Kong User Calls, 12 Oct 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Introducing Kong Gateway 3.0 [User Call]

Description

Kong Gateway 3.0 is now out and it's bringing a ton of useful functionality for you all to use. It's adding OpenTelemetry tracing, advanced secrets management and a brand new router in addition to the usual performance increases.

Michael Heap (Director of Developer Experience at Kong) takes you through the deprecations, breaking changes and brand new features that you need to be aware of in Kong Gateway 3.0.

A

So hello, everyone, my name- is Dalia I work as a community manager here at Kong. I would like to welcome all of you I'm, so happy you're joining us.

A

um So today we have a very special presentation for you: long awaited uh what's new in concave way, 3.0 um I know a lot of you have been asking for a long time when we're going to have a user call about that. So I'm very, very happy to have Michael here today uh who is going to walk us through what is new in concave way 3.0, so you will see a q a function on the bottom.

A

Please ask all of your questions there and we'll take them after my question is with his presentation: um yeah Michael. Please take it away from here.

B

Thank you very much Dalia uh good morning afternoon and or evening wherever you're joining us from um you might notice that this says Kong Summit 2022.. uh This is a presentation uh that I gave alongside uh Guan lundai who's. One of our engineering directors at Kong Summit in person at The, Palace Hotel in San Francisco I, wanted to make sure that you got the same experience that the people in the room got um so we're going to be using exactly the same slide deck exactly the same demos.

B

And hopefully you will have a couple of questions afterwards.

B

So I'm Michael I'm, the one on the right, uh unfortunately cannot be here today, but a lot of the the content um in this slide. Deck comes from the work that he and his team have been working on towards can't get with 3.0 foreign to cover today. So we're going to split this session into the first part is where we'll run through all the flagship features uh that we shipped in concave with 3.0 and any live demos, and then I'll. Take you on a bit of an architecture, deep dive.

B

Now con Gateway 3.0 is our biggest release. Yet there are a ton of features in there to make your life easier and I wanted to take this opportunity to call out a few of them and really the ones that stand out for me as a user of con Gateway myself.

B

Now we won't have time to go through all of these today, but you can find more information about all of them at docs.conhq.com,.

B

Luckily, for you, we do have time to dive into some of them, um we'll be covering Secrets management, a shiny new Kong manager, interface, built-in, open Telemetry support and a new router implementation, which is not only more powerful and more performant. This also written invest, which gains us some internet points and speaking of the router. That's where we're going to start so, let's Dive In.

B

I'd say that the new router is the most crucial feature that we introduced in 3.0.

B

I guess the first question is well. Why is it? Why do we need it like? Didn't? We have a root of the works? Just fine and the reason we needed a new router is that as Kong is deployed in um companies that scale faster and faster and faster and the scale of which Comcast to operate, gets bigger and bigger. Some of our larger customers have over 10 000 routes configured and that causes issues when you're changing Kong's routine configuration in production.

B

So to solve this, we've built a new router from the ground up to handle the challenges that we've been presented with.

B

Our previous router was highly optimized for runtime, which means that once it had been rebuilt once it had the the mapping of routes to upstreams- and it was very very very fast.

B

But we made an assumption when we built Kong's router initially, where we thought people would create their roots and then that would be it and they change infrequently kind of every couple of days, maybe once a week, but that's actually not what ended up happening.

B

A lot of customers are changing routes dynamically, like on demand in response to other events with the old router.

B

um Imagine that you've got more than ten thousand boots and you change one you've got to wait for the whole routing table to be rebuilt, but with the new router that ships in three dollar zero it becomes feasible to only incrementally rebuild the routing table, so only update the routes that changed.

B

My design goals were to have compound performance, so we wanted better rebuild time but comparable runtime performance, and we also wanted to start thinking about the future. How can people Express more complex routing needs? How can we have a unified implementation for our lower based control plane for Coco, which is our next Generation go-based control? Plane I'll talk a little bit more about that later and Kong manager, which is a web UI.

B

I mentioned that we, besides think about expressing these new routing rules on the left. You can see the the old style which still works, which is still the default and can't go with 3.0, but you can also opt for this new DSL based approach and like the DSL based approach, because it means it's very explicit.

