets around 300 hits per second. Setting up and managing LVS can be made a lot easier using a tool that our ex staff wrote called LVSadmin. Written in perl it is easily configurable and provides a curses based front end to manage the servers. Setting up a new LVS cluster is really easy.

For our new cluster we have 2 servers that I want to load balance with LVS:


And I want them presented with the following hostname:


We want 2 LVS instances, for redundancy in case one dies which will run on the following servers:


The target platform is Ubuntu Dapper, which is our platform of choice at the moment until Hardy is out!

On lvs0/lvs1 Grab the source code (lvsadmin, LVS.pm) for LVS from SourceForge and place it in /usr/local/bin, lvsadmin should be +x.

Then install the following packages

apt-get install perl-modules libcurses-perl libcurses-widgets-perl keepalived

A few variable changes need to be done in the LVS.pm:

* Change the $MASTER to the hostname of the master server, in our case lvs0
* Change $IF to the interface that packets will be coming from, in our case eth0
* There is a br0 further down the script that needs to reflect the $IF change so again change that to eth0
* Change $PASSWORD to the keepalived password you want
* Find the lvs_id and change that to a new unique instance for this LVS cluster.

LVS.pm on both servers should be identical.

The following files need to then be made in /etc/keepalived

portlist (this is a list of all the realserver ports LVS will manage)


serverlist (these hostnames are resolved to create a meaningful display)


serverstate (the default state of the servers, lvsadmin will read and write it’s state to this file when you change things)

services (list of virtual server ports)


viplist (list of virtual server IPs) eth0

Then on each of the real servers we need to create the virtual IP for them to listen on in /etc/network/interfaces add:

auto lo:23
iface lo:23 inet static
pre-up echo 1 >/proc/sys/net/ipv4/conf/all/arp_ignore; echo 2 >/proc/sys/net/ipv4/conf/all/arp_announce

then on each real server start the interface:

[email protected]:~# ifup lo:23

That’s all the configuration done, we just now have to start the LVS system, the first time is a little flaky but from them on in it will work smoothly.

On each lvs server start lvsadmin, then press go to info and press Shift-S to save, it will then create /etc/keepalived/keepalived.conf

! Configuration File for keepalived

global_defs {
lvs_id LVS_XEN

vrrp_instance VI_1 {
state MASTER
interface eth0
virtual_router_id 51
priority 150
advert_int 1
authentication {
auth_type PASS
auth_pass eggsandham
virtual_ipaddress { dev eth0
# Virtualserver:
virtual_server 8889 {
delay_loop 60
lb_algo wrr
lb_kind DR
protocol TCP
virtualhost www.digitalbrain.com
# Realserver:
real_server 8889 {
weight 30
connect_timeout 3
nb_get_retry 3
delay_before_retry 3
# Realserver:
real_server 8889 {
weight 30
connect_timeout 3
nb_get_retry 3
delay_before_retry 3

Then start keepalived

/etc/init.d/keepalived start

You should now be able to telnet to the correct port on the virtual server if it’s working!

[email protected]:~$ telnet lvs-dev-blobdirector 8889
Connected to lvs-dev-blobdirector.btn.dbplc.com.
Escape character is '^]'.

To test LVS redundancy take down the master (lvs0) and see if you can still connect to the virtual server.

[email protected]:~# /etc/init.d/keepalived stop
Stopping keepalived: keepalived.
[email protected]:~# ps aux | grep keep
root 3859 0.0 0.1 3940 900 pts/1 R+ 14:04 0:00 grep keep

[email protected]:~$ telnet lvs-dev-blobdirector 8889
Connected to lvs-dev-blobdirector.btn.dbplc.com.
Escape character is '^]'.

And there we have it, an easy way to create a scalable, highly available server platform!

SVN COPY 502 Bad Gateway error

Our developers were experiencing a weird problem recently with our SVN installation where they couldn’t copy any files in SVN, they would always get the following error

svn: COPY of /project/!svn/bc/5121/trunk/path/file.gif: 502 Bad Gateway (https://svn)

A quick fix of course would have been to just create a new file and copy the contents but this wouldn’t have kept the file history. A quick google lead to this page and a solution!

Our specific problem was we had tried to be lean and set up a default https config that all our SSL sites used which specified the SSL parameters in, but we hadn’t explicitly enabled SSL in the svn vhost, so although SSL was working fine, Apache and mod_ssl actually thought the request was coming through on port 80, and thus http, instead of port 443, https, so the request was being translated to technically copy a file from one svn repository to a completely different one.

The solution was to put the SSL engine and cerficiate options back in to the svn vhost so Apache would pick up that the connect was indeed https and not http!

Version Control with Subversion

For more SVN advice, I recommend¬†Version Control with Subversion by O’Reilly. It contains everything you need to know when using or managing SVN repositories.

New Years Resolutions

A lot of people hate the idea of new year resolutions, but if you want to make some changes and the fact it’s the start of a year will give you motivation, then so be it! Lets see how many I keep in 2009!

* Give up smoking
* Give up caffeine
* Stick to diet and gym
* Less red meat and fatty meats
* No takeaway pizza, McDonalds or Burger King
* Attend martial arts classes more regularly
* Join a yoga/pilates class to improve posture/flexibility
* Visit another contenant

Last year my resolution was to give up McDonalds, I only wavered once when stranded at Liverpool St. Station!

Resize Xen Filesystem

We run a lot of Xen instances for our development and test servers and a few were starting to get full. Fortunately the disks in the real servers were very large and the xenlet partitions were made using LVM so resizing them to add more space was possible!

[email protected]:~# df -h
Filesystem Size Used Avail Use% Mounted on
/dev/hda1 4.0G 3.8G 200M 95% /
varrun 257M 48K 257M 1% /var/run
varlock 257M 0 257M 0% /var/lock
udev 257M 40K 257M 1% /dev
devshm 257M 0 257M 0% /dev/shm

Basically we just have to shut down the xenlet, resize the partition and then restart the xenlet again, simple!

[email protected]:~# xm shutdown dev-myfiles0
[email protected]:~# lvextend -L40G /dev/vg0/dev-myfiles0-disk
Extending logical volume dev-myfiles0-disk to 40.00 GB
Logical volume dev-myfiles0-disk successfully resized
[email protected]:~# e2fsck -f /dev/vg0/dev-myfiles0-disk
e2fsck 1.40.2 (12-Jul-2007)
Pass 1: Checking inodes, blocks, and sizes
Pass 2: Checking directory structure
Pass 3: Checking directory connectivity
Pass 4: Checking reference counts
Pass 5: Checking group summary information
/dev/vg0/dev-myfiles0-disk: 16541/524288 files (0.9% non-contiguous), 138346/1048576 blocks
[email protected]:~# resize2fs /dev/vg0/dev-myfiles0-disk
resize2fs 1.40.2 (12-Jul-2007)
Resizing the filesystem on /dev/vg0/dev-myfiles0-disk to 10485760 (4k) blocks.
The filesystem on /dev/vg0/dev-myfiles0-disk is now 10485760 blocks long.
[email protected]:~# cd /etc/xen
[email protected]:/etc/xen# xm create dev-myfiles0.cfg
Using config file "./dev-myfiles0.cfg".
Started domain dev-myfiles0

Wee, lots of free space now!

[email protected]:~# df -h
Filesystem Size Used Avail Use% Mounted on
/dev/hda1 40G 3.8G 37G 10% /
varrun 257M 40K 257M 1% /var/run
varlock 257M 0 257M 0% /var/lock
udev 257M 40K 257M 1% /dev
devshm 257M 0 257M 0% /dev/shm