Conf42 Golang 2023 - Online

Go and Redis, better together!

Video size:

Abstract

This talk will demonstrate how Go and Redis can work well together and are (probably) built for each other - Simplicity, Speed and Developer-productivity being their driving principles. I intend to help simplify the learning curve and reduce friction for developers in their journey with Redis and Go

Summary

  • Abhishek Gupta is a developer advocate at AWS. databases go and open source in general are his focus areas. Don't hesitate to connect with him if you have questions or just want to chat about tech in general.
  • This talk is geared towards folks who are looking to get started with Redis and go. In many ways, so far, I've seen that Go has become this lingua franca, or language of the cloud. Hopefully some of the tips and tricks will help you simplify that learning curve.
  • Redis and go have a few things in common. One of the things which really, really stands out to me is the simplicity. The goal is to give you simple components to help build complex things and hide the complexity of the language itself behind a simple facade.
  • Redis is nothing but a key value store, where the key can be a string, or even a binary. Redis is primarily an in memory system, but you can configure it to be persistent or rather persist that data to disk as well. You can get offerings from pretty much any cloud provider, big or small.
  • The common one is string, which can be used for something as simple as storing a key value pair. sorted sets are like this big brother to set, and they make it possible for you to implement things like leaderboard. Redis just feels like an extension with all these data structures.
  • To build something useful you need to have clients. Redis has a very, very rich ecosystem of clients. Goredis is by far the most popular client out there. I'll be using this client for most part of this presentation.
  • Readygo is a fairly stable, fairly tenured client in the go ecosystem. But the API for this particular client is a bit too flexible, at least in my personal opinion. And then there is another client called Ruidus. In terms of the ecosystem and where go stands, as of now, it's very close to the redis and Java client.
  • Redis can be used to connect to a local redis server. You can also set a TTL or expiry. Here we take a look at how to work with strings. Keep this fairly high level and fairly quick.
  • edge set is actually very flexible. It can take key value pairs like what you can see here. Can take a string slice, it can take a map built. Can actually take in a struct as an input as well. This is a MUch easier, simpler, and a natural way of representing your objects.
  • Using an s add function to add a bunch of tags to a set. If I want to check, hey, is this sre tag included for conf fourty two? And in this case it's not. So sets very, again very simple to work with conceptually from a client library as well.
  • Pipeline is one of my other favorites in addition to sharded pub sub and few others. Think of it as batching. You can send a bunch of commands in one go, and you can get the response in a single request. I would encourage you to actually try it out.
  • There's this data structure called hyper log, and it's this sets on steroids. It can store a lot of data, much, much more than set. Run this for yourself and see the kind of memory usage which you notice in your application.
  • A use case around worker queues and how they make that possible. Start a redis server and a mock SMTP server. These worker instances are going to distribute the tasks of sending emails amongst themselves. This can be used in many use cases.
  • Use real time technology called websocket to create much, much more powerful solutions. Show us an application of the pub sub capabilities. Pub sub is about broadcasting messages to a channel. Users subscribe to that channel and receive messages. Use combination of WebSocket and redis pops up.
  • The connection mode that you use in your client application will depend on whether you're using a standalone redis setup or a redis cluster. If you don't use the right mode of connection, obviously you'll get an error. Make sure that you have the right configuration in terms of the VPC and the security groups.
  • You can either scale redis up or out, right? Up, down and in and out. But you need to put a lot of thought into scaling out horizontally. You should opt for a redis cluster based setup instead. And once you have scaled out, don't forget to make use of those replicas.
  • Sharded pub sub was introduced in redis seven. It helps scale your traditional pub sub even further. It's very similar to the original subscribe and publish methods or functions. Something to be mindful of and make sure that you're using the sharding pub sub in your Redis cluster topology.
  • Here are a few resources which you might find interesting. A link to the GitHub repo for the redis client as well as the Discord channel where you can interact with the core maintainers. If you want to build practical solutions or try out advanced use cases, give it a go.
  • All right, that's that. I would love to have your feedback. All you need to do is take out your phone, scan this QR code, or you can use the link directly here. Once you finish submitting your feedback, you should get a download link to this presentation as well.