B

It means you don't have to understand that on the left when it says X my header Fubar, but that's evaluated as an or and the way you specify multiple headers X, another header and X, my header, that's an and right, that's tribal knowledge, that's just how con Gateway works. But if you take a look, take a look at the bottom on the right. You can see the explicit X. Another header equals black and X. My header is through I buy. It just makes the the routing rules very, very explicit.

B

Whether these uh this router works is we build up two runtime indices.

B

um One is a hash table and we're working on building a red, extreme approach as well, and you can see that um if we do a get for slash login, it actually matches two routes: Route 2, which is the exact match, http.path equals login and Route 1, which is any path that starts with a slash. That's what that little cover and then equals means it's a prefix operator. So it says any path that begins with this.

B

The router will evaluate root 2 as the strongest match and route to that Upstream.

B

This happens, benefits for the new router uh reduced router rebuild time. um That's key, and this is the issue that customers have been hitting with lots and lots of routes. So we wanted to optimize for that, and we've also seeing increased runtime performance when reaching requests for common deployment sizes.

B

With 10 000 routes, we actually reduce P99 latency from 1.5 seconds down to 0.1 seconds.

B

That means that 99 of requests are completing almost 10 times faster during configuration, changes on large deployments, and we can express those complex routing needs in a single route rather than having multiple, which means it's faster again, if we can express them as a single route, that's just one route to it and lies and process, rather than having three four five routes to do. The same thing, foreign.

B

We mentioned the um single view to use it everywhere, and this is a great proof of concept and this is building an upper an AST. So this is how the router actually works internally, um but you can change this. So obviously the protocol is http um I. Actually, let's put a new one in the path starts with Slash Foo and the headers start with hello and the headers end with world, and it's a HTTP or https request.

B

um So this this request uh will match it any request where the the X3 header starts with hello and ends with world on both HTTP and https requests.

B

um I've done a lot of talking, so I'm actually going to show you this.

B

um I have a concave wave running. What I'm going to do is just create a service and create a route, and you may have noticed that this is not something that you're used to seeing this Expedition. This replaces methods and paths and everything else when you're creating a route and 201 created, we can see the com Gateway accepted this. To try this out. You can set root of flavor to expressions in your kong.com file, foreign.

B

We can see that this works because it starts with hello and ends with world and that matches the the expression that I configured so starts with hello ends with world.

B

This also works, even though it's got one two, three, four, five, six in the middle: it's because it starts with hello and ends with world foreign ER ends with world and that returns to 404 from the router.

B

um That is a super quick demo of how the the new router works and how you can configure it.

B

Performance is super important. It's why we rebuilt the router and that's why I'm excited to introduce you to open Telemetry, which is a brand new plugin in con Gateway 3.0, open, Telemetry or autel is a collection of tools apis and sdks to instrument collect and Export Telemetry data from your software.

B

You can use it to understand what your software is doing, how it's performing and where time is being spent during execution. Now you can see this um Trace here.

B

We can see that over the course of a request, um 21 microseconds were spent in the router of two milliseconds, like that's pretty quick and most of our time was was taken by the mail limiting plugin I'll show you how to actually configure this in just a moment.

B

Understanding how con Gateway is performing is a key part of any deployment and we're excited to announce extensive tracing support in both Kong open source and Enterprise. This is the first time this is available in Concord, open source gone other days when a customer complains that Gateway is too slow. They don't know why going forwards that can never open, Telemetry and see that detailed breakdown of where time is being spent by con gateway to help increase performance and throughput.

B

There are two ways to get started. You can use the out of the box. Open Telemetry plugin, send Auto spans directly to any compliant backend or to an open Telemetry collector I've been testing Kong with honeycomb. That's all you'll see later and it just works.

B

You can also integrate with third-party platforms um such as honeycomb or datadog, using the plugin development kit. um Sorry using the open, Telemetry collector and get all of your traces in there. So I've used auto collector with datadog, and it looks great as well.

B

We also provide a plug-in development kit to enable our customers to instrument their own custom plugins and understand where time is being taken um in those custom. Plugins.

