Copyright © Alexandre Cassen – Linux Virtual Server OpenSource project 1
Keepalived for LVS
User Guide
Alexandre Cassen
http://www.keepalived.org � acassen@linux-vs.org
Copyright © Alexandre Cassen – Linux Virtual Server OpenSource p
LicenceLicenceLicenceLicence
This document is copyright 2001, 2002 Alexandre Cassen. It is released
under the terms of the GNU General Public Licence. You can redistribute
it and/or modify it under the terms of the GNU General Public Licence as
published by the Free Software Foundation; either version 2 of the
Licence, or (at your option) any later version.
roject 2
Copyright © Alexandre Cassen – Linux Virtual Server OpenSource project 3
CONTENTSCONTENTSCONTENTSCONTENTS
CONTENTS ..............................................................................................................................3
I. INTRODUCTION.................................................................................................................4
II. TERMINOLOGY................................................................................................................4
III. SOFTWARE ARCHITECTURE.....................................................................................5
3.1 GLOBAL VIEW AND LOCALIZATION .....................................................................................5
3.2 SOFTWARE DESIGN ............................................................................................................6
IV. HEALTHCHECK FRAMEWORK .................................................................................7
V. FAILOVER FRAMEWORK : VRRP FRAMEWORK ..................................................7
VI. INSTALLING KEEPALIVED .........................................................................................8
VII. KEEPALIVED CONFIGURATION SYNOPSIS .........................................................9
7.1 GLOBAL DEFINITIONS SYNOPSIS ........................................................................................9
7.2 VIRTUAL SERVER DEFINITIONS SYNOPSIS...........................................................................9
7.3 VRRP INSTANCE DEFINITIONS SYNOPSIS ........................................................................11
VIII. KEEPALIVED PROGRAMS SYNOPSIS .................................................................12
8.1 KEEPALIVED DAEMON.......................................................................................................12
8.2 GENHASH UTILITY ............................................................................................................12
8.3 RUNING KEEPALIVED DAEMON .........................................................................................12
IX. CASE STUDY : HEALTHCHECK ...............................................................................14
9.1 MAIN ARCHITECTURE COMPONENTS.................................................................................14
9.2 SERVER POOL SPECIFICATIONS ........................................................................................14
9.3 KEEPALIVED CONFIGURATION..........................................................................................15
X. CASE STUDY : FAILOVER USING VRRP..................................................................18
10.1 ARCHITECTURE SPECIFICATION .....................................................................................19
10.2 KEEPALIVED CONFIGURATION........................................................................................19
XI. CASE STUDY : MIXING HEALTHCHECK & FAILOVER ....................................21
11.1 KEEPALIVED CONFIGURATION........................................................................................21
Cop
I. IntroductionI. IntroductionI. IntroductionI. Introduction
Load balancing is a good solution for service virtualization. When you design a load balanced
topology one must take special care of:
• Real server availability using health-checks.
• Load balancer availability using failover protocol.
Load balancing real services, provides a global Highly Available virtual service. To increase the
load balanced service availability we need to monitor each real server node. This problem is mainly
handled using a health-check framework manipulating a real server pool.
On the other hand, when using a load balancer director we introduce a Single Point Of Failure
for the virtual service. So load balancer high availability must also be handled, using dedicated routing
protocols for director failover/virtualization.
Keepalived tries to address these two problems by adding, on the one hand, a strong & robust
health-check framework, and on the other hand, implementing a Hot Standby protocol. These two
frameworks can deal with the Linux Virtual Server (LVS) framework to manipulate LVS real server
pools by adding or removing real servers based on health-checks’ decisions.
II. TerminologyII. TerminologyII. TerminologyII. Terminology
In
LV
WAN
LVS Router
WAN Interface
LVS stands for “Linux Virtual Server“. LVS is a
patched Linux kernel that adds a load balancing
facility. For more information on LVS, please refer to
the project homepage: http://www.linux-vs.org. LVS
acts as a network bridge (using NAT) to load balance
TCP/UDP stream. The LVS router components are:
• WAN Interface: Ethernet Network Interface
Controller that will be accessed by all the
yright © Alexandre Cassen – Linux Virtual Server OpenSource project 4
this document, we will use the following keywords:
S component:
• VIP: The Virtual IP is the IP address that will be accessed by all the clients. The clients only
access this IP address.
