6 ms·

Lambda@Edge – Preview

110 points by forrestbrazeal 10y ago

hackcrafter 10y ago
> Because your JavaScript code will be part of the request/response path, it must be lean, mean, and self-contained. It cannot make calls to other web services and it cannot access other AWS resources. It must run within 128 MB of memory, and complete within 50 ms. I'm having a hard time conceptualizing what kind of logic you would run on these edge functions? You have access to headers, so also things like client cookies, so could you use that to potentially personalize/templatize the body (once they have body mutation abilities)? Or is it really just about static content routing? Detect google bot user agent and send them to resource A versus B etc...
- ash_gti 10y ago
  Couldn't you have your apps primary routing handling in this?
  - forrestbrazeal 10y ago
    The blog post says that one of the primary use cases is to help preview content show up correctly on Facebook, etc. So I don't think client-side routing for static apps will work in this case.
  - mifreewil 10y ago
    The linked post uses CloudFront as an example, which is commonly used simply as a CDN in front of static content on S3. This allows you to take that model and add some dynamic content to it. I just recently used this model to host a JS widget used on external websites, now I could for example add some basic logic to disallow certain domains (by checking referrer headers) from using the widget.
- Alex3917 10y ago
  > I'm having a hard time conceptualizing what kind of logic you would run on these edge functions? Two use cases: 1) Routing Facebook and Twitter to a pre-rendered version of your single page app based on their user agents. 2) Adding custom headers to your responses, e.g. for a Content Security Policy. There is no good way to do either of those things currently. You could have one off options for each of those and the 50 other common use cases, but this is actually a really elegant solution. Very well done.
  - jensvdh 10y ago
    Would you be able to route facebook/google to a pre-rendered version of your app without sending it a 301 redirect though?
- jalk 10y ago
  This is totally cool if the origin is just a dumb S3 bucket. From what I can tell, you can implement the same kind of rewrite logic you would otherwise have in i.e. httpd/nginx conf. And of course doing so at the edge means that you potentially shave off some latency during redirects.
- wiredfool 10y ago
  It's a potentially cleaner way to send back a redirect from API gateway/lambda rather than pattern matching on http in the error string of the Jason response body.
saurik 10y ago
While having minimal access to the request to do things like add headers is of course quite useful, the people I've known who most want code execution at the edge are either already using or looking at using ESI (Edge-Side Includes, as provided by Akamai and Varnish), in order to express things like "use the cached content for almost the entire page but add some per-user information to the response via a secondary virtual request (the content for which might also be cached at the edge!) for people who are logged on; they say that at some point they will allow access to and potentially modification of the body, but it isn't clear to me that this will be as potent as (even if it eventually becomes technically as powerful as) ESI.
Alex3917 10y ago
For what it's worth, here is the original thread: https://news.ycombinator.com/item?id=13011120 https://news.ycombinator.com/item?id=13011120 This should definitely meet our needs in both cases: serving prerendered content when necessary, and also adding the security headers to the front end. And it's definitely a 10x better solution than having an extra EC2 instance with your private certs on it, just to run an nginx proxy with twenty lines of code.
- saurik 10y ago
  Wait, by "original thread" are you implying this feature was thrown together in the past week and a half since you asked that question?
  - linkregister 10y ago
    Yes, the HN thread is prominent in the article.
    - saurik 10y ago
      Ah, OK: the answer is "no" (as the question did not lead to the feature), not "yes" ;P; and also "I am dumb" (FWIW, I actually got to this thread by searching for it after already knowing about @Edge, so I didn't even click the article link: I saw the @Edge stuff today in the ECS FAQ, searched for it on Google and saw it had been announced 15 minutes earlier, and then came to HN to read comments about it).
  - Alex3917 10y ago
    Of course not. But it still explains what this new feature does that AWS wasn't able to do previously.
  - shade23 10y ago
    >Without spilling any beans I let him know that we were very aware of this use case and that we had some interesting solutions in the works. Other customers have also let us know that they want to customize their end user experience by making quick decisions out at the edge. It was in the pipeline already I guess.
