Introduction

This document aims at explaining how to set up an active-active Denodo Cluster with 2 nodes behind the HAProxy load balancer. While the configuration is specific to the chosen load balancer, the concepts should be general enough to serve as guidelines for other load balancers configuration.

In detail, we are going to cover the following topics:

• Configuration of the load balancer.
• Implementation of HTTP session persistence (or stickiness).
• Setup of the deployment scripts in the Solution Manager.
• Considerations on the Denodo internal metadata databases in clustered deployments.
• Testing recommendations.

Software Versions

Denodo Platform	9
HAProxy	HA-Proxy version 2.0.13-2 2020/04/01 - https://haproxy.org/
Operating System	Ubuntu Server 20.04 on all the Machines

Cluster Architecture and introduction to HAProxy

The architecture discussed in this article is shown in the following diagram:

Figure 1: General Architecture with 2 Denodo Platform active nodes and a load balancer

Client applications (on the left) access the Denodo platform nodes (node01.denodo and node02.denodo) through the load balancer machine (loadbalancer). The two Denodo nodes are referenced in the Solution Manager (sm.denodo on the right) with which they have normal interactions (license check, monitoring, promotion, …). The bidirectional Solution Manager -> nodes traffic bypasses the HAProxy Load Balancer.

External (from client applications) and internal (between Denodo servers) communication is secured via SSL/TLS encryption. This is not just a typical recommendation for production deployments, but a requirement to be able to perform promotions in the Solution Manager. Indeed, starting from this version, to enable promotions the Denodo Security Token must be enabled in the involved servers and the Denodo Security Token in turn requires SSL/TLS.

In the following table, we can see all the components of the Denodo Platform that can be load balanced:

Component	Protocol	Default Port	Comments
Virtual DataPort - JDBC	TCP	9999
Virtual DataPort - ODBC	TCP	9996
Virtual DataPort - Arrow Flight SQL	RCP	9994
Scheduler	TCP	8000
Scheduler Index	TCP	9000	Not treated in this article
Data Marketplace	HTTP	9090 (SSL: 9443)
Scheduler Web Admin Tool	HTTP	9090 (SSL: 9443)
RESTful Web Service	HTTP	9090 (SSL: 9443)
REST Web Services	HTTP	9090 (SSL: 9443)
Diagnostic & Monitoring Tool	HTTP	9090 (SSL: 9443)

Components of the Denodo Platform that can be load balanced

The load balancer is based on HAProxy, a free and open source software that provides a high availability load balancer and proxy server for TCP and HTTP-based applications that spreads requests across multiple servers. It is written in C and has a reputation for being fast and efficient in terms of processor and memory usage.

Setup of HAProxy in the Load Balancer Machine

Install the package via apt (you may want to install it from source for the latest version):

$ sudo apt install haproxy

The main configuration file is /etc/haproxy/haproxy.cfg and the installation creates a systemd service:

$ systemctl status haproxy ● haproxy.service - HAProxy Load Balancer          Loaded: loaded (/lib/systemd/system/haproxy.service; enabled; vendor preset: enabled)          Active: active (running) since Mon 2020-08-10 12:36:00 CEST; 5h 55min ago            Docs: man:haproxy(1)                  file:/usr/share/doc/haproxy/configuration.txt.gz         Process: 48519 ExecStartPre=/usr/sbin/haproxy -f $CONFIG -c -q $EXTRAOPTS (code=exited, status=0/SUCCESS)    Main PID: 48530 (haproxy)           Tasks: 2 (limit: 1075)          Memory: 3.9M          CGroup: /system.slice/haproxy.service                  ├─48530 /usr/sbin/haproxy -Ws -f /etc/haproxy/haproxy.cfg -p /run/haproxy.pid -S /run/haproxy-master.sock                  └─48531 /usr/sbin/haproxy -Ws -f /etc/haproxy/haproxy.cfg -p /run/haproxy.pid -S /run/haproxy-master.sock

Every time you modify the configuration you have to restart the service in order to get it taken into account.