B

There are a few non-limitations today, unfortunately, that pdk is only available for lower plugins um and it does not support all observability providers. Functionality right only supports tracing you'll need additional plugins um such as the deadstock plugin for advanced use. Cases such as metrics and logging right back to the live demo.

B

So what I'm gonna do I've created just an open, Telemetry demo service and a root.

B

And I've just configured the open, Telemetry plugin, so all you need to do is say, use open, Telemetry, give it an endpoint and any authentication. Don't worry. I will be rotating that key as soon as this call is done and now, if we start making requests up, that is the wrong command sites, just slash Aussie root, so localhost 8000.

B

Let's see and I'll just make a couple of requests and what con Gateway is doing now is generating all those uh traces all those spans. In the background and if I go back to my browser.

B

And go to Honeycomb fingers crossed they are showing up. So you can see that gets Otis hello. Let's make that a bit bigger that can go into the span. You can see that uh con Gateway took 425 milliseconds again. The router was super fast and the most violations it was added by the Upstream itself. This is a great way to say: like concave was taking too long.

B

It's usually not it's usually the Upstream, but now we can prove it using open, telemetry foreign.

B

We're flying through this like this is just how easy it is to get started with these new plugins like create service like you know how to create a route and then just enable the plugin.

B

Call manager uh for those of you that aren't familiar con manager is a UI that we provide to help customers configure and monitor their com deployment and Kong manager, 3.0 ships with a new design and improved user experience, including my tool tips.

B

We've revamped Kong manager in 3.0, making core Gateway entities such as service route, consumer and plug-in configuration more intuitive understanding, co-in entity is configured, is now presented front and center, rather than being hidden behind that view config button, so the left was a common manager on Gateway 2.8 on the right is 3.0, we've also updated empty States to make it easier to get started. We've added tool, tips to explain configuration options, and it's now harder to accidentally delete a service or a plugin screenshots aren't the same.

B

I have a running con manager instance we'll just go and click around it. So you can see what it looks like.

B

um So one of the the key things to look at here is these Christmas lights makes it very easy to see at a glance how your services are performing see here that most of my requests are 2xx, which means they're all care 2.41 average over it. That's okay, ish as well I can see when my license expires. uh I'm using a development license, so I've got about 2 600 days left. Yours will tell you how long you have left on your license and then we can click into a workspace and then choose a service.

B

Let's choose service staging. You can see at a glance how this service is configured. What the configuration is, what protocols it listens on, what ports we could do this for roots I've had a lot more colors. You can see this response to get post, delete and options which hosts and which paths it responds on one of my favorite things which I didn't know until consummate when guanlan presented this section um was, you can actually click. These I can enable and disable plugins.

B

Just by clicking this.

B

um This little enable button.

B

If we go ahead and create a new route previously, you had to pause uh paste uuids into this service, but now it's just auto complete it filters down as you type it makes it really easy to configure your services and routes, foreign.

B

Manager, I highly recommend clicking around in your own environment and just seeing what kind of information you can get out of your own services and routes.

B

So, as I said, I didn't know that you could click the enabled and disable buttons like that was one of our best key Best Kept Secrets until she shared it with us all at Summit and speaking of Secrets. I would like to introduce you to the final Flagship feature that we're talking about today, which is Secrets management.

B

Now secrets management allows you to store sensitive information securely in external vaults.

B

um You can use environment variables for open source or you can use AWS, Secrets manager, Google, Cloud, Secrets manager and hashcode vault. If you have an Enterprise license and then you can have Cong access those values at runtime, a secret is a piece of sensitive information that Gateway requires to learn. It could be your database password. It could be a secret key through a plugin.

B

It could be an SSL certificate, no matter what it is, you probably don't want it being publicly accessible to anyone that has access to your config by storing sensitive values are secrets. You make sure that they're not visible in plain text through the platform, either in com.conf in your declarative, config files in your logs or even in the Kong manager. Ui like today. If you've got a secret value in your proxycash plugin to authenticate with redis, anyone can go in click edit plugin and see that secret, using Secrets management that goes away.

B

All this is the Vault reference.

B

In addition, using Vault Secrets allows you to have perv environment values for development staging and production, while still using the exact same declarative configuration file, because each environment can point its own Vault. That uses the same path but different values.