slaymaker1907 10y ago
I somewhat wonder if this will be used by malicious websites for cloaking. For those who are unaware of the term, cloaking is used by malicious websites/ads to render different content to search engines and bots than for real users to avoid detection by those automated agents.
dustingetz 10y ago
you can use this to serve dynamic websites out of CDN, for example github pages, which we talk about as a static site but underneath it is a slow moving database (your git repo)
jotto 10y ago
For anyone needing a minimal effort prerendering solution (server side render for most JS frameworks), i.e. the use case described by Jeff - I made https://www.prerender.cloud/ https://www.prerender.cloud/ (you could use the prerendercloud npm middleware from the lambda edge)
dkordik 10y ago
Interesting... looks like something we might want to use to do URL rewriting for an S3-hosted SPA test environment, so that deep links resolve to the base of the SPA without having to use a more "dynamic" host like an EC2 instance running nginx. [edit] ah, nevermind, you have to be using CloudFront too.
- brutopia 10y ago
  This so much! Cloudfront/lambda/s3 combo is missing this piece for which you have to use unnecessarily complex api gateway or ec2 varnish/nginx setup. Thanks Jeff!
pedalpete 10y ago
This seems it might be a way to check if a user is logged-in or not and provide the appropriate page. I'm thinking if the user details are in a jwt or cookie, you could render the correct page on insecure routes.
kiwidrew 10y ago
This is very, very cool. It allows you to deploy application logic extremely close to the user (in terms of latency), and Cloudfront has an order of magnitude more edge locations than the other AWS services like EC2 and S3. In fact, if the promised ability to read/write the request/response bodies does appear, it might be possible to deploy your entire website or app as a single Lambda@Edge function. (Not every app would fit within the size, memory, and CPU constraints, of course.) The main limitation would probably be the fact that for each incoming HTTP request, your choices are limited to either making zero origin requests (i.e. the function generates the response directly) or one origin request. It's not clear how this could work if you wanted to make two origin requests and combine them to render a single response. I'm totally going to sign up for the preview and play around with it! Especially if, like the regular node.js Lambda functions, you can include arbitrary Linux binaries with the function code. Compiled request-processing code would easily meet the 128MB/50ms criteria.
scrollaway 10y ago
So we're getting C# and a new Node.js runtime for Lambda. Jeff, surely you're reading this - when are we getting Python 3 support as an additional runtime? I've been waiting for months for re:invent in the hope that it would be announced today, only to be disappointed. I'm not the only one in this situation: stuck on Python 2.7, hitting issues left and right because of an ancient runtime that we no longer would care about were it not for Lambda. Just today I hit another issue in production which impacted our webhook system. Wouldn't you know it, `json.dumps()` produces illegal json in Python 2.7 when you give it an IntEnum: >>> from enum import IntEnum; from json import dumps, loads >>> class E(IntEnum): ... A = 1 >>> dumps(E.A) 'E.A' >>> loads(dumps(E.A)) Traceback (most recent call last): ... ValueError: No JSON object could be decoded Please. Do something about this. It's wasting so much time :(
- Mizza 10y ago
  Was really expecting a lot more out of the announcements today as well. Python3, GPUs, custom containers, larger packages.. the list goes on.. That being said, re: Python 3, you should keep a close eye on Zappa (https://github.com/Miserlou/Zappa https://github.com/Miserlou/Zappa) - as we've got a surprise in store that may delight you!
  - scrollaway 10y ago
    I've looked at Zappa in the past; it's very impressive, but it's overkill for what we're doing - we're only processing files and running a Django ORM instance behind it. And because we have quite a high throughput, running a Python 3 shim behind Python 2 would increase our costs by too much. Thank you for making it though. It's a really cool part of the Lambda ecosystem. Excited to hear there's something in store for Py3 :)
  - movedx 10y ago
    So this looks very cool. Great job... but... It would be pretty simple to wrap this up and automate it easily using a Terraform file. All it requires is an S3 resource for the ZIP, a Lambda resource using the S3 resource for the code, and finally an API Gateway resource to tie it all together. Thoughts?