$ sudo systemctl restart haproxy

Before restarting you may want to check that the configuration file is valid

$ haproxy -f /etc/haproxy/haproxy.cfg -c Configuration file is valid

HAProxy Configuration File

Let’s have a look at the configuration file

global         log 127.0.0.1 local2 info         chroot /var/lib/haproxy         pidfile /var/run/haproxy.pid         maxconn 256         user haproxy         group haproxy         daemon         ## enabling the HAProxy Runtime API         stats socket ipv4@127.0.0.1:7777 level admin         stats socket /var/run/haproxy.sock mode 666 level admin         stats timeout 2m defaults         mode                 http         log                 global         option                 httplog         timeout         connect 10s         timeout         client 30s         timeout         server 30s ## http/9090 => Proxy for Apache Tomcat Based Web Applications ## - Data Marketplace ## - RESTful Web Service ## - REST Services ## - Scheduler Web Admin Tool ## - Design Studio ## - Diagnostic & Monitoring Tool frontend http-in         bind                         *:9090         default_backend backend_servers_tomcat_http         option                         forwardfor backend backend_servers_tomcat_http         balance          roundrobin         cookie       SERVERID  insert                 server                  node01 192.168.141.121:9090 cookie ck_node01 check                 server              node02 192.168.141.122:9090 cookie ck_node02 check ## https/9443 => Proxy for Apache Tomcat-based Web Applications using ## a secure (SSL/TLS) channel frontend https-in         bind                        *:9443 ssl crt /etc/haproxy/denodo_server_key_store.pem         default_backend        backend_servers_tomcat_https backend backend_servers_tomcat_https            balance          roundrobin         cookie SERVERID  insert         server           node01 192.168.141.121:9090 cookie ck_node01 check         server           node02 192.168.141.122:9090 cookie ck_node02 check ## tcp/8000 => Proxy for Scheduler Server traffic frontend tcp-in-scheduler         mode                             tcp         bind                            *:8000         default_backend                backend_servers_tcp_in_sched             option              tcplog backend backend_servers_tcp_in_sched         mode                     tcp         balance             roundrobin                 server          node01 192.168.141.121:8000 check         server          node02 192.168.141.122:8000 check ## tcp/9999 => Proxy for JDBC traffic frontend jdbc-in         mode                         tcp         bind                         *:9999         default_backend         backend_jdbc_servers         option                         tcplog backend backend_jdbc_servers         mode                  tcp         balance         roundrobin # These names must match the names defined in the Solution Manager         server                 vdp.node01 192.168.141.121:9999 check         server                 vdp.node02 192.168.141.122:9999 check ## tcp/9996 => Proxy for ODBC traffic frontend odbc-in         mode                 tcp         bind                 *:9996         default_backend backend_odbc_servers         option                 tcplog backend backend_odbc_servers         mode                tcp         balance         roundrobin         server                 node01 192.168.141.121:9996 check         server                 node02 192.168.141.122:9996 check ## tcp/9994 => Proxy for Arrow Flight SQL traffic frontend arrow-in         mode                         tcp         bind                         *:9994         default_backend         backend_arrow_servers         option                         tcplog backend backend_arrow_servers         mode                  tcp         balance         roundrobin # These names must match the names defined in the Solution Manager         server                 vdp.node01 192.168.141.121:9994 check         server                 vdp.node02 192.168.141.122:9994 check

• The global section instructs about process-wide parameters. In this example, we are telling haproxy to use the rsyslogd daemon for logging and specifying the user and group of the process.
• The defaults section lists some default parameter values.
• Finally comes the proxies section, one for each open port we want to make available through the load balancer. Each proxy consists of a frontend section and a backend section.

• The frontend section has to specify

• the protocol (tcp, http, …)
• the listening IP address and the port (ex. *:9999 indicates that we want to listen on all the available IP address of the machine)
• the default_backend
• And any additional option

• The backend section has to specify the list of servers in the backend with the following syntax

server <name> <address>[:[port]] [param*]

Every backend section name should match one and only one default_backend name in the configuration. Address can be a hostname (for example defined in /etc/hosts) but it is resolved at haproxy service start time.

• A note regarding the SSL/TLS configuration of the load balancer. If traffic between client applications and the Denodo servers is secured through SSL/TLS you should choose how the load balancer deals with secure connections.