B

We're shipping with three Enterprise volts, AWS, hashcode, Vault and gcp, and an open source environment variable driver.

B

You can reference Secrets using an intuitive format, and it looks like this on screen on 0.2. So volts is the protocol and then the the driver, so it could be AWS or hcv or gcp, and then the path to the secret and Kong will automatically resolve those secrets of runtime ensuring that they're not available for people to read um and once again this works a lot better when you can actually see it work.

B

So, let's go on to the demo.

B

um So I just need to export My Vault address I, have Vault running on localhost.

B

What 8200, nope and I didn't mean to put that at the beginning, so I can do. Vault gets heavy.

B

Here we are, this is my testing, my memory uh volts KV get Mount secrets uh Venice. Now there there is no secret there. So let's go ahead.

B

And set a secret.

B

And it's put not set, so this is where my memory is failing me there we are so now. If we get that secret, uh we can see that in slash reddish password, we have the value demo.

B

If we start a ready server, we're doing require pass demo and just to prove to you. It works I can't access it without Authentication.

B

But now it will show me that we have no keys.

B

So at this point, I have my ready server running I have a vault with the secret key. The last thing to do is actually create um a service and a plug-in config that uses that so once again, I'm just going to create a service create a root starts with Slash SM. So.

B

Localhost 8000.

B

Totally less Sharp as well.

B

Let's do that again: curl Dash I localhost 8000, Secrets management, hello, there. We are and there's nothing in these headers right now about caching.

B

So let's go ahead and add the the proxy caching plugin here I'm, adding it it will cache any get requests it's using redis to Cache I'm running con Gateway within Docker, but running redis on my local machine. So I'm saying use my Docker host and here's the key part I'm saying look in hashcode Vault in the red is Secret in the password path, foreign.

B

B

So if I go back and make that request again, this time we should see X cache key. We can see that the cache status was missed, but if I go back to my registry and list out the keys, we actually see the item just there.

B

If I make my requests again, it comes back super fast because proxy caching is enabled and the cash was hit.

B

So this is an example of how you can use Secrets management to secure your redis credentials, but, like I said you can use it in anything in kong.conf and any plug-in marked referenceable. If you check out the the Kong plugin Docs see the word referenceable, it means that you can use a vault reference in there.

B

All right, thus, all of the the demos we'll just do a quick, deep dive through the architecture and then I will come on to the the questions that you you've all been asking great to see. So many questions.

B

uh This is beauty on the inside stuff. uh This isn't necessarily something you would see there today, but it's super important for keeping kongate where performance and stable the first thing we shipped is a new time and Library I'm. Using the time moves on avoidable. We have to do a job every once in a while or after a delay.

B

um The HTTP log uh Funk plugin, for example, flushes the access log on each request and what used to happen is open investor created a fake request and the cover routine and each of those fate requests takes up a certain amount of memory and it occupies a connection just like a normal request.

B

Open investor uses a linked list to manage all those cover routines and, as the number of core routines increases, it becomes more expensive to read and write the link table um because it's all in complex to once the time task expires.

B

We pick a timer from the timer pool to run the task um and when the timer pool is empty, like we just let that task queue expire and as a result, uh even if we keep starting the timer, the number of fake requests and cover routines in the systems will be relatively stable, which means that we don't receive a big performance impact thanks.

B

We tested on a modern server without any load for each timer.

B

um If we did this in a real world deployment with my compute heavy jobs, this becomes less pronounced, but you can see the number of requests per second, with the original timer uh decreases. Logarithmically like it drops very fast, but with the new timer Library, it's a linear decrease um and the number of requests per second stays quite High.

B

This is probably a good time to talk about new events Library as the the time the library is heavily centered around events in the old model, a worker would write one or more event items into a shared dipped into shared memory and then post a message to notify all of the other workers using the event system.

B

Then every worker starts to pull events to try and get message from social memory. At the same time, one of the workers will receive say a router change request.

B

It will go to all of the worker nodes to make sure that the change request is propagated to all of them and the way it works is it uses the shared dictionary of events so with 32 workers um and only 13 items in the show date like when a worker is reading from the sheddict it locks that, and what happens is everything becomes locked and performance drops.