Transcript

This transcript was autogenerated. To make changes, submit a PR.
Hey folks, welcome to this session. Thank you so much for tuning in. I've been a part of Con 40 before, but not the Golang edition, so I'm delighted to be speaking at this wonderful go conference. My name is Abhishek Gupta. I work on a bunch of things, but databases go and open source in general happen to be the key focus areas in my role here as a developer advocate at AWS. My social media activity is kind of lukewarm, I would say, but I'm fairly active on GitHub because that's where I keep pushing all my stuff that I'm building. But don't hesitate to connect with me if you have questions or just want to chat about tech in general. Now this talk is geared towards folks who are looking to get started with Redis and go. Or perhaps you're already experienced with both these topics and in that case this might be a good refresher. Now I have a very, very simple agenda for you. I'll start off by setting the context about redis and go in general, followed by some hands on stuff and wrap up with some gotchas and things you should watch out for. Now, I cannot possibly go over or cover everything in a single talk, so I'll leave you with some resources at the end as well. And by the end of this presentation, you should have a fairly good overview of Go and Redis. And this includes things like the client options you've got. How can you implement common use cases and patterns? And as you start your journey with Redis and go, hopefully some of the tips and tricks will come in handy and try and help you simplify that learning curve and help reduce some of that friction. As you might have guessed already, I love Redis and I've been a fan and a redis user since a few years, but I'm not the only one out there. Redis is not brand new anymore, but I still like to refer it to it as relatively young because it was first open source in 2009, so that makes it around 14 years old now. But I'm pretty confident that there are a lot of innovations we are going to witness in the years to come. As far as Redis is concerned now, since it was released, it did not take Redis too long to win the hearts and minds of the developer community. As you can see, as per DB Engine's statistics, Redis has been actually topping the charts since 2013. That's just four years after its release and on the stack overflow annual survey. It's been voted one of the most loved databases for many years in a row now. And I also remember seeing something like Redis being one of the most pulled docker images on Docker hub as well. So all in all, pretty popular. And I also love Go. And in many ways, so far, I've seen that Go has become this lingua franca, or language of the cloud, so to speak. It powers many, many cloud native projects. Apparently 75% of CNCF projects are written in Go. And if you're using perhaps Docker or Kubernetes, Prometheus, terraform, you're actually using Go indirectly and the power of Go. And in fact, there are many, many databases that have been written in go, like influxdb or etCD and vitas etc. And it is not just that Go caters to, excuse me, go caters to a wide variety of general purpose use cases as well, be it web apps or APIs or data processing pipelines. Think about infrastructure as code, SRE DevOps solutions, and also building Clis. In fact, that happens to be one of the very, very popular use cases for Go. And no wonder Go has become a very popular language. Again, pointing to this, I love referring to stack overflow. Now, as you might think, go is down at the bottom, but if you notice carefully, it is the only statically typed language in that list after rust. And of course there's c sharp, but that's down the list. And in fact, this is from 2022. If you look at data from 2021 down to 2018, you will notice that Go has actually maintained its top five spot pretty consistently. Now, with that, let's talk about redis and go. They have a few things in common, but in my experience, one of the things which really, really stands out to me is the simplicity. Now, Redis is a key value store, but the values can be any data structure that you sort of see here. These are all data structures that we use as developers day in, day out. Think of lists or a set, or a map or a sorted set. So that kind of gives you the feeling that redis is just this simple extension of these core programming languages constructs that we use day to day. And with go, it actually comes in various forms. This ranges from the excellent yet simple go tooling or the comprehensive standard library, easy to use concurrency. And sometimes it's not about what is there, but what is not there. And it's this notion of not bloating go with unnecessary language features. Just to cite an example, it took a while for go generics to be added to the language. Now, I'm not trying to trick you or trying to sell you into thinking that go is simple, or for that matter, any programming language is simple, right? That's not the case. But with go, the goal is really to be able to give you simple components to help build complex things and hide the complexity of the language itself behind a simple facade. And there are folks who have actually explained this in a much better fashion and in much greater detail. And I would really encourage you to watch this talk by Rob pike, one of the co creators of Go. It is from 2015, but very much applicable and to the essence and spirit of go, so to speak. All right, so let's quickly get a very, very high level view of redis, or rather an introduction to redis before we move on. So at its very core, like I alluded to earlier, redis is nothing but a key value store, where the key can be a string, or even a binary for that matter. But the important thing to note is that as far as the value is concerned, you can choose from a variety of data structures. Now, these are data structures like strings, hashes, lists and sets and sorted sets and so on and so forth. And by the way, contrary to popular opinion, redis is not just a cache. In fact, it's