There are several techniques:

• SSL Termination: client application-to-loadbalancer is SSL/TLS secured, loadbalancer-to-backend is not secured
• SSL Pass-through: the load balancer does not modify the connection, it just proxies it.
• Hybrid: combination of Termination and Pass-through

Each approach brings benefits and drawbacks, see the article linked in the reference section for an in-depth discussion and implementation tips in HAProxy. In our case we chose SSL Termination.

HTTP Session Stickiness

When behind the load balancer we put web applications that serve different content based on logged-in users, we must find a way to avoid same-session switching between nodes. One of the methods to deal with this is to implement the so-called application layer persistence, in which the load balancer inserts a Set-Cookie: header that is included in every following request from the same client.

You enable the persistence by adding the cookie NAME insert instruction in the backend section along with the cookie keyword followed by a unique cookie name in each server instruction.

For example:

cookie SERVERID  insert server node01 192.168.141.121:9090 cookie ck_node01 check server node02 192.168.141.122:9090 cookie ck_node02 check

In this case we are associating each newly routed connection to node01 the cookie ck_node01. The next time the same client sends a request to the load balancer it will include the cookie and the load balancer will force the routing to node01, thus bypassing the load-balancing algorithm. The same behavior applies for node02.

Configuration of Runtime API

In a later section we will see how to set up Deployment Scripts in the Solution Manager. To do that we must allow runtime modifications to the HAProxy configuration, for example to cut or enable access to a given backend server.

It turns out that we can do that by using the HAProxy Runtime API (link in the reference section).

For that to work we just need to add the following lines in the global section of the configuration file:

stats socket ipv4@127.0.0.1:9999 level admin stats socket /var/run/haproxy.sock mode 666 level admin stats timeout 2m

Setup of the deployment scripts in the Solution Manager

As a preliminary step of the deployment scripts configuration in the Solution Manager you should define:

• an environment (here dev01)
• a cluster (here cluster01_dev01) in this environment
• a Scheduler, Virtual DataPort and Data Marketplace servers of each node in the cluster.

Environment, cluster and servers definition in the Solution Manager

As an example, here is the definition of vdp.node01:

Definition of a VDP server in the Solution Manager

Deployment Scripts

In the Solution Manager we can define scripts that disable the servers in a cluster before a promotion deployment and bring them back online once the deployment is finished. The idea of these scripts is to communicate to the load balancer to temporarily cut communication to the backend node being updated.

To do that, we need to apply two settings:

1. Define the Deployment type as Without service interruption . We apply this setting in the DEPLOYMENTS pane of the environment, as shown in the image below:

Setting the Deployment Type in the Environment

2. Upload the deployment scripts in the DEPLOYMENT SCRIPT pane of the environment. We can either cut access at the cluster and node level, in our example we chose the latter.

NOTE: to get the script working, ensure that

• The sshpass package is installed in the machine hosting the Solution Manager Server.
• The socat package is installed in the machine hosting the load balancer.
• The host public key of the loadbalancer machine is known in the Solution Manager Server machine. This means that an entry with that hostname must be present in file ~/.ssh/known_hosts. You can do that with:

$ ssh-keyscan -H loadbalancer >> ~/.ssh/known_hosts

Here is the script to be uploaded:

#!/bin/bash ##################### ## This script allows enabling or disabling a node on a ## HAProxy based load balancer ##################### nodename=$1 status=$2 loadbalancer_user=$3 loadbalancer_pwd=$4 loadbalancer_host=$5 set_instruction="set server backend_jdbc_servers/${nodename} state ${status}" echo "echo ${set_instruction} | socat stdio /var/run/haproxy.sock" | sshpass -p${loadbalancer_pwd} ssh ${loadbalancer_user}@${loadbalancer_host} 'bash -' echo "EXIT CODE=$?"

In this script we are telling HAProxy via a remote SSH command (sshpass) that we want to change the status of backend_jdbc_servers/${nodename} to ${status}. As such this script can be used both for enabling (state: ready) and disabling (state: maint) the relevant node. In this example we only deal with the Virtual DataPort JDBC traffic; in a more realistic scenario you would need to take into account all the components that should be disabled/enabled (e.g. Scheduler and Data Marketplace).

