HAProxy Reviews

We are using HAProxy for our internal load balancers between the front-end and back-end services. HAProxies are configured either manually (LVS redundancy included) or automatically (ambassador pattern by using a Consul & Consul-Template).

It reduced the load on our main load balancers.

• Flexible health-checking
• Good config format to auto-configure

We need to handle new connections by dropping, or queuing them while the HAProxy restarts, and because HAProxy does not handle split config files.

No stability issues.

No scalability issues.

We did not need technical support because the documentation is good.

The initial setup is pretty straightforward.

We use NGINX as well. However, because the health checks are a paid feature, I like to avoid it whenever possible.

We pretty much know all the software out there and its capabilities, so we did not evaluate anything else.

Use a split front-end and back-end instead of using the "listen" directive. Therefore, you are clear what you are setting up. 

Be aware of connections dropping during restarts, etc.

We have the following use-cases for HAProxy:

1. To load-balance dozens of Apache 2.4 Servers mod_proxy. (Internal load-balance Tomcat, Jetty, JBoss app containers, using TCP load-balancing).
2. To load-balance hundreds of MySQL and PostgreSQL databases using TCP load-balancing. We manage inventory of these through Ansible automation.
3. To provide a layer of security (username/passwd) authentication for legacy back-end Web apps that may not have username/passwd implemented yet. Some financial Web apps were created over 15 years ago and focused on reports, files, logs, and market share stats and were written in Perl. We also had a very old Kibana interface to visualize those logs. Such Web apps required HAProxy to tunnel the requests with un/pw authentication.
4. To redirect traffic internally based on /URL to the relevant services (DNS nameserver) and as a gateway to tunnel traffic to customers who explicitly require reverse-IP authentication. The DNS nameserver was a trendsetter that we learned quickly and now cannot live without.

As our traffic began increasing nine years ago, we desperately needed to load-balance TCP requests (for DBs). We originally used round robin on an array[] which stored the IPs of half a dozen DBs. But with HAProxy, we didn't need to maintain such complexity. We later exploited many more features.

The most important features would be the load-balancing of HTTP and TCP requests, according to multiple LB algorithms (busyness, weighted-busyness, round robin, traffic, etc). 

Another important feature that we cannot live without is the username/passwd authentication for legacy systems that had none.

The web stats UI, which provides the status of the health and numbers, could greatly benefit from having a RESTful interface to control the load-balanced nodes. Although there is a hack around the UI (by issuing a POST request to HAProxy with parameters), a RESTful interface would greatly improve the automation process (through Chef and Ansible).

I have deployed over 50 instances of HAProxy over the past 15 years and never encountered any stability issues. Most HAProxy instances have continuously run for over two years until the server required a kernel upgrade.

I have used it for over 10 years. I started using it as a Web application (Tomcat, Apache, JBoss) load-balancer when it had a few stable releases. When I first start using it, HAProxy was primarily used to load-balance HTTP requests. Since we are a B2B company that deals primarily with hotel inventory, IP authentication was a must. Therefore, our customers had single end-points to send and receive RESTful requests. To make this viable, we had to use a central server as a proxy to tunnel out the requests. We will continue to use HAProxy as our entry-point and exit-point of the system.

Primary use case of this solution is during PoC, demos, and in development environments. I have found HAProxy very helpful in replicating production environment architecture in a development and testing environment.

It has been a great help for all the use cases noted above, and economical.

Honestly, I have only evaluated the load-balancing feature because that is what I needed in my environment. As you can understand, for development and end-to-end testing, procuring a hardware load-balancer is a pretty expensive a deal.

I would like to evaluate load-balancing algorithms other than round robin and SSL offloading.

Also, it would be helpful if I could logically divide the HAProxy load-balancing into multiple entities so that I would install one HA Proxy LB application which could be used for different Web servers for different applications. I am not sure if these features are available.

I am using HAProxy on a live production Web Server (Apache load balancing). It's a very good open-source solution.

Four to five years back, we were struggling with Web load-balancing; we were facing issues with a single Apache. We got the best, and open-source, solution.