• Real server: A real server hosts the application accessed by client requests. WEB SERVER 1
& WEB SERVER 2 in our synopsis.
• Server pool: A farm of real servers.
Hub / SWITCH
WEB SERVER 1 WEB SERVER 2
LAN Interface clients.
• LAN Interface: Ethernet Network Interface
Controller to manage all the load balanced
servers.
• Linux kernel: The kernel is patched with the
latest LVS and is used as a router OS.
• Virtual server: The access point to a Server pool.
• Virtual Service: A TCP/UDP service associated with the VIP.
VRRP component:
• VRRP: The protocol implemented for the directors’ failover/virtualization.
• VRRP Instance: A thread manipulating VRRPv2 specific set of IP addresses. A VRRP
Instance may backup one or more VRRP Instance. In our “Case study: Failover”, we are
dealing with 4 VRRP Instances. One owning (VIP1,VIP2), one owning (VIP3,VIP4), one
owning (DIP1) and one owning (DIP2). It may participate in one or more virtual routers.
• IP Address owner: The VRRP Instance that has the IP address(es) as real interface
address(es). This is the VRRP Instance that, when up, will respond to packets addressed to
one of these IP address(es) for ICMP, TCP connections, …
• MASTER state: VRRP Instance state when it is assuming the responsibility of forwarding
packets sent to the IP address(es) associated with the VRRP Instance. This state is illustrated
on “Case study: Failover” by a red line.
• BACKUP state: VRRP Instance state when it is capable of forwarding packets in the event
Copyright © Alexandre Cassen – Linux Virtual Server OpenSource project 5
that the current VRRP Instance MASTER fails.
• Real Load balancer: An LVS director running one or many VRRP Instances.
• Virtual Load balancer: A set of Real Load balancers.
• Synchronized Instance: VRRP Instance with which we want to be synchronized. This
provides VRRP Instance monitoring.
• Advertisement: The name of a simple VRRPv2 packet sent to a set of VRRP Instances while
in the MASTER state.
III. Software ArchitectureIII. Software ArchitectureIII. Software ArchitectureIII. Software Architecture
3.1 Global View and localization
Pool de serveurs
WAN
Hub
LoadBalancer
Linux Virtual Server
Server 1 Server 2 Server 3 Server 4
Kernel Space
User Space
Routing
Decision
FORWARD
RULES
Incomnig IP Packet
ProcessINPUTRULES
OUTPUT
RULES
VS Rules
Table
VS Connection
Hash Table
setsockopt
proc filesystem
Firewall FrameWork
VS Schedule
Controle Module
Outgoing IP Packet
VIP
setsockopt
proc filesystem
NETLINK Multicast
HealthCheck Framework VRRP Framework
Cop
Our software architecture deals with 4 Linux kernel components:
• LVS kernel framework: Using the setsockopt call for kernel 2.2 and the setsockopt netfilter call
for kernel 2.4.
• IPCHAINS framework: For kernel 2.2 in LVS NAT architecture we use an internal IPCHAINS
wrapper to send MASQ chains to the kernel. This is only used when running Linux 2.2 kernel.
On kernel 2.4 IPVS code handles specific NAT rules using the netfilter call.
• NETLINK Interface: For the Hot Standby protocol (VRRP), we use the NETLINK interface to
set/remove VRRP VIP.
• MULTICAST: For the VRRP part, advertisements are sent to a MULTICAST group.
3.2 Software design
The following figure illustrates the Keepalived internal software implementation components.
Keepalived uses a fully multithreaded framework based on a central I/O multiplexer. The 2 main
components are:
• Health-checker worker threads: Each health-check is registered to our global scheduling
framework. These workers perform health-checks using on the Keepalived health-check
Th
VRRP Bootstrap
Socket Pool thread
Global scheduling framework
I/O multiplexer
VRRP Packet Dispatcher
thread
VRRP state handler
Kernel Space
User Space
VRRP Instance
VI_1
VRRP Instance
VI_2
VRRP Instance
VI_n
LOW LEVEL PRIMITIVES
Netlink Multicast SIOCGIF
SMTP
Notification
IPVS framework
Multilayer HealthCheck Framework
MISC CHECKER External
Forked process call
TCP CONNECT
THREAD
HTTP GET send
SSL GET send
MD5SUM over
HTML content
THREAD
Layer 4 Layer 5/6/7
DAEMON INITIALIZATION
HealthCheckers Workers
thread registration
yright © Alexandre Cassen – Linux Virtual Server OpenSource project 6
framework. The health-check frameworks currently implements 3 checkers :
o TCP CHECK: Performing a LAYER3 check.
o HTTP GET: Checking a remote HTTP server html content integrity.
o SSL GET: Checking a remote SSL server html content integrity.
o MISC CHECK: Performing user defined integrity checks.