Parameter passing is done by graphically defining parameters. In our example there are five of them:

param_number	name	type	value
0	nodename	load balancing variable	name
1	state	literal	ready
2	loadbalancer_user	load balancing variable	loadbalancer_user
3	loadbalancer_pwd	load balancing variable	**********
4	loadbalancer_host	load balancing variable	loadbalancer_host

Parameters can be of type literal, where you specify the value at creation time, or load balancing variable, that are defined at the cluster or the server level and can be handy when a parameter corresponds to a node (nodename) or to a cluster (such as loadbalancer_user,loadbalancer_pwd and loadbalanceer_host). Sensitive parameters, such as passwords, can be stored encrypted.

Deployment Scripts configuration: enabling (top) and disabling the node (bottom)

Definition of Load Balancing variables at the server and the cluster level

Testing of Deployment Scripts

We can test that your deployments scripts work as expected by deploying a revision in the Solution Manager.

Afterwards, we can inspect the load balancer logs (/var/log/haproxy.log)

Aug 12 16:10:45 localhost.localdomain haproxy[55119]: Server backend_jdbc_servers/vdp.node01 is going DOWN for maintenance. 1 active and 0 backup servers left. 0 sessions active, 0 requeued, 0 remaining in queue. Aug 12 16:10:45 loadbalancer haproxy[55119]: [WARNING] 224/161045 (55119) : Server backend_jdbc_servers/vdp.node01 is going DOWN for maintenance. 1 active and 0 backup servers left. 0 sessions active, 0 requeued, 0 remaining in queue. Aug 12 16:10:51 localhost.localdomain haproxy[55119]: Server backend_jdbc_servers/vdp.node01 is UP/READY (leaving forced maintenance). Aug 12 16:10:51 loadbalancer haproxy[55119]: [WARNING] 224/161051 (55119) : Server backend_jdbc_servers/vdp.node01 is UP/READY (leaving forced maintenance). Aug 12 16:10:51 localhost.localdomain haproxy[55119]: Server backend_jdbc_servers/vdp.node02 is going DOWN for maintenance. 1 active and 0 backup servers left. 0 sessions active, 0 requeued, 0 remaining in queue. Aug 12 16:10:51 loadbalancer haproxy[55119]: [WARNING] 224/161051 (55119) : Server backend_jdbc_servers/vdp.node02 is going DOWN for maintenance. 1 active and 0 backup servers left. 0 sessions active, 0 requeued, 0 remaining in queue. Aug 12 16:10:56 localhost.localdomain haproxy[55119]: Server backend_jdbc_servers/vdp.node02 is UP/READY (leaving forced maintenance). Aug 12 16:10:56 loadbalancer haproxy[55119]: [WARNING] 224/161056 (55119) : Server backend_jdbc_servers/vdp.node02 is UP/READY (leaving forced maintenance).

From the logs, we clearly see that the node access has been cut off and restored one after another.

We can also have a look at the deployment summary in the Solution Manager Administration Tool that confirms that VQL deployments have been bracketed among SCRIPT_SERVER tasks that correspond to script executions in the load balancer.

Deployment status showing that the enable and disable scripts have been run before and after the VQL deployment

Considerations on the Denodo internal metadata databases in clustered deployments

Each Denodo component maintains a dedicated metadata database. When deploying in clusters the following aspects should be taken into account for the architecture design:

Scheduler

In cluster mode, the Scheduler servers that constitute the cluster must share the Scheduler internal metadata database. You can consult the Scheduler Cluster Settings manual page to set up this scenario and to understand the technical implications of having such deployment.

Data Marketplace

In cluster mode, the Data Marketplace servers that constitute the cluster must share the Data Marketplace internal metadata database. You can consult the Data Marketplace External Database Setup manual page for the configuration steps.

Virtual DataPort

In cluster mode, the Virtual Data Port servers may share the internal metadata database.

If each node interacts (read/write) only with its local metadata database, there must be an external agent that guarantees the synchronization between all the local metadata databases, one for each node.

