I'm trying to optimize my web application using Google's Page Speed API which has highlighted the absence of "Keep-alive" in my HTTP response headers as a major page speed weakness.
In talking with my back-end devs and sys admins, they've told me that using Keep-alive on the site is impossible because we use a load balancer.
I'm wondering, is this accurate? Are there load balancers that support Keep-alive?
It seems strange to me that the Page Speed API would complain about Keep-alive if it were impossible to use with load balancers because I would imagine a fair amount of applications and large sites use load balancers.
Thanks!
I don't know what type of load-balancers do you have... but I don't think that it would prevent the use of keep-alive connections.
The load balancer will handle each incoming connection to one of the backend servers. Now for each object the browser needs to make a new connection just to fetch that object (for example all small images). Establishing and closing TCP connections takes some time. This is why the Google Page Speed suggests to have keep-alive turned on. Another option is put all your small images into one big image and use css sprites to display part of it on different places on your page.
But back to the load balancer. If you have network load balancer, it should work without any questions - it will just redirect incoming TCP connection to one of the backend servers. If you have HTTP load-balancer, it will accept the connection, read the request, send the request to backend server, wait for it to answer and send the answer back to the browser. If you enable keep-alive, the load balancer should forward the next request it receives over the same connection.
For dynamic pages you don't need keep-alive. Keep-alive is mainly useful for static content (js, images, css) as for each one html page you have usually more than 10 static objects. So I would suggest to continue serving html trough that load-balancer and serve static content over different hostname (static.example.com).
Related
I created an instance group through an instance template, and aligned this instance group to a backend service which is used by a http load balancer.
Now when I open a url to an instance vm from the instance group I created, I can do GET POST and DELETE requests and all of the requests are fast, and everything works as expected.
When I open up the url to the static IP for the load balancer. I can do GET and POST requests, but DELETE requests throw a 400 BAD REQUEST with a response page saying:
That’s an error.
Your client has issued a malformed or illegal request. That’s all we
know.
Other load balancer issues:
The site is quite slow through the load balancer. Perhaps
there is a setting I'm missing, I'm pretty sure I set everything to
us-central-1b.
Sometimes the site doesn't even show up. It will work for http, but then
it won't work for https and visa versa. The load balancer has very strange
behaviour.
My VM api access is set to This instance has full API access to all Google Cloud services
I'm using Django as my api layer, I turned on debugging on this host and saw that the DELETE requests weren't even coming through when making requests through the loadbalancer static ip. Is there a firewall setting I'm missing?
Please help me make this fast again and allow the DELETE requests to happen.
Thanks!
Are you sending anything in the body of the request?
Google load balancer will respond with 400 BAD REQUEST if you try to send anything in the body. Easy way to check if this is the problem is fire up Chrome Developer tools and check the Request Payload section is empty/doesn't exist.
The HTTP spec doesn't explicitly say wether you can pass anything in the body so this isn't wrong, just undefined.
Is the load balancer slow for all requests or just pages with lots of elements on?
My server side setup is like this : I have a google compute engine instance running nodejs server. It sits behind https load balancer.
So the issue is that if OPTIONS type request is sent with body(payload) to load balancer ip, it sometimes fail with 502. I have checked that in all these requests, node server is returning and logging correct responses. It works perfect if the request body is left empty.
However, when I directly hit GCE's ip, it works fine whether or not request body is there or not.
For e.g
this is the load balancer ip (with ssl enabled)
OPTIONS https://130.211.14.60/health
this is the direct GCE machine ip (without ssl)
OPTIONS http://104.199.159.212:8002/health
I have checked by sending requests multiple times(literally hundreds of times) using Postman app.
And this issue is only with DELETE and OPTIONS type requests. GET/POST/PUT works perfectly fine.
Is there anyone who can point me what the issue could be and how to solve it.