B

Using lower Vesta worker events, we use a timer and sleep to actively pull events from this show date. Now, we've rebuilt this and can't get with 3.0. Instead of having that sleep on pulse sleep and pull where everything's trying to access the shame shared dictionary with lockings we've introduced a new pub sub system, um we have a an event broker, which is a service run in one and only one nginx worker and then an event publisher and an event subscriber that come with any nginx worker.

B

Events uses a Unix demand socket um for communication, so it it posts messages across that domain. Socket and all of the the other workers are listening um on ninja next thread for messages to respond to so, instead of constantly pulling they're just waiting for messages to be pushed to them,.

B

There's a big impact for this. It's in the first case um lure Vesti events and the invested worker events are about the same rest, events being the old implementation and work with events being the new one, but in the second case uh well, instead of running 1000 requests every half a second, where we run 10 000 requests every 0.1 seconds.

B

Worker events is significantly better. It's about 40, more performant than the old locking messaging system.

B

This is the final beauty on the inside change, which is lmdb, which stands for lightning memory mapped database manager. This is a transactional embedded key Value Store in our previous way of using storage. Was that shared dictionary, which is a global blocking event?

B

That means there can only be a one nginx worker reading from a shed Dixon at once, maybe so those numbers 1 to 13 one worker per slot, and this shows up as less than ideal scalability issues. That's the number of com workers increases and the workers are waiting their turn to read with lmdb. The database structure is multiversioned, so readers come down with no locks, writers cannot block readers and readers don't block the letters.

B

It supports a multi-process concurrent rights and reads um so you can submit all your transactions safely and it supports persistent and is fully asset compliant, and that means that when we're writing data, you know it's not going to be lost.

B

Lndb allows Kong to run faster with less memory usage. It lets you handle very large configs with ease, but it will help us bring even more advanced features. We're just getting started.

B

It'll allow us to build in things like incremental sync, for hybrid modes and through leveraging lmdb's transactional support.

B

It's the foundation, that's going to support our vision of making hybrid modes, the prefer methods for Kong deployments when you push config the request per second serving from a data plan, is notably better with lmdb.

B

So this goes back to uh when we're doing things like a routing update, uh we can see that the number of requests per second stays fairly consistent with lmdb and is about a hundred thousand requests per second faster than the shed data clutch.

B

That was all the beauty on the inside changes. I showed this slide at the beginning of the session, and we ran through these four features.

B

Which means we didn't even touch upon these ones, and these are five Flagship features too, like this slides. This nine are just the the highest impact, the highest visibility things, and there are dozens of smaller features that we ship to. If you want to learn more head over to docs.comhq.com Gateway and take a look at the release notes, thank you for your attention. So far, I think it's time to turn it over for questions.

A

Thank you so much Michael. We have a few questions in the Q, a chat.

B

Cool I will take them in order in Q a um hi Mohan. uh What is Coco great question, so Coco is the next Generation Kong control plane the idea behind it being the um nginx open best day or a wonderful for proxying uh traffic, but it's not really ideal for building a control plane for handling. All of that, um the con Gateway configuration that you need to do so. Coco is next Generation control plane for deploying Kong in hybrid mode.

B

If you want to give it a test drive today, you can head over to Kong incubator. Let me pull that up right now and share my screen again, because this is something that you should all see.

B

So this is incubator.comhq.com.

B

You can go down and see Coco. It adds a little bit of background. It's the Next Generation control playing, gives you a comparison with the existing control plane and gives you a quick start. um Do not run this in production today.

B

um We want you to to keep the tires. Try it out.

B

um This is a technical preview. We make no guarantees. I will show that this is what's powering con connect or Cloud offering so it is fairly battle tested at this point like Kong, is using this internally I think the phrases drinking of one champagne rather than dog food and got one products but yeah. That's what cocoa is next Generation control plane being used by con connect, and you can give it a go by heading over to incubator.comhq.com.

B

Okay and next up Jay, can you tell me more about the com.log dot serializer changes um honestly, I cannot uh not without a little bit more investigation.

B

um If you head over to the con Gateway GitHub repo and there's an issue um tag me I'm at M Heap, um we can take a look together.

