跟讀練習: What is Event Driven Architecture (EDA)? - 通過影片學習英語口語

What are event-driven architectures? And how can  they enable real-time user interactions, pluggable microservices, and extensible streaming and data  analytics? My name is Whitney, I'm on the Cloud team here at IBM. Before I dig in on event driven  architecture, specifically I'd like to talk about system design, more generally and specifically  the reactive manifesto. The reactive manifesto is a set of community-driven guidelines that are  intended to give a cohesive approach to systems design. So, at the core of the reactant manifesto  we have, we want our system to be message driven.

And so, specifically asynchronous messaging.

And then also, we want our our  system design to be scalable, and we want it to be resilient.

So this implies distributed systems or Kubernetes.  So by scalable we mean that we want the hardware used to expand as the workload  expands and contract as the workload contracts and by resilient we mean we we don't  want any single point of failure and if the system does fail we want it to be able  to tolerate tolerate that elegantly so with this foundation in place we should be able to build a  system that is responsive responsive is the value on top of that we can expect  a system that is maintainable so it means it's easy to change easy to fix a bug  or do an update and we want one that is extensible so we should be able to expand the system easily so now that we have the core of what we  want our systems designed to look like now let's talk about let's level set  here and talk about like what is an event so an event is a statement of fact an immutable  statement of fact about something that happened in the past so let's give that some context  we'll talk about a retail application so a retail application has a checkout service and that checkout service is going to want  to communicate with an inventory service a shipping service and a contact service so with the messaging model if the inventory wants  to know what the checkout is doing the checkout will send a message directly to inventory  to let the inventory know a checkout happen and another one directly to shipping and directly  to contact so with messaging we have targeted delivery um not only that but inventory  can maybe send a message back to checkout it can be a conversational  delivery conversational messaging and then finally let's talk about where that  message lives before it's sent it's it's lives on the host service or on the producing  service when it's received it lives on the um it lives on the host service until it's received  by the consuming service or in the case of a message broker it might sit on a machine in  between but regardless the the data persistence is transient now let's compare that to eventing so if  our checkout service is producing events it's producing an event every time a checkout  happens without any concern to which service is consuming that and so with that we have a  system that's highly scalable so we have um three services here interested in the checkout  but even if we expand it to a hundred services listening to check out that wouldn't affect  what the checkout service is doing in any way not only that when we have all the checkouts  in a row we come up with an event log and it can be replayable  so we have a stream history and a history of everything that that  every checkout event that has happened and then our data persistence depending on how the  cluster is set up it can be set up to be immutable it can be there forever as long as you have the  physical hardware to back it up so we have um instead of having transient data persistent  so we have uh permanent or possibly permanent not immutable but permanent data persistence e and  t um so this is the foundation of our event driven architecture so we have a resilient a reactive  responsive system design and then that message driven element is is event messaging specifically  so when we go into our proper event design the first thing we're going to consider are our event  producers so we might have a web application that produces events and we  might have a a mobile application and let's say we also have an edge device that's  producing events so uh internet of things device and all of these are producing  events into our system so so this piece of the system here that  has all the events running through it this is called the event backbone so what does this system enable well first of all we have our event logs being  produced by all the producing applications so the event logs can do a few things first it  can trigger an action so if we have an internet of things device let's say our internet of things  device is a robot vacuum cleaner and when we turn on that robot vacuum cleaner it spins up a pod  on the infrastructure and that pod is a function as a service so function as a service is a modular  piece of code often associated with an edge device that doesn't exist on a server on the server until  it is turned on and that's when it is spun up so that's called serverless function as a service  is sits on top of serverless infrastructure but the point is that event triggers an action  of the with the robot vacuum cleaner to spin up bounce all around the house make everything  squeaky clean and then when the function is finished running it powers down and perhaps  that'll cause our function as a service to send another event back into the system to let  it know that the the function is complete another cool thing about the event log is it  can be used to help optimize and customize data persistence so if our web application is our maybe  it's our retail application that we talked about before and it's doing a stream of checkout events  well our inventory service can be reading from that and it's keeping track of a local data  store that's keeping track of the inventory so our inventory service will consume from  that data stream from the web application it'll modify the local data and then it'll produce  again into the event backbone and so this new stream is giving the most current inventory  to any other application in the system that wants to consume from it and so while we're  talking about data another thing that happens we can save all of the data that comes through the  system so all the raw data or all of the transform data here at the end once it goes through the  system we'll save it all in a huge data lake and this is what's going to power  really data-heavy applications like ai for example and the data lake is  accessible from anywhere in the system that any service that you let consume from it so another thing that our event-driven  architectures can enable are a stream processing so this is built on top of  the apache kafka streams api so let's say that our mobile app is producing  let's say it's a music streaming service and every time let's say up until your in your whole  life up until this point you've been listening to only miley cyrus and britney spears but today you  decide you know what i want to see what jay-z is about and you turn jay-z on well immediately  your streaming app is showing you kanye west and lauren hill and giving you ideas about what  you might want to listen to next so that stream processing is a really powerful and what gives  the gives real-time calculations like if it's a ride share app you um that you know exactly how  far away your ride is but it can also be used to implement business rules and business policies  and it can be used for real-time analytics and then let's also talk about with  our system the communications layer so specifically it's called pub sub communication for publish subscribe so we could have any  microservice a containerized microservice and just like we talked about before with it being  extensible we can very easily plug it into we can write a new microservice and plug it into our  system and immediately have it start consuming from any stream and producing to a new stream or  even producing to a stream that already exists and none of the other applications need  to be concerned with it at all so it's highly plugable not only that but because  of the stream history we talked about before when you plug it into the system you don't have  to have it start consuming from right this moment it can be consuming from as long as from the  beginning of your application if you've been saving your stream history for that long so  as long as that stream history persists when you plug in a new system it can consume from  either the old data or can be consumed from the present moment so with that we talked about  the major benefits of event driven architecture that sit on top of the reactive manifesto ideas  for system design with eventing at its core