How timeouts can make or break your Backend load balancers

[@hnasr]

If you enjoyed this, check out my Introduction to NGINX on udemy grab a discount coupon here www.husseinnasser.com/courses

[@Springfielde]

It's amazing how you understand relations and responsibilities between OSI layers. Great video, thanks!

[@sagartyagi2450]

Can't thank this guy enough. Been following his vidoes lately, and in this one, I was able to figure out almost 85% of the stuff he mentioned about networking aspects without even him mentioning about them. Thanks man🤘, becoming a better engineer.

[@DarkKnight-ree]

Learned more in the past month from you than my entire career 😅

[@BaishaliGhosh13]

There are so many courses explaining backend concepts. I’ve always found a big gap in those when I think of how does this happen at a network level? Thank you for explaining these concepts at a “on the wire” level. It was not easy to understand the nginx documentation but you explain the rationale behind the existence of the different types of timeouts so well. Thank you for the great work you do.

[@hnasr]

Thank you!! Nginx doc is not ideal unfortunately it was designed for low level network devs who already understood those concepts..

[@saurabhsharma7123]

Dude. You explain sooo well. Love from India.

[@hnasr]

❤️

[@surajitlovetomakefrenz8270]

@Hussein Nasser- Hussein just to understand something. In the case of a SLOW LORIS attack, where an attacker slows down the packet transfer to the reverse proxy, how does that affect other TCP connections between the reverse proxy and other clients? Does it mean because the reverse proxy is trying to wait for the packets for a single request, it cannot accept requests from other clients during that time period?

[@SergeyPopovX]

Concerning headers timeout, there might be the following reasons to have them. Let's consider the reverse proxy that does routing, rather than load balancing, that is making decisions based on the start of request, where the request will eventually be forwarded to. This decision should be rather quick as it's the bottleneck for the whole infrastructure. Or another example: if authorization is based on headers (cookies and bearer fall in this category), we want kick out strangers as soon as possible, at the same time giving more forgiving timeout for actual users.

[@hnasr]

Pretty neat use case thanks for sharing Sergey,

[@SergeyPopovX]

There is also a pretty obvious use case when the payload is a stream that does not suppose to end (WebSockets, Server Sent Events and streaming video content lies in that category). In such scenario the criteria to keep a client connected is client keeping up with the stream (you mentioned from another point of view --- through timeouts). See --- there is another way to bring time into consideration --- a speed. Whenever you may trick timeouts because it's discrete, there is not that much you can do with the requirement for the average upload speed. Also it can be easily computed from bytes uploaded and seconds elapsed.

[@kiragod722]

Me : I am going to learn something new about backends . Learns a new word also .. 🌄🌄🌄🙂🙂 Just wow to hussein nasser

[@megazord5696]

My man! Keep the content coming!

[@debjitchattopadhyay599]

Hey Hussein, what is the timeout that can be configured for upstream service response timeout? Most of the configs I found were between successive packets instead of whole response timeout?

[@dineshkinibailoor340]

Could you just make an video about 499 status introduced by nginx? I recently noticed this while loading bulk data on the page and server was taking prolonged period of time to process the data.

[@kamleshshete6948]

Awesome sir 😊😊 keep it up.🙏🙏

[@deshkarabhishek]

105k subs

[@antoniopafundi3455]

Great video

[@Mokstar35G]

Hussein, having 1 tool for each use case, it's good idea if you have the money, teams and capacity to manage this multi vendor strategy,, but think when you, as a company, need 10 different vendors for 10 different tools and definitely not helping to have the simplicity you are mentioning. Probably having 1 tool to do 10 use cases (in any cloud/ environment) as Nginx do, you can simplify drastically (architecturally speaking) and also, the include the cases where tools doesn't work in different environment, so need extra subscription. As you said, Is all depend on the goal or target.

[@hnasr]

Valid point👍

[@Mokstar35G]

@@hnasr Thank you Master!