110 points by forrestbrazeal 10y ago

> Because your JavaScript code will be part of the request/response path, it must be lean, mean, and self-contained. It cannot make calls to other web services and it cannot access other AWS resources. It must run within 128 MB of memory, and complete within 50 ms. I'm having a hard time conceptualizing what kind of logic you would run on these edge functions? You have access to headers, so also things like client cookies, so could you use that to potentially personalize/templatize the body (once they have body mutation abilities)? Or is it really just about static content routing? Detect google bot user agent and send them to resource A versus B etc...

ash_gti 10y ago
Couldn't you have your apps primary routing handling in this?
- forrestbrazeal 10y ago
  The blog post says that one of the primary use cases is to help preview content show up correctly on Facebook, etc. So I don't think client-side routing for static apps will work in this case.
- mifreewil 10y ago
  The linked post uses CloudFront as an example, which is commonly used simply as a CDN in front of static content on S3. This allows you to take that model and add some dynamic content to it. I just recently used this model to host a JS widget used on external websites, now I could for example add some basic logic to disallow certain domains (by checking referrer headers) from using the widget.
Alex3917 10y ago
> I'm having a hard time conceptualizing what kind of logic you would run on these edge functions? Two use cases: 1) Routing Facebook and Twitter to a pre-rendered version of your single page app based on their user agents. 2) Adding custom headers to your responses, e.g. for a Content Security Policy. There is no good way to do either of those things currently. You could have one off options for each of those and the 50 other common use cases, but this is actually a really elegant solution. Very well done.
- jensvdh 10y ago
  Would you be able to route facebook/google to a pre-rendered version of your app without sending it a 301 redirect though?
jalk 10y ago
This is totally cool if the origin is just a dumb S3 bucket. From what I can tell, you can implement the same kind of rewrite logic you would otherwise have in i.e. httpd/nginx conf. And of course doing so at the edge means that you potentially shave off some latency during redirects.
wiredfool 10y ago
It's a potentially cleaner way to send back a redirect from API gateway/lambda rather than pattern matching on http in the error string of the Jason response body.

saurik 10y ago

While having minimal access to the request to do things like add headers is of course quite useful, the people I've known who most want code execution at the edge are either already using or looking at using ESI (Edge-Side Includes, as provided by Akamai and Varnish), in order to express things like "use the cached content for almost the entire page but add some per-user information to the response via a secondary virtual request (the content for which might also be cached at the edge!) for people who are logged on; they say that at some point they will allow access to and potentially modification of the body, but it isn't clear to me that this will be as potent as (even if it eventually becomes technically as powerful as) ESI.

Alex3917 10y ago

For what it's worth, here is the original thread: https://news.ycombinator.com/item?id=13011120 https://news.ycombinator.com/item?id=13011120 This should definitely meet our needs in both cases: serving prerendered content when necessary, and also adding the security headers to the front end. And it's definitely a 10x better solution than having an extra EC2 instance with your private certs on it, just to run an nginx proxy with twenty lines of code.

saurik 10y ago
Wait, by "original thread" are you implying this feature was thrown together in the past week and a half since you asked that question?
- linkregister 10y ago
  Yes, the HN thread is prominent in the article.
  - saurik 10y ago
    Ah, OK: the answer is "no" (as the question did not lead to the feature), not "yes" ;P; and also "I am dumb" (FWIW, I actually got to this thread by searching for it after already knowing about @Edge, so I didn't even click the article link: I saw the @Edge stuff today in the ECS FAQ, searched for it on Google and saw it had been announced 15 minutes earlier, and then came to HN to read comments about it).
- Alex3917 10y ago
  Of course not. But it still explains what this new feature does that AWS wasn't able to do previously.
- shade23 10y ago
  >Without spilling any beans I let him know that we were very aware of this use case and that we had some interesting solutions in the works. Other customers have also let us know that they want to customize their end user experience by making quick decisions out at the edge. It was in the pipeline already I guess.

