This page describes how to set up HAProxy as a load balancer and transparent proxy for an on-premise Fredhopper installation.

HAProxy is only available for Unix and Unix-like systems. Please contact Fredhopper for options to configure load-balancing on Windows.

Introduction

Load-balancing FAS is required to ensure that the frontend can always access a FAS query server even when a reindex operation is currently running. Load-balancing FAS requires at least two "live instances" in addition to the indexer.

Fredhopper recommends to use multiple servers for the FAS live instances itself to avoid a hardware failure bringing down the complete FAS service. Similarly the load balancer setup can be duplicated for additional redundancy.

Step-by-Step Instructions

Step 1: Install HAProxy

HAProxy is available in the package repositories of the common Unix versions and Linux distributions, but can also be installed manually from the source archives.

On a RedHat-based system (RHEL, CentOS, Fedora) use yum install haproxy, on Debian-based systems (including Ubuntu) apt-get install haproxy should work.

Step 2: Create Configuration

The HAProxy configuration file consists of usually three sections: general settings, global defaults, and a section for every environment that should be available via the load balancer. A default template is available in the Fredhopper installation directory as config/fredhopper-haproxy.default.cfg.

Start by copying this example file to config/fredhopper-haproxy.cfg, and open it for editing in a text editor.

Global and Default Settings

The common configuration and defaults could look like this:

global
	# log 127.0.0.1	local0 local1 local5 notice info 
	# maxconn <num>
	# nbproc 1
	user fredhopper
	group fredhopper
	pidfile /home/fredhopper/fredhopper/tmp/fredhopper-haproxy.pid
	log 127.0.0.1 user info
	# stats socket /home/fredhopper/fredhopper/tmp/fredhopper-haproxy.socket user fredhopper group fredhopper level admin

defaults
	log	global
	mode	http
	option	httplog
	option	dontlognull
	option	allbackups
	option	redispatch
	option  log-separate-errors
	# Options used with 1.5+ to improve performance
	# option  tcp-smart-accept
	# option  tcp-smart-connect
	# option  splice-auto
	option  abortonclose
	retries	3
	maxconn	2000
	contimeout	5000
	clitimeout	60000
	srvtimeout	60000
	grace 100

The default configuration assumes that Fredhopper is installed in /home/fredhopper/fredhopper, and that the Fredhopper user is called 'fredhopper'.

Environment-specific settings

The environment-specific settings are derived from the config/topology.txt, and link all live instances of a particular environment to be available using haproxy. In this example the config/topology.txt file looks like this:

# See https://www.fredhopper.com/learningcenter/x/BAobAQ for more
# information about the contents of this file.
#
# instance name | host | preset | indexer |  comment
demo|192.0.2.1|1|-|Indexer and Preview environment for demo setup
demolive1|192.0.2.2|1|demo|Live Query Server for demo setup
demolive2|192.0.2.3|1|demo|Live Query Server for demo setup

The load balancer should be available at http://192.0.2.1:8080, and balance over the demolive1 and demolive2 instances of the demo environment.

Load-Balancing for the Indexer The indexer instance should not be included in the load balancer configuration together with the live instances, as it contains the unpublished configuration.

For each environment a section like this needs to be added to the config/fredhopper-haproxy.cfg file:

listen query localhost:8080
	balance leastconn
	option httpchk GET /status/
	
	# Server templates:
	# server NAME HOST:PORT check inter 3000 rise 2 fall 3 maxconn 50 [weight INTEGER] [backup]
	server demolive1 192.0.2.2:8180 check inter 3000 rise 2 fall 3 maxconn 50
	server demolive2 192.0.2.3:8180 check inter 3000 rise 2 fall 3 maxconn 50

Load Balancing Strategy HAProxy supports multiple strategies for load-balancing, the example uses the "leastconn" strategy as it distributes load based on the number of currently running queries on each server, avoiding uneven server load due to query processing time differences.

Step 3: Configure FAS

FAS needs to be able to connect to itself in certain situations, for example for accessing parts of the Business Manager. If FAS is queried directly it determines its own address based on the query, but a load balancer or transparent proxy can interfere with this logic.

This logic can be disabled by configuring each FAS explicitly to point to itself for queries by setting an option:

Execute the following command for each instance of the environment, replacing HOST, INSTANCE and PORT according to the settings from config/topology.txt.

bin/deployment-agent-client --location HOST set-option INSTANCE /com/fredhopper/transformservlet/TransformServlet/@xml-url-server-port=PORT

Determining the correct port For FAS 7.1 and earlier the port is typically calculated as "instance number" * 1000 + 7180, so that instance 1 would use port 8180, instance 2 would use port 9180, etc. For FAS 7.2 and later the port can be determined using this command: bin/deployment-agent-client --location HOST get-option INSTANCE qserver_http_port

Step 4: Start HAProxy

The following command is used to start haproxy as a daemon process using the config/fredhopper-haproxy.cfg configuration file:

haproxy -D -q -f /home/fredhopper/fredhopper/config/fredhopper-haproxy.cfg

Debugging When testing the configuration it is helpful to see some verbose messages from haproxy on the console rather than have it run as a daemon. This can be achieved by starting haproxy in the foreground: haproxy -d -f /home/fredhopper/fredhopper/config/fredhopper-haproxy.cfg

Step 5: Test

Ensure that all FAS instances are running first. It should then be possible to open the preview pages of FAS using the load balancer IP and port, for example http://192.0.2.1:8080/preview. If this is correctly working you can execute a bin/fresh-index-to-live INSTANCE. While this is running haproxy should continue to answer queries, and should print information about servers "going up" and "going down" in the order in which the fresh-index-to-live proceeds to update them.

Step 5: Enable HAProxy to run continuously

As a last step haproxy should be configured to run at boot time as a service.