If a shared metadata database is configured, the replication is done automatically.

By default the Virtual DataPort metadata database is not shared. To understand how the metadata database can be shared and to read some considerations and insight on this choice you can refer to the Storing the Metadata on an External Database manual page.

Diagnostic & Monitoring Tool

The Diagnostic & Monitoring Tool does not support a shared metadata database so the configuration, environments and servers must be synchronized manually.

Summary

Denodo Platform Component	Support for shared DB for metadata	Compulsory shared DB in clusters
Virtual DataPort	yes	no
Scheduler	yes	yes
Data Marketplace	yes	yes
Diagnostic & Monitoring Tool	no	no

Summary of support for shared metadata databases of the Denodo components

Testing

You can test your Denodo clustered deployment setup by:

• trying to reach all the Denodo services through the load balancer.
• verifying that the load is balanced by checking that the nodes receive the expected portion of a simulated workload, taking into account the workload distribution algorithm logic. To this end, you may find it useful to monitor both the Denodo nodes via the Diagnostic & Monitoring Tool and the logs of the load balancer.

 

Web based applications

• Data Marketplace: Login, do some typical actions (search, query, categorize, administer..) then logout. Repeat.

• HTTP: http://loadbalancer:9090/denodo-data-catalog/AutoLogin
• HTTPS: https://loadbalancer:9443/denodo-data-catalog/AutoLogin

• Scheduler Web Admin: Login, do some typical actions (explore jobs, create a new one, delete one job, edit one job, …) then logout.  Repeat.

• HTTP: http://loadbalancer:9090/webadmin/denodo-scheduler-admin/login
• HTTPS: https://loadbalancer:9443/webadmin/denodo-scheduler-admin/login

• Design Studio: Login, do some typical actions (explore catalog, open a view, create a derived view, execute it, ..) then logout. Repeat.

• HTTP: https://loadbalancer:9090/denodo-design-studio/#/        
• HTTPS: https://loadbalancer:9443/denodo-design-studio/#/        

• RESTful Web Service: Login and do some typical actions (navigate views, search with url parameters, …) then logout. Repeat.

• HTTP: http://loadbalancer:9090/denodo-restfulws/
• HTTPS: https://loadbalancer:9443/denodo-restfulws/

• REST Web Services: Query a deployed REST web service

• HTTP: http://loadbalancer:9090/server/<virtual database name>/<web service name/views/<view name>?param_1=val_1&param_2=val_2[...]
• HTTPS: https://loadbalancer:9443/server/<virtual database name>/<web service name/views/<view name>?param_1=val_1&param_2=val_2[...]

• Diagnostic & Monitoring Tool: Login and do some typical actions (create a new monitored environment, refresh the requests table, …) then logout. Repeat.

• HTTP: http://loadbalancer:9090/diagnostic-monitoring-tool/Home
• HTTPS: https://loadbalancer:9443/diagnostic-monitoring-tool/Home

JDBC Client

        Successful connection to Virtual DataPort using JDBC

ODBC Client

Windows ODBC Data Source Creation

Scheduler

You have several options to test the configuration:

• Graphically: Connect to the scheduler server using the Scheduler Web Administration Tool specifying the loadbalancer hostname
• Command line: Customize the sample script provided under $DENODO_HOME/samples/scheduler/scheduler-api/ that leverages the Scheduler RMI Client API. You will probably have to modify the parameters in samples/scheduler/scheduler-api/src/com/denodo/scheduler/demo/GlobalNames.java, launch $DENODO_HOME/samples/scheduler/scheduler-api/scripts/test_schedulerclient.sh. After these steps, you can run a job from the command line:

$DENODO_HOME/samples/scheduler/scheduler-api/scripts/test_schedulerclient.sh -start job_extract_iv_sales -h loadbalancer -p 8000

• Command line: Leverage the Scheduler REST Client API

References

HAProxy in Wikipedia

HAProxy Home page

Dynamic Configuration of HAProxy via API

HTTP Session stickiness

HAProxy tutorial

Using SSL Certificates with HAProxy

Configuring Deployment Scripts in the Solution Manager