a solid messaging platform as well. And you can configure redis to be highly available by setting up primary replica replication that's asynchronous in nature. Or perhaps take it a step further with something like a redis cluster topology. And Redis is primarily an in memory system, but you can configure it to be persistent or rather persist that data to disk as well. And there are solutions like Amazon Memorydb that can actually take it a lot further by actually making it possible to use redis as a persistent primary database. Now, since Redis is open source and wildly widely popular, you can actually get offerings from pretty much any cloud provider, big or small, or perhaps even run it on kubernetes, on cloud, on prem, or even in hybrid mode. The bottom line is that with redis, if you want to put it in production, you can be rest assured that there will not be any dearth of options for you if you were to choose to run and operate it. Now, what you see here is a list of code data structures. So the common one is string, which is really simple at first, but actually quite powerful. It can be used for something as simple as storing a key value pair, to implementing advanced patterns like distributed locking or rate limiting and even session storage and so on and so forth. Now, a hash is very very similar to a map in Java or a dictionary in Python. And it is used to store object like structures. Perhaps you have a user profile or a customer information, things like that. Well, a set behaves as promised. It helps you maintain a unique set of items with the ability to list them, to count them, execute operations like union intersection and so on. So it's very, very close to the mathematical version. And sorted sets are like this big brother to set, and they make it possible for you to implement things like leaderboard, which is a very useful sort of solution in areas like gaming. So for example, you can store player scores in a sorted set. And when you need to get the top ten players, you can simply invoke the specific sorted set and the specific command and get that data. And the beauty is that all the sorting happens on the database side. So there is no client side logic which you need to apply here. And lists, right. They are a very, very versatile data structure as well, and you can use them to store many, many things, but using it as a worker queue is a very common use case and a pattern. And there are popular open source solutions like perhaps you have often used them as well, like sidekick or celery, that already support redis as a back end for job queuing solutions. Redis streams, they were added in redis five, if I remember correctly, and that can be used for streaming use cases. And there is also ephemeral messaging with pub sub. It's a broadcast, it's a publish and subscribe mechanism where you can send messages to channels and their consumers which receive from those channels. There's this geospatial data structure as well, and a really, really cool data structure, data structure called hyperlog log, which is can alternative to a traditional set. And it can store millions of items while actually optimizing for data storage. And you can count the number of unique items with really high accuracy. And like I mentioned earlier, Redis just feels like an extension with all these data structures. It just feels like an extension to everyday programming language constructs which we as developers use day in, day out. Moving on. So Redis has a relatively simple protocol, but you can always use it from a terminal perhaps. But to build something useful you need to have clients. And Redis has a very, very rich ecosystem of clients. And that applies to go as well. So let's explore that. Now, what you see here is a bunch of selected go clients. There are others as well, but these are the key ones. So what I'm going to actually talk about is, or rather use heavily in my sort of demos and discussion today is Goredis. So Goredis is by far the most popular client out there, the go client for Redis out there, and it has what you'd expect from a good client. It has good features, decent documentation, a very, very active community. And to be honest, Redis was already super popular and widely used, and it recently actually moved under the official Redis GitHub organization, which is just an icing on the cake, to be honest with you. So like I said, I'll be using this client for most part of this presentation, so I'll not spend a ton of time on this. So moving on to another client named Readygo, which is fairly stable, fairly tenured client in the go ecosystem, which supports all the standard redis stuff, as you might expect, the redis data types, the features such as transactions and pipelining, and it is actually used in other Redis go client library as well, like Redisarch or Redis time series go client. But that being said, the API for this particular client is a bit too flexible, at least in my personal opinion. Now some may like it, but for me it does not feel like a good fit when I'm using a typesafe language like go. Right? And like I said, it's just my personal opinion. But apart from this, the biggest drawback to me is the fact that this client does not support the redis cluster topology, right? So that's a big let down to me. And then there is another client called Ruidus. I hope I'm pronouncing this correctly. Forgive me if I'm not. So this is, I'd say it's a relatively new client, again at the time of recording this talk, but quickly, it's very very quickly evolving as well. So it supports the RSP three protocol features like client side caching, and supports a variety of redis modules as well. And as far as the API is concerned, this client adopts a very very interesting approach. So it provides this do function which is very very similar to the Readygo client which you just looked at. But the way it allows you to create commands is via this builder pattern, so this helps retain the strong typing, unlike the readygo client which you just saw. And to be honest with