slaymaker1907 10y ago

I somewhat wonder if this will be used by malicious websites for cloaking. For those who are unaware of the term, cloaking is used by malicious websites/ads to render different content to search engines and bots than for real users to avoid detection by those automated agents.

dustingetz 10y ago

you can use this to serve dynamic websites out of CDN, for example github pages, which we talk about as a static site but underneath it is a slow moving database (your git repo)

jotto 10y ago

For anyone needing a minimal effort prerendering solution (server side render for most JS frameworks), i.e. the use case described by Jeff - I made https://www.prerender.cloud/ https://www.prerender.cloud/ (you could use the prerendercloud npm middleware from the lambda edge)

dkordik 10y ago

Interesting... looks like something we might want to use to do URL rewriting for an S3-hosted SPA test environment, so that deep links resolve to the base of the SPA without having to use a more "dynamic" host like an EC2 instance running nginx. [edit] ah, nevermind, you have to be using CloudFront too.

brutopia 10y ago
This so much! Cloudfront/lambda/s3 combo is missing this piece for which you have to use unnecessarily complex api gateway or ec2 varnish/nginx setup. Thanks Jeff!

pedalpete 10y ago

This seems it might be a way to check if a user is logged-in or not and provide the appropriate page. I'm thinking if the user details are in a jwt or cookie, you could render the correct page on insecure routes.

kiwidrew 10y ago

This is very, very cool. It allows you to deploy application logic extremely close to the user (in terms of latency), and Cloudfront has an order of magnitude more edge locations than the other AWS services like EC2 and S3. In fact, if the promised ability to read/write the request/response bodies does appear, it might be possible to deploy your entire website or app as a single Lambda@Edge function. (Not every app would fit within the size, memory, and CPU constraints, of course.) The main limitation would probably be the fact that for each incoming HTTP request, your choices are limited to either making zero origin requests (i.e. the function generates the response directly) or one origin request. It's not clear how this could work if you wanted to make two origin requests and combine them to render a single response. I'm totally going to sign up for the preview and play around with it! Especially if, like the regular node.js Lambda functions, you can include arbitrary Linux binaries with the function code. Compiled request-processing code would easily meet the 128MB/50ms criteria.

scrollaway 10y ago

So we're getting C# and a new Node.js runtime for Lambda. Jeff, surely you're reading this - when are we getting Python 3 support as an additional runtime? I've been waiting for months for re:invent in the hope that it would be announced today, only to be disappointed. I'm not the only one in this situation: stuck on Python 2.7, hitting issues left and right because of an ancient runtime that we no longer would care about were it not for Lambda. Just today I hit another issue in production which impacted our webhook system. Wouldn't you know it, `json.dumps()` produces illegal json in Python 2.7 when you give it an IntEnum: >>> from enum import IntEnum; from json import dumps, loads >>> class E(IntEnum): ... A = 1 >>> dumps(E.A) 'E.A' >>> loads(dumps(E.A)) Traceback (most recent call last): ... ValueError: No JSON object could be decoded Please. Do something about this. It's wasting so much time :(

Mizza 10y ago
Was really expecting a lot more out of the announcements today as well. Python3, GPUs, custom containers, larger packages.. the list goes on.. That being said, re: Python 3, you should keep a close eye on Zappa (https://github.com/Miserlou/Zappa https://github.com/Miserlou/Zappa) - as we've got a surprise in store that may delight you!
- scrollaway 10y ago
  I've looked at Zappa in the past; it's very impressive, but it's overkill for what we're doing - we're only processing files and running a Django ORM instance behind it. And because we have quite a high throughput, running a Python 3 shim behind Python 2 would increase our costs by too much. Thank you for making it though. It's a really cool part of the Lambda ecosystem. Excited to hear there's something in store for Py3 :)
- movedx 10y ago
  So this looks very cool. Great job... but... It would be pretty simple to wrap this up and automate it easily using a Terraform file. All it requires is an S3 resource for the ZIP, a Lambda resource using the S3 resource for the code, and finally an API Gateway resource to tie it all together. Thoughts?