Templar: A proxy to improve HTTP API interactions | Hacker News Reader

Templar: A proxy to improve HTTP API interactions

github.com

68 pointsevanphx25 comments

Based on the first sentence of the description, I was expecting a proxy that added callback support to arbitrary APIs. I hit "https://my-templar-proxy/?real_url=$x&real_param_a=10&callba..., it hits the real API and waits for the response, then hits my callback with it.

I feel like that might be more useful than what appears to be just a special-purpose varnish/squid

Author here! That's certainly an interesting feature that could be added. I have on the future list to support sending responses back via AMQP. Your callback scheme isn't much different than that.

Congrats on the announcement. Templar looks like a great project.

Re AMQP, I want to mention that our Zurl project does something similar, but with ZeroMQ. It leads to some interesting possibilities: http://blog.fanout.io/2014/02/18/fun-with-zurl-the-http-webs...

Zurl is not an HTTP proxy though, but a ZeroMQ gateway. Templar with async responses would be much more generally useful to HTTP developers.

I agree, async responses are a great tool. It's a question of figuring out a good way of delivering them back to the client. Another option is a long poll endpoint within Templar that it can deliver a stream of responses on. AMQP is an easy option because it keeps Templar simple.

This could be really useful for sending outbound webhooks, if it grew a couple of extra features:

1. "Fire and forget" ability. It would be great if I could send a request through Templar with an X-Templar-Fire-And-Forget: 1 HTTP header which means "deliver the request in your own time, but return a 200 OK to my client straight away so I don't have to wait for it.

2. It would be great if it could "validate" the URLs going through it somehow. When sending webhooks, it's important to be sure that they are going to the rest of the internet and not being used to probe internal IPs/hostnames of your own infrastructure. This is a bit of a fiddly problem, which is why it would be nice to have a proxy like Templar take it on. Take a look at the blocked_hosts section in http://search.cpan.org/~bradfitz/LWPx-ParanoidAgent-1.02/lib... for an example. I'm not sure how this would translate into X-Templar HTTP headers though.

Fire-and-Forget is on the todo list. Since you're asking about it I'll bump it up. The later, Templar could have an option that tell it to only connect to public ips. Would that do it for ya?

Yes, that would be excellent - do you mean just public IP ranges?

One possible snag: I heard from a co-worker that if you are running on EC2 and you make an HTTP request to another site that coincidentally is also hosted on EC2 that traffic can sometimes be routed over a private IP range.

That's good point actually, I could see that happening.

What is your concern about the url used? That they point at something a user is trying to coerce you to hit? If so, that could be a public IP too..

It's more about hitting internal services, where there may not be adequate protections in place. There's some earlier discussion here: https://news.ycombinator.com/item?id=7139176

Ah! Ok, I got it. Feels like the right way to handle this is the allow a blacklist to be defined via config file, then applied as request, something like "X-Templar-Blacklist: internal". The list would be a set of ip ranges and thus you'd have to construct the list so that the EC2 => EC2 problem doesn't crop up, but it's doable!

Looking at that CPAN module, it wouldn't be hard at all to add a blacklist to Templar via a config file that is checked. The EC2 => EC2 issue is probably the biggest stumbling block to getting what you want. I'll have to investigate that, it seems like that would only happen if internal to EC2 they do ICMP redirects...

The HTTPS implementation looks dangerous at first sight - it looks like an attacker could very easily just strip the X-Templar-Upgrade out from the request, and then end up with everything being un-encrypted - without the client throwing any warnings.

Could you possibly provide more details of how you avoid this?

Templar is designed to be used within your own infrastructure on a trusted network, where you talk to it and it talks to the outside world.

An attacker would have to be already within your private network, and if they were, they could observe the traffic in plain text already.

I think the confusion is around where Templar sits in relation to your app making HTTP API calls and the services you want to talk to. I'm going to draw up a diagram to help explain this better.

Note that this is by the excellent Evan Phoenix, creator of Rubinius, the Puma concurrent web server for Ruby and tons of other awesome projects.

Thanks Veejay!

Maybe I'm missing the point here, but I'm not entirely sure I see the point here.

What can Templar do for me that, say, nginx can't? All else equal, if nginx is already in my stack and Templar is not, why do I want to adopt Templar, or indeed even look twice at it?

A great question. Nginx is not typically configured as a normal proxy but it can certainly do it. A difference is that Templar gives you control of options applied to each request differently rather than the same the same to all.

From timeouts to caching, one Templar can be used for all different kinds of upstream APIs.

Well, if it's worth adding a whole new layer to my stack to deal with this kind of stuff, surely it's even more worth extending an existing layer to deal with it; as a corollary, if I know enough about the problems I'm having to know what I need Templar to do for me, then I know enough about them to know what I need nginx to do for me, too.

Again, though, maybe I'm just missing the point here. Perhaps it would help to know what use case prompted Templar's development.

In my opinion, the draw is the ability to use it when connecting to third party endpoints (so you can't control how they cache or how they respond) with third party libraries (which means you lack fine grained control over timeouts and caching on your end).

Personally, I'd rather gain the necessary control through the request mechanisms, but for a quick-and-dirty solution, this would probably work pretty well.

It's very true that things like timeouts are done better within the client library. But Templar can combine timeouts with caching, allowing an API to appear to still be active and returning something even when it's just very slow.

Because a cache is best when it's populated often, it make sense to put that into a service that many different processes on different machines can use.

Everything within templar can be done by adding capabilities to client libraries, no argument. But most client libraries don't provide these capabilities in one programming language, let alone many of them. By having the functional available as a service, Templar provides these capabilities fairly transparently (and even more transparently in the future when there is per/host and per/url configuration).

That assumes nginx has the capability to do what Templar does and I don't believe that's true.

Request collapsing looks like a nice feature . Regarding HTTPS support, I am not really sure if I understood the details mentioned in the Docs properly .

Does Templar implement SSL Termination ?

Templar does not do SSL termination because it does not terminate requests for your api. It's designed to mediate connections between you and any APIs that you use, protecting you against those APIs being unreliable.

The caching part looks really interesting -- have you considered adding support for more fine-grained caching control, such as respecting etags and last-modified times?

Thanks for sharing this!

Yup! That's on the todo list. Auto-caching, as I'm calling it, would figure out a how long to cache something for and then be able to later on use HEAD to check if the url in question has changed.

That cache time could be as long as 5 seconds, so that Templar is checking pretty often if the upstream has changed.

Reply on news.ycombinator.com