you, I haven't used this client a lot, but it looks like it's packed with a lot of useful features, so I can't really complain too much now. But for those who are looking to sort of dive deeper into perhaps the performance numbers and things of that nature, go to its website. There is a benchmark comparison with the Redis library, which you might actually find pretty interesting. Now let's take a look from a holistic ecosystem perspective, right, not just from a go point of view. Now, for folks who are not actually using go with redis, or maybe are not aware of it, the popularity of the redis driver might actually come in as a surprise to you. So Java workloads are a huge chunk of the redis ecosystem, and Jedis is the bread and butter client when it comes to Java application. But I was actually very, very surprised to see redison, this client called Java client redison, topping the charts. And it is followed by the node client and then Python and back to Java. And again, the point is sort of not to do chest thumping based on GitHub starts. I mean, they matter to an extent, but this is meant to give you a sense of things, right? And of course, another thing to note here is I was looking at repositories with more than 10,000 stars. So when I was prepping for this, the PHP redis client was very, very close. Actually it had around 9.6 thousand, 9.6k stars. So that's that. In terms of the ecosystem and where go stands, as of now, it's very close to the redis and Java client. So anyway, let's switch gears and move back from stars to the ground reality, as they say. It's time to see some code now. So I'm going to switch to the ide. Okay, so let me start off with the basics. So what you're looking at is some sample and demo code here. So I'm going to walk you through this code and I'll keep this fairly high level and fairly quick because I have to move on to some of the use cases as well, and patterns. All right, so first off, if you look at this, I hope this is big enough. If you look at this init function, we need to obviously connect to redis, right? So we have init function for that where we use the new client function to connect to a local redis server here. Now this is actually not going to establish the connection, and that is why we ping, we issue a ping command, which is a first class redis command, to make sure that we are indeed connected. This is a good practice, it's not Mandatory, but it's good to fail fast, as they say. Now let's take a look at how to work with strings. So here, as you can see, I'm doing a very, very simple set. I'm just using this set function and setting this key to a particular value and then using the get function to extract that same value. But you can also set a TTL or expiry. Right, so here the TTL was zero, which means that this key is never going to expire. You'll have to delete it yourself. But if I were to set a TTL, perhaps a two second TTL in this case, and after perhaps 3 seconds, if I were to try and search for this key, I will not find this because Redis is going to automatically delete this key after its expiry time. So it applies to any key in redis for that matter. Okay. All right, so that's it in terms of strings. Like I said, I'm moving quick here, very, very high level. Just giving you a sense of how to work with the client. Just give me a second while I wrapped up. All right, so moving on to the hash. Now, as you can see, I'm using h set to set a bunch of key value pairs, right? So this is the name of my set, the name of the key, and I am adding these key value attributes, the name of the website, the URL of the website, the date. Right. So edge set is actually very flexible. It can take key value pairs like what you can see here. It can take a string slice, it can take a map built, it can actually take in a struct as an input as well. So if you see here, if I were to scroll down real quick, I have declared this Conf 42 struct. And notice this redis tags which I'm using. Right. So internally, the go Redis library uses this to map to the name and attributes of the hash which it creates and deals with. So I can then create an instance of this struct like I did here, and then pass it on to this edge set method. So this is a MUch easier, simpler, and a natural way of representing your objects and then sort of letting the library deal with the hash level, doing the et set and the get, and then populating your structs accordingly. Right. So it's much more convenient than using a random set of key value attributes, which is kind of static, to be honest with you. Moving on, let's look at how to use a set again. As you might expect. Right, no surprises here. I'm using an s add function to add a bunch of, this is the name of my set Con 42, and I'm adding a bunch of tags to this set. In this case, I have duplicates here as well. So if I were to then use the scard function to print out. How many tags are there for conf fourty two, it will give me the result, it will only give me the unique number. Of course, in this case it will be nine. And if I want to check, hey, is this sre tag included for conf fourty two? And in this case it's not. So it's going to return false and so on and so forth. Right? So sets very, again very simple to work with conceptually from a client library as well. And pipeline, pipeline is again one of my other favorites in addition to sharded pub sub and few others. Because from a conceptual point of view, redis is primarily a request response system. You give it a command like set get, and it gives you back a response, right? So there is this back and forth going on, but there are use cases where you might not use it. Perhaps you want to bulk load some data, and using the pipeline feature you can actually send it multiple commands. Think of it as batching, right? And you can send a bunch of commands in one go, and you can get the response in a single request. So aws, you can see here, I'm using a redis pipeline. I've got a redis pipeline object from the client object here. And then what I'm doing is inserting 1 