From Google's docs -
The HTTP(S) load balancer does not support sending an HTTP DELETE with a body to the load balancer. Such requests will receive an error message: Error 400 (Bad Request)!! Your client has issued a malformed or illegal request. Only DELETE requests without bodies are supported.
More info here.
This feature is now supported by Google Load Balancer, released in Dec '18.
More info here.
Are there any specific spec'd processes that a browser client can use to dynamically encourage a server to push additional requested items into the browser cache using HTTP/2 server push before the client needs to actually use them (not talking about server-side events or WebSockets, here, btw, but rather HTTP/2 server push)?
There is nothing (yet) specified formally for browsers to ask a server to push resources.
A browser could figure out what secondary resources needs to render a primary resource, and may send this information to the server opportunistically on a subsequent request with a HTTP header, but as I said, this is not specified yet.
[Disclaimer, I am the Jetty HTTP/2 maintainer]
Servers, on the other hand, may learn about resources that browsers ask, and may build a cache of correlated resources that they can push to clients.
Jetty provides a configurable PushCacheFilter that implements the strategy above, and implemented a HTTP/2 Push Demo.
The objective of server push is that the server send additional files (e.g. javascripts, css) along with the requested URL (e.g. an HTML page) to the browser before the browser knows what related files are required, thus saving a round-trip and improve webpage load speed. If the browser already know what resources are needed it can request with normal HTTP calls.
We have recently fixed a nagging error on our website similar to the one described in How to stop javascript injection from vodafone proxy? - basically, the Vodafone mobile network was vandalizing our pages in transit, making edits to the JavaScript which broke viewmodels.
Adding a "Cache-Control: no-transform" header to the page that was experiencing the problem fixed it, which is great.
However, we are concerned that as we do more client-side development using JavaScript MVP techniques, we may see it again.
Is there any reason not to add this header to every page served up by our site?
Are there any useful transformations that this will prevent? Or is it basically just similar examples of carriers making ham-fisted attempts to minify things and potentially breaking them in the process?
The reasons not to add this header is speed performance and data transfer.
Some proxy / CDN services encode the media, so if your client is behind proxy or are you using a CDN service, the client may get higher speed and spend littler data transfer. This header actually orders proxy / CDN - not to encode the media , and leave the data as is.
So, if you don't care about this, or your app not use many files like images or music, or you don't want any encoding on your traffic, there is no reason not to do this (and the opposite, recommended to).
See the RFC here: http://www.w3.org/Protocols/rfc2616/rfc2616-sec14.html#sec14.9.5
Google has recently incorporated the service googleweblight so if your pages has the "Cache-Control: no-transform" header directive you'll be opting-out from transcoding your page in case the connection comes from a mobile device with slow internet connection.
More info here:
https://support.google.com/webmasters/answer/6211428?hl=en
I'm keeping the backend API as a separate project from the frontend HTML5 app consuming it. I'm using Yeoman for the frontend development. Yeoman runs on localhost:3501 and the backend on localhost:3000. When I make the API request from the browser (using AngularJS's $http), I hit the same origin policy:
XMLHttpRequest cannot load http://localhost:3000/venues. Origin http://localhost:3501 is not allowed by Access-Control-Allow-Origin.
AFAIK, same origin policy should kick in only when making request across different domains. Why is it whining when we do a request from localhost to localhost (albeit to different port)?
How can I make this work and will this cause problems in production?
The ports also count for cross domain requests, therefore http://localhost:3000 and http://localhost:3501 are 2 different domains (from browser's point of view).
If you need both applications (client and backend) to run on different ports, consider using http://enable-cors.org/
According to W3C, "An origin is defined by the scheme, host, and port of a URL", so the different port is causing your problems.
Two possible solutions:
CORS (Cross Origin Resource Sharing)
Using JSONP requests
Both would require changes to your backend (I'm not familiar enough wo. CORS would probably mean the least changes to your frontend (I think AngularJS supports it out-of-the-box).