Load balancing, mainly. We are using two Apaches with HAProxy load-balancing.

Some people use HAProxy to load balance MySQL databases as well.

They should introduce one feature that I know many people, including me, are waiting for: HAProxy should have provide hot-swipe for back-end servers, no downtime while shifting back-end servers.

They also need a more detailed GUI for monitoring and configuration.

I have been using this solution for the last four to five years with no downtime.

No issues with scalability.

Easy to set up.

If you don't have expertise then go with the licensed version. Otherwise, open-source is the best solution.

I will always give HAProxy a 10 out of 10. We never have any downtime with it. The HAProxy team regularly updates the product. I would definitely recommend HAProxy. You must go with this product.

Proxying the requests for our API to a number of back-ends. We are using it in Linux with Keepalived to ensure high availability.

It has allowed us to evenly distribute the load across a number of servers, and check their health and automatically react to errors. It also allows us to do graceful reloads, so not a single request is lost.

• Performance
• Stability
• Security
• Active health checks of back-end servers
• A lot of configuration options which let you deeply customize it.

It needs proper HTTP/2 support.

It serves as a fast front-end solution to our services.

Also, it uses the mobile detection module that I developed and maintain.

It improves our scalability and responsiveness services to meet our demanding customer requirements. We now have much better metrics regarding clients types.

• The Lua module to reach an increasing user base.
• HTTP/2 support
• Multi-thread support

Maybe HAProxy could be more modular. 

The SSL termination was a nice, useful addition.

Load-balancing between hosting sites, using Level 4 and Level 7 (with proxy-protocol, depending on the requirements).

It makes the hosting on multiple platforms/datacenters very easy, without having to worry about the high availability required by our customers.

The anti-DDOS PacketShield filtering solution (embedded in the physical appliances) as well as the BGP route injection are great features and heavily used.

The firmware upgrades are extremely easy, fast and convenient. You only have to upload it and reboot. It is as fast and easy to roll back to the previous version.

These features are really an advantage over a "homemade" solution.

The basic clustering is not usable in our very specific setup. The clustering is mainly a configuration replication and is great in a case of active-passive usage. In the case of an active-active (or with more than two nodes) where the configuration is not fully identical, it cannot be used as-is.

We did encounter some stability bugs, but all were ironed out diligently by support.

No scalability issues, ever.

Excellent so far. We have always been able to deal with any problem, quickly and efficiently.

We used something else a long time ago. The old solution wasn't addressing our needs.

Straightforward, but received some help for the migration.

Very good value for the money. One of the simplest licensing schemes in this category of products.

Test/lab virtual machines can be installed without a licence. They can't be used for performance testing but otherwise behave like production nodes.

We are mainly using HAProxy Community Edition on our own servers. We also use this solution depending on our needs, but maintenance and upgrades are more complicated.

Play a lot with the VM version. The product is powerful but some features are a bit hidden (read the HAProxy docs).

E2E load balancing of Layer 7 and Layer 4 applications.

Software defined load balancing allows us to dynamically adjust and codify routing decisions. This speeds up development.

Performance and SSL proxy/offloading capability. Compared to nginx it’s a lot cleaner and quicker.

Dynamic update API. More things should be possible to be configured during runtime.

We would like to see dynamic ACL and port update support. Our infrastructure relies on randomly allocated ports and this feature would allow us to update without restarting the process. The ACL add/update would help with some direct routing challenges that currently require us to work around them with a map and static back-ends.

No issues with stability.

No issues with scalability.

A+++. Super-quick to respond, and always on target with answers specific to the current issue.

We did use hardware loadbalancing, and still use nginx for some Layer 7 routing challenges. We switched because software defined loadbalancing allows us to dynamically adjust and codify routing decisions. This speeds up development.

The learning curve is small if one is familiar with routing/networking in general, but it takes some time to fully understand the impact of some configuration settings. The support for all major Linux distros makes running and testing a breeze though.

NTLM/F5 hardware, nginx.

Use it for some small, non-critical systems first, get comfy with the stats, and then scale out. Codify your configuration and keep it as simple as the requirements allow.