million keys. And then all I do is execute the pipeline. Now this is a very, very important step. If you do not invoke the exec function, your pipeline will not get executed and you'll be wondering like what the heck happened? So make sure you call the exec function and yeah, that's it. And I would encourage you to actually try it out. Try how it works by using say, client set right versus a pipeline to insert in this case 1 million items, or maybe even more, and just compare the time it actually takes, right? You'll be surprised with the speed and efficiency of pipeline here. And other one I want to talk to you about is you saw how to use a set, okay? Now there's this data structure I mentioned earlier very briefly called hyper log, and it's this sets on steroids, so to speak, okay? So it is a very, very efficient and enhanced version of set, because what it can do is it can store a lot of data, much, much more than set, with almost a finite storage, right? So if you were to run this code, and I'm not showing it to you, just in the essence of time, but if you were to perhaps add 1 million, perhaps there is a use case you have, right? You want to track the number of unique views and you're tracking ips, right? So if I were to add ips to a set and do the same thing to a hyperlog log using this PF add commands and if you were to inspect within redis as to how much space these keys take, you will see a huge amount of difference when I can. My tests with the set, it took around 50 mb of memory, approximately 50 mb with the hyperlog log. It was, I think it was around zero point mb or something like that, right? You can try this out, try adding 1 million entries. I'm using a faker library just to generate random fake IP addresses. Run this for yourself and see the kind of memory usage which you notice in your application. Okay, now moving on, let me show you some things in action. So a couple of use cases which I want to demonstrate to you. The first one is this use case around worker queues and I told you lists and how they make that possible. So I'm going to run an application to see that in action. Okay? Now before that I'm going to start a redis server here. First I'm going to use docker to make it very simple here. Excuse me. So just give me a second. All right, so I'm being to start this server again. All right, so my redis server here is up and running in docker nice and quick. Now another thing I'm going to do is start a mock SMTP server, right? Because the application I'm trying to show you is a very, very typical simple web application. Well it's not a web application but a use case of a web application where someone registers and then they suddenly get a welcome email, right? So that's what I'm trying to display here. So let me start off another container here for the fake SMTP server. Excuse me, this is not what I want to run. Let me copy over a lot of copy paste going on. In the interest of time, don't worry about this, it should be up and running pretty soon. So I have my SMTP server running. Now what I'm going to do is run my worker application, okay? So let me hop over again and make sure I am in the right folder and then run this command go run in the worker folder. Worker go. So what you will see here is a worker has started and it has published its unique id. I've just given a unique id for each random unique id for each worker. You'll see why I've done that pretty soon. And I'm going to run the producer application as well here side by side. So run producer, producer co. So what it's going to do is just randomly submit data here and I'm going to show you the code as well. So that gives you more clarity. But what's happening right now is my producer is generating data within redis and my worker applications are actually working. Okay, what are they doing? They're sending emails. They are sending emails to the SMTP server. So let's quickly switch over and take a look at that server. Just give me a second real quick while I make this happen. Yeah. All right, we have this server here and if you see this is my SMTP mail server and you will see a bunch of emails flying back and forth. Okay, so this is what our worker application is doing and if you notice here it says processed by this particular worker. So right now we just have one instance of the worker running. So all you will see is this particular worker instance, right. This particular message. Okay, that's all well and good. Now let's add another worker instance. So that's the magic of the list data structure, right? Scaling out horizontally. So it's very simple. I'm going to run another instance here. So go run another worker. And there you go. So now this has a different id notice it ends with 1119, starts with two a. So what's going to happen is now everything is going to happen automatically. These worker instances are going to distribute the tasks of sending emails amongst themselves, right? So now you have a balance. You have two workers crunching all these new users who are registering on your website. So let's go back to the email client which we were using and just refresh it and you'll notice that there are different workers. So if you see this is the second worker which we had started and if I were to show you perhaps another one, this is the same worker and yeah, this is the first one which we had, right. If you notice closely. So like I said, work distribution, this can be used in many use cases. Of course I showed you a very simple one of website users registering and then this asynchronous process starting where these notification emails are sent out. Right now let's look inside redis what is happening and also take a look at the code. So I'm going to go back to my terminal here and fire up the redis cli real quick to log into my local redis instance. And let's take a look at this key called celery. Okay, so there is this key called celery and let's look at what type it is. Right. So I'm going to type celery. So it's of type list. Now actually what's happening is behind the scenes, the