• VRRP Packet Dispatcher: Demultiplexing specific I/O to handle VRRP Instance
corresponding.
ese 2 main components use the following low-level primitives:
• SMTP notification: An SMTP wrapper using asynchronous stream process. This primitive
enables Keepalived to send email notifications.
Copyright © Alexandre Cassen – Linux Virtual Server OpenSource project 7
• IPVS framework: The LVS kernel interface for real server pool manipulation. All the IPVS
load balancing methods are implemented eg: LVS NAT, DR & TUN.
• Netlink: Kernel routing interface for the VRRP part. Provides VRRP VIP manipulation.
• Multicast: For sending VRRP adverts we use multicast (need to handle multicast binding to
specific interface, …)
• IPCHAINS framework: This is only used if running Linux kernel 2.2 to automatically set
MASQ chains. This part is obsolete for newer kernel since LVS natively deals with MASQ
rules as it is a NETFILTER module. Only implemented for compatibility purpose.
• SYSLOG: All daemon notification messages are logged using the syslog daemon.
IV. Healthcheck frameworkIV. Healthcheck frameworkIV. Healthcheck frameworkIV. Healthcheck framework
As described in the global Keepalived software design, the current multilayer health-check framework
implements the following checker modules:
• TCP_CHECK: Working at layer4. To ensure this check, we use a TCP Vanilla check using
nonblocking/timed-out TCP connections. If the remote server does not reply to this request
(timed-out), then the test is wrong and the server is removed from the server pool.
• HTTP_GET: Working at layer5. Performs a GET HTTP to a specified URL. The get result is
then summed using the MD5 algorithm. If this sum does not match with the expected value,
the test is wrong and the server is removed from the server pool. This module implements a
multi-URL get check on the same service. This functionality is useful if you are using a server
hosting more than one application server. This functionality gives you the ability to check if an
application server is working properly. The MD5 digests are generated using the genhash
utility (included in the keepalived package).
• SSL_GET: Same as HTTP_GET but uses a SSL connection to the remote webservers.
• MISC_CHECK: This check allows a user defined script to be run as the health checker. The
result must be 0 or 1. The script is run on the director box and this is an ideal way to test in-
house applications. Scripts that can be run without arguments can be called using the full
path (i.e. /path_to_script/script.sh). Those requiring arguments need to be enclosed in
double quotes (i.e. “/path_to_script/script.sh arg1 … argn”)
The goal for Keepalived is to define a generic framework easily extensible for adding new checkers
modules. If you are interested into checkers development, you can read the HackingGuide which
exposes the internal software implementation and style guide used (this guide is currently under
writing process).
V. Failover framework: VRRP frameworkV. Failover framework: VRRP frameworkV. Failover framework: VRRP frameworkV. Failover framework: VRRP framework
For director failover Keepalived implements the VRRP protocol. To quickly describe this protocol :
“VRRP specifies an election protocol that dynamically assigns responsibility for a virtual router
to one of the VRRP routers on a LAN. The VRRP router controlling the IP address(es)
associated with a virtual router is called the Master, and forwards packets sent to these IP
addresses. The election process provides dynamic fail over in the forwarding responsibility
should the Master become unavailable. This allows any of the virtual router IP addresses on
the LAN to be used as the default first hop router by end-hosts. The advantage gained from
using VRRP is a higher availability default path without requiring configuration of dynamic
routing or router discovery protocols on every end-host.” [rfc2338].
Copyright © Alexandre Cassen – Linux Virtual Server OpenSource project 8
NB: This framework is LVS independent, so you can use it for LVS director failover, even for other
Linux routers needing a Hot-Standby protocol. This framework has been completely integrated in the
Keepalived daemon for design & robustness reasons.
The main functionalities provided by this framework are:
• Failover: The native VRRP protocol purpose, based on a roaming set of VRRP VIPs.
