theideaofcoffee 10 months ago

HAProxy has been and continues to be one of my favorite tools because of features like this. Stuff where you might have to start digging around, say, arcane and fragile Apache rewrites, becomes a series of surprisingly readable functional-like statements that make reasoning about the implications of those statements easy. Even for HAProxy newbies, they can get going really fast and have the confidence that their changes aren't going to lead to major breakage because of its readability.

Just today I deployed some haproxy acls to transparently partition traffic for new API features off of a customer's main legacy application onto a dedicated cluster of API machines. It still surprises me just how easy doing something like that is. The most difficult part of that process was remembering the business logic--to append a query string value that that API expected to see in the request for this part of the migration. This is completely unlike munging around Apache rewrites and hoping that you didn't leave off a 'L' and make it start looping, or forget to append the query string, stuff like that.

I think that if you work with anything web-facing, regardless of the scale, you'd be well served by learning even HAProxy's most basic features and building from there. You'd be astounded at how far you can get with even a basic configuration.

  • phil21 10 months ago

    Yep. Obviously I'm a fan of HAProxy based on my post history - but it truly is one of those transformational tools you only come across a few times during a career.

    It's basically one of those tools you end up looking for problems for it to solve it's so elegant. I have my HAProxy hammer, and everything is a nail!

    Cannot give Willy and team more kudos!

    • theideaofcoffee 10 months ago

      I am the same way, I advocate a lot for replacing legacy LBs (F5? Citrix? blech) and software not really suited for LB functions, a la Apache, with HAProxy. It's working out well so far.

      Combine that with a nice network and a tasty [EI]GP and be in pure web-serving bliss (of which I've only worked on two networks with that kind of design forethought, which is a shame).

      • amerine 10 months ago

        What do you mean by nice network?

        • subway 10 months ago

          Based on the E/IGP comment, I'd guess they mean a routed layer 3 network that can speak routing protocols down to the host level. As opposed to the big ol flat networks that get pushed around frequently

zaarn 10 months ago

I absolutely love HAProxy ACLs. I use them to rewrite LetsEncrypt requests to another destination so I can cleanly request certificates on the proxy host without downtime or configuring backends or backend servers.

I've also recently migrated one proxy from Traefik to HAProxy, previously I had two proxies, one with T and one with HAP, for different IP ingests, now it's one HAP. Though I don't have failover since the networking in LXC seems to dislike multiple interfaces on the same subnet and doesn't seem to like HAProxy trying to listen on specific interfaces either, that'll have to move to VMs. (I need separated frontends with separate IPs)

The easy part there is that I fully automated HAProxy deployment including certificates with ansible, so moving from container to VM will be a breeze.

Lastly, performance is amazing. I've never had HAProxy even hiccup, even when I got DDoS'd at some point HAProxy held up like a champ.

  • jSherz 10 months ago

    Any losses from Traefik now that you're fully HAP?

    • zaarn 10 months ago

      I didn't rely on T's docker integration, that would have been a loss, I guess. HAP has much better behaviour though, T tended to buffer requests a lot which lead to several times in which either T crashed or the host system started swapping like crazy because the server had balooned to a few gigabytes memory.

znedw 10 months ago

I switched from nginx to HAProxy for my internal reverse-proxying needs and it is excellent, the configuration is a bit more verbose but it handles every app (including WebSockets, weird RPC, cookies etc) with ease.

regecks 10 months ago

Damn. All these years and I never realized anonymous/inlined ACLs were possible.

joshbaptiste 10 months ago

anyone tried Traefik?

Thinking of using it for a new home based reverse proxy solution over HAProxy

  • ahoka 10 months ago

    We are using it on production. Rules get configured from container tags.

  • haolez 10 months ago

    It’s cool, but it’s HTTP/HTTPS only, I believe.

    • arthurk 10 months ago

      It lists "Websocket, HTTP/2, GRPC ready" under Features

      • zaarn 10 months ago

        HAProxy can do TCP traffic, I do that to proxy SMTP and SSH into backend servers where they are needed.

      • ahoka 10 months ago

        All of those are HTTP :-)

bearble 10 months ago

I use HAProxy through PF-Sense to route external traffic to my internal network over SSL behind the firewall. It's been painless and great.

nickadam 10 months ago

There are some strange and obvious errors in this post. Don't get me wrong I appreciate the post and I did learn some stuff, but do some proofreading.

> 192.168..122.3 Is this some new IP format I'm not aware of?

> acl evil path_beg /evil > acl evil path_beg /evil This is the same line twice to demonstrate a logical OR (I guess one is supposed to be path_end)

  • rogerdonut 10 months ago

    Thanks for pointing this out. We have corrected this.

VvR-Ox 10 months ago

Thank you for sharing this, I think many ppl are happy for every example and code snippet they can get.

The configuration hassle with HAProxy however let me switch to nginx and traefik in the meantime...