application which I'm using, it's this go library which is using the celery protocol to do all this magic. Okay. And internally it creates a list called salary, and that's where all my jobs go. Okay, so this is the list. What's the length of this list? Right, so 139, there are around 139 tasks at this moment. Now this is going to change very, very quickly. And these are the number of tasks which are there in the list. And that's what our worker applications are actually processing. So, cool. Let's take a look at one of the items. So I can use a list command l range. And perhaps let take a look at the first item here. The first item, the task in the list. So this is the body which you see, it's actually JSon encoded. Okay, so let's make this simple for ourselves and decode this. I'm going to open another terminal here. So going to do is throw in this and I'm going to do a base 64 decoding. And this is just to show you right, as to what's going on behind the scenes here. So if you see, this is the JSON payload, this is the celery compatible payload, which is being sort of shared back and forth, right? So this is the name of our user here that the producer application is sending. And at the end of the day, the worker is actually churning this data and sending emails, fake emails in this case, of course. Right? So with that said, let me actually show you the application itself very quickly. So here is the worker which we have. So if I were to quickly show you the important stuff here, right, so the most important part here is this salary client object and how I register the function and to a specific named task here. Right? So this is all going to shape the salary message. And the function which I'm being here is the send email and it's just using the SMTP libraries to send emails. Very simple, nothing magical going on here. And if I were to look at the producer application here. So all it's doing is sending messages, in this case email addresses, to the same task, and that's why the worker is picking it up. Right. And just fyi, this is the library which I'm using. It's called go celery. Right. So yeah, just a very quick example of how the list capabilities are actually abstracted behind the scenes here. We are not using the native list commands directly, but we are using this library instead, which uses redis lists behind the scenes to do all this magic and task distribution and all the good stuff. Right. So let me close this and move on to another example. I'm going to show you an application of the pub sub capabilities. So at a very, very simple level, pub sub is about broadcasting messages. So you publish messages to a channel, and there are consumers which subscribe to that channel and receive messages as long as they're online and as long as they are subscribed to that channel. Now this is very powerful, but it can actually be combined with this technology called real time technology called websocket to create much, much more powerful solutions. And it also actually helps overcome some of the limitations which WebSocket has because it's a very, very stateful protocol. Now, I've built a very, very simple, yet canonical chat example, very simple chat server. So let's see how that works before we go into the code. So I'm going to quickly go into another directory here and start off this application, my chat server here, and you will see that my chat application has started and let's connect to it and exchange a few messages like you would do in a typical chat application. Now, what I'm going to do is a Websocket client, a command line websocket client called AWS can, and I'm connecting to my server as this user one. Right, it's a sample user. Okay, so this user is connected. Now what I'm going to do is connect another user, but this time I'm going to change the name and call it user two. Now, these users can exchange messages over by using the combination of WebSocket and redis pops up. So I say, hello there, this will go off. Hi, how are you? It doesn't matter what you're typing, this is just for demo and so on and so forth. And I can keep on adding users to this. And this is being to ensure that all the users get all these messages. So another user called as user three, like hey there. Yeah, so you kind of get a sense of where this is going. Okay, now I'm going to show you the code. Just focus on the key parts like I did before. So let me close this and this and go over to my chat application. So there are a few things which you should note here. Of course, this is, like I said, very small application. But again, we are making connections to the redis client, so on and so forth, setting up our routes here. But the important thing is, let me scroll down here real quick, the part where we create the websocket connection. Now minded, let me just scroll up a bit. I'm using this client called WebSocket, part of the larger gorilla system of libraries, group of libraries. So I'm using this websocket client, I create this Websocket connection, and what I do is associate that username which I had entered with the actual connection object itself. Right? So this is a Websocket connection here. And when a message comes in via that websocket connection, when that user sends hi. Hey. Hello. What I actually do is publish messages, that particular message, to a redis channel. Okay? So I'm using this function called publish here. And then what happens is when a message is actually received on that channel, I have this broadcaster function which is running within a go routine. So what it does is whenever a message is received on this channel, it broadcasts it to all the connected users and it does it through that websocket object, that session object which we have, right? So it kind of goes over that map where I had initially stored that mapping of user and websocket connection, and it simply broadcasts, right again, like you see, we are combining websockets and redis channel to overcome some of the constraints of WebSocket and making this sort of simplified but useful chat application. This