• VRRP Instance synchronization: We can specify a state monitoring between 2 VRRP
Instances. It guarantees that 2 VRRP Instances remain in the same state. The synchronized
instances monitor each other.
• Nice Fallback
• Advert Packet integrity: Using IPSEC-AH ICV.
• System call: During a VRRP state transition, an external script/program can be called.
VIVIVIVI. Installing Keepalived. Installing Keepalived. Installing Keepalived. Installing Keepalived
Before installing Keepalived, any previously installed version should be removed.
1. Download the latest Keepalived source code from the web site and unzip/untar it. In order
to compile Keepalived you need to have the following libraries installed:
• OpenSSL,
: This library is needed for MD5 and SSL
support.
• popt, : Used for command line
parsing.
You will also need the Linux kernel source with the ipvs patches if you intend to use
Keepalived with LVS.
2. Then simply compile the daemon and the genhash utility.
3. All the binary and template configuration file are installed. You may need to create a call to
the keepalived daemon in your rc file. If you are using RedHat Linux, an example initialization
setup would be:
If you are running Debian Linux this would be:
Note: The link should be added in your default run level directory.
[root@lvs keepalived]# ./configure
[root@lvs keepalived]# make
[root@lvs keepalived]# make install
[root@lvs keepalived]# ln -s /etc/rc.d/init.d/keepalived.init /etc/rc.d/rc3.d/S99keepalived
[root@lvs keepalived]# ln -s /etc/init.d/keepalived.init /etc/rc2.d/S99keepalived
Copyright © Alexandre Cassen – Linux Virtual Server OpenSource project 9
VII. Keepalived configuration synopsisVII. Keepalived configuration synopsisVII. Keepalived configuration synopsisVII. Keepalived configuration synopsis
The Keepalived configuration file uses the following synopsis (configuration keywords are
Bold/Italic):
7.1 Global definitions synopsis
Keyword Definition Type
global_defs identify the global def configuration block
notification_email email accounts that will receive the notification mail List
notification_email_from email to use when processing “MAIL FROM:” SMTP command List
smtp_server remote SMTP server to use for sending mail notifications alphanum
smtp_connection_timeout specify a timeout for SMTP stream processing numerical
lvs_id specify the name of the LVS director alphanum
Email type: Is a string using charset as specified into the SMTP RFC eg: “user@domain.com”
7.2 Virtual server definitions synopsis
global_defs {
notification_email {
email
email
}
notification_email_from email
smtp_server host
smtp_connect_timeout num
lvs_id string
}
virtual_server (@IP PORT)|(fwmark num) {
delay_loop num
lb_algo rr|wrr|lc|wlc|sh|dh|lblc
lb_kind NAT|DR|TUN
(nat_mask @IP)
persistence_timeout num
persistence_granularity @IP
virtualhost string
protocol TCP|UDP
sorry_server @IP PORT
real_server @IP PORT {
weight num
TCP_CHECK {
connect_port num
connect_timeout num
}
}
real_server @IP PORT {
weight num
MISC_CHECK {
misc_path /path_to_script/script.sh
(or misc_path “/path_to_script/script.sh ”)
}
}
}
real_server @IP PORT {
weight num
HTTP_GET|SSL_GET {
url { # You can add multiple url block
path alphanum
digest alphanum
}
connect_port num
connect_timeout num
nb_get_retry num
delay_before_retry num
}
}
}
Copyright © Alexandre Cassen – Linux Virtual Server OpenSource project 10
Keyword Definition Type
virtual_server identify a virtual server definition block
fwmark specify that virtual server is a FWMARK
delay_loop specify in seconds the interval between checks numerical
lb_algo select a specific scheduler (rr|wrr|lc|wlc…) string
lb_kind select a specific forwarding method (NAT|DR|TUN) string
persistence_timeout specify a timeout value for persistent connections numerical
persistence_granularity specify a granularity mask for persistent connections
Virtualhost specify a HTTP virtualhost to use for HTTP|SSL_GET alphanum
protocol specify the protocol kind (TCP|UDP) numerical
sorry_server server to be added to the pool if all real servers are down
real_server specify a real server member
Weight specify the real server weight for load balancing decisions numerical
TCP_CHECK check real server availability using TCP connect
MISC_CHECK check real server availability using user defined scr