is fully functional, by the way, and if someone were to exit, that user is going to be deleted from the actual map within the code, and everything is going to work as expected. So of course you can try this out. But like I said, very, very simple example to demonstrate a couple of very, very common patterns and use cases. One of the worker queue pattern and the other of a pub sub real time applications combining pubsub and websocket. So we only have so much time. I did not cover a lot of things, to be very frank with you, things like geospatial or redis streams. But I hope you have a fair idea of the concept of redis data structures. How do you use some of these commands, the sets, the string, the set, the list and so on and so forth within the Godadis client. So with that, let's move on to the next part of this talk. All right, back to the presentation. Now let's go over some of the most common things you should watch out for when working with redis, especially when you're getting started. Now, there are actually great resources around this topic, and I happen to write a blog post which I will share at the end of this presentation. Now, in order to do anything else you first need to connect to the redis server. Kind of obvious, but believe it or not, sometimes that can be challenging. So let's take a look at a couple of low hanging fruits in this area. And this first one is kind of obvious. Now I'm going to call it out anyway because this is one of the most common getting started mistakes which I often see folks make. Now, the connection mode that you use in your client application will depend on whether you're using a standalone redis setup or a redis cluster, or perhaps even redis sentinel if you're using it. For example, if you're using the go redis client, you will need to either use the new cluster client or the new client function depending upon the topology. Now most Redis clients draw a clear distinction between these types of connections which you're making, but interestingly enough, the goer redis client has this option of a universal client which is a built flexible right. It gives you that flexibility of using one function to choose and connect to any redis topology out there. And this is actually available in the go redis v nine, if I remember correctly. Right, so v nine onwards. And if you don't use the right mode of connection, obviously you'll get an error. But sometimes, depending upon which redis instance you're using, which provider and so on and so forth, the root cause actually might be hidden. It might be hidden behind a generic error. So you have to be watchful to avoid frustrations and those challenges in getting started. And now this is also something which trips up folks all the time, especially when they are working with redis on cloud, which is very common to be honest with you. Now say for example, the example here which you see is with Amazon elasticache where your redis nodes are in a VPC. Now if you have a client application deployed to a compute service like AWS lambda or Kubernetes cluster in eks or can ecs and so on and so forth, you need to make sure that you have the right configuration in terms of the VPC and the security groups. And of course this might depend, this is going to vary depending upon the compute platform which you're ultimately using. But something to bear in mind, and like Samantha rightly said, you need to read the please, please read the documentation for your specific provider once you're setting all these things up, including of course AWS. Redis is actually wicked fast and a single server which actually it's going to actually take you pretty far in your journey. But also know that Redis is primarily an in memory system. I've alluded to that before. So you will actually want a bigger machine with more ram and memory if you want to handle more data. And at the same time, it's important to know that Redis is a single threaded system. So what that means is you cannot throw more cpu cores at it. It's probably not going to benefit you a lot. So that's where you'll have to think about scaling. And as is the case with most systems, you can either scale redis up or out, right? Up, down and in and out, right? That's the way scaling up and down is. I would say relatively simpler. Not simple, relatively simpler. But you need to put a lot of thought into scaling out horizontally scaling in and out, especially if it's for a stateful system like a database, like redis, for example. So you have to be very, very clear about the type of workload you're optimizing for. So perhaps you are looking to solve a problem where you have a lot of reads, so you can choose to add more replica nodes to the existing primary redis node. Or perhaps you want to increase write capacity. Your applications have a lot of it's write heavy workload, so you will find yourself getting limited by the primary replica node. And you should opt for a redis cluster based setup instead. So you can increase the number of shards in your cluster. And this is because the primary nodes can only the primary nodes can accept writes and each shard can only have one primary node. But overall, once you have this setup, this has the added benefit of increasing the overall high availability of your redis setup in general. And what you see here is an illustration from elasticache documentation, but this is applicable to any setup, whether in the cloud or on prem. Doesn't really matter, right from a conceptual perspective. And once you have scaled out your please, please don't actually forget to make use of those replicas in your code. That's what matters at the end. And I'm calling this out because the default behavior in most redis cluster clients, including the Redis CLI and the go client as well, is to redirect all the reads to the primary node. Now, if you have added read replicas to scale traffic with the default mode, they're actually going to sit idle, believe it or not. So you need to make sure that you switch to a mode called the read only mode, wherein it's going to ensure that the read replicas handle all the read requests and they are not just passive participants. Okay, so in the go redis client. You can set this using this attribute called read only. You just need to set it to true. Or you can also use this attribute called route by latency, or randomly routing it to any node and activating these is going to also activate the read only mode. So be careful about that and make sure that you use these or activate these configuration options. Now, instead of using consistent hashing like a lot of other distributed systems and databases, redis actually uses this concept of a hash slot. Now, there are 16,384 hash slots in total, and a range of hash slots is assigned to each primary node in your cluster, and each key belongs to a specific hash slot, and thereby assigned to actual to a particular node inside of your cluster. But the thing to know is that multi key operations executed on a redis cluster cannot work if they belong to different hash slots, but you're not completely at the mercy of the redis cluster. Here it is actually possible to influence the key placement by using something called hashtags. Of course, we know hashtags from our social media applications, but you can ensure that specific keys go to the same hash slot. So, for example, if you're using storing customer data with id 42 in a set, storing orders for a customer in a set, and you have perhaps the customer profile information in another hash called customer 42 profile, you can actually use these curly braces to define which is the specific substring which your client application uses for the hashing, right, to allocate to that specific hash slot. In this case, our keys are customer 42, right? So we can be confident of the fact that the profile and the orders for a specific customer are going to land up in the same hash lot, right? So something to definitely bear in mind and also apply to your client application code sharded pops up. This is actually one of my favorite ones, to be honest with you. My favorite features this was, again, relatively, I wouldn't say recent, but relatively recent, because this was introduced in redis seven and it actually helps scale your traditional pub sub even further. But you'd actually be surprised by the fact that many folks actually don't know this. And I would really, really encourage you to check out the documentation that covers this very, very well. And the usage is very similar. If you have used pubsub in your client application before, it's very similar to the original subscribe and publish methods or functions, with the exception that you can now use those hashtags, right, those braces to the channel name in order to make sure that you can influence the shard placement so something definitely, like I said earlier, something to be mindful of and make sure that you're using the sharded pub sub in your Redis cluster topology. Now, there are cases where you need to actually execute bulk operations on multiple keys across your clusters. Now, the redis client has a handful of utilities here as well. For example, as you see this, for each shard function on the cluster client. And the key thing to note here is the fact that this is concurrent. This is all going to happen concurrently, so the library actually takes care of fanning whatever operation you specify. It fans it out to multiple go routines. Definitely helps with scalability there. And you can also use this for each primary node, so there is a function for that too. I think I've forgotten to mention that, I apologize. Anyway, so I think we have covered a few scenarios so far, but in the interest of time, I'm going to pause. There are a lot more, but I'm going to pause here, like I said, for time and sort of try and wrap up things. But before I do that, here are a few resources which you might find interesting. A link to the GitHub repo for the redis client I mentioned this before, as well as the Discord channel where you can interact with the core maintainers, ask questions, exchange ideas with the community members, and the documentation. The Redis I actually find the redis documentation to be very helpful in general. So something which you should sort of follow very closely once you're building anything with redis. And in case you're using redis on AWS, you might perhaps find some of these blog posts pretty helpful as well in terms of using your best practices and optimizing your client applications to the fullest possible. And while I am at it, while I'm sharing all these resources, I might as well do a bit of self promotion because I've put a fair bit of content out there. Perhaps when you see this in future there'll be more. And all these blog posts and content which I've put out, they all have individual GitHub repos. You can sort of try everything step by step, whether you're getting started or you're interested in learning about good practices. If you want to build practical solutions or try out advanced use cases, give it a go and give me feedback. I would love to hear more and whether there is something specific you'd like me to cover around redis and go. So please don't hesitate to reach out to me. All right, that's that. I really, really hope you found this useful I would love to have your feedback. It should only take a minute, I promise. So all you need to do is take out your phone, scan this QR code, or you can use the link directly here as well. And once you finish submitting your feedback, you should get a download link to this presentation as well. So your feedback is actually going to generally help me improve my content and at the end of the day understand your needs better. So with that, I'm going to wrap this up. Thank you once again for tuning in and enjoy the rest of your conference.
...

Abhishek Gupta

Principal Developer Advocate @ AWS

Abhishek Gupta's LinkedIn account Abhishek Gupta's twitter account



Awesome tech events for

Priority access to all content

Video hallway track

Community chat

Exclusive promotions and giveaways