B

Merge sometimes I have pain deciding when I need to use con, Gateway or service mesh Kuma. Can we say that con Gateway 3 has moved me away from service meshes a little more? It really depends on your use case. If you're using the service mesh for observability and the open, Telemetry Point con Gateway is enough, then perhaps if you're looking for secured inter-service communication east to west traffic through mtls service mesh is probably still a good choice.

B

It really depends what you're trying to achieve.

B

Merge again on the topic of open Telemetry: does it cause a slight delay between client requests and the services no more than any other plugin um we're talking microseconds?

B

This is because the the spans are recorded internally and then flushed um to the Upstream um asynchronously. So we don't have to wait for the Upstream service to respond, and so the Upstream open Telemetry service to respond before sending back the response from your actual Upstream.

B

Is Kong manager available for the open source version? It is not, however, you can get it for free, so we offer two distributions: uh Congo open source and Kong Enterprise Kong Enterprise is available free of charge, including Kong manager, and the only difference is that the source is not available, but you can run it for free forever.

B

um So there there's no commitment. All you have to do is install The, Conch Enterprise package and you will get con manager for free.

B

uh Kang Kong support both 3.0 and 2.0 routes. At the same time, um old routes in the 2.0 router engine and The Roots by 3.0. It cannot Denis. You have to make a choice when you start con Gateway by setting that root of flavor. Your options are traditional, which uses the two Dozier router traditional compatible, which uses 2.04 map with the 3.0 router for the increased route to rebuild time or you can set router flavor to Expeditions and use the new routing DSL. But it is unfortunately an All or Nothing decision.

B

And up if a plugin interacts with postgres then does open Telemetry plugin show the DB interaction time. Yes, it does. It will show you the DNS time to look it up and how much time the the plugin took. If you use the lower pdk and wrap your database queries as separate spans, they will show up in open Telemetry as well.

B

uh Session are there any new plugins that support integration with Splunk, not Splunk, specifically, but I, believe you can send logs to Splunk using all existing plugins um either the the file log plugin with the Splunk, forwarder or I, think the HTTP log plugin will also work for Splunk, but I would have to double check that one.

B

uh Creating custom, plugins and deploying getting hdp node is a manual process, unlike old conversion will be able to create custom plugins using car manager, slash the admin API. If not, is this in scope for upcoming releases? uh You are right, it is currently a a manual process.

B

um Is any scope for upcoming releases. I know that the team has been discussing uh streaming, plugins from the CP to the DP, but no commitments on that whatsoever. It's more of an exploration phase today, but is it possible and then we will see if it's possible and decide if we're going to ship it and when we're going to ship it.

B

In case of using Kong as an Ingress controller for kubernetes, is it possible to use the DSL based routing I believe so because the the Ingress controller laps can't get away? You will need to make sure that your um CP and DP pods I started with the con gluta flavor environment variable such as expressions, but otherwise you should be able to use DSL best routing foreign.

B

Plug-In for Kong, three uh honestly I, don't know, feel free to there's an issue on GitHub I talk to your account manager. If you have one.

B

uh Some of vanan is runtime groups feature only available in SAS. uh Yes, today, it is only available in con connect. uh So for those that aren't familiar, runtime groups is a feature that gives you virtual control plans.

B

You can have a single pen of glass to view all of your services, all of your routes, all of your plugins, but isolate config completely. So each runtime group gets its own Services Roots consumers plugins everything.

B

It works great in a Federated model, where teams on their own configuration and use the collaborative config and deck to push things up. But yes, unfortunately, it is only available in SAS. Today.

B

uh Where can we find the permissible differences between 2.0 and 3.0? Do we have one link? uh Yes, we do that will be docs.conhq.com Gateway, slash change, log. Everything that changed is on that page.

B

um I was wondering if you can explain, Dynamic plug-in ordering a bit um sure so Dynamic plug-in ordering is a new 3.0 feature, and previously all com plugins had hardcoded priorities saw the sorry just music off.

B

um All the plugins had high quality priorities, so the authentication plugin would always run before very limiting as an example.

B

Now this is great because you might want to limit based on the consumer. But what happens if someone is trying to brute force your Authentication? You might want to run that limiting first based on IP.

B

So what plugin you're doing does is let you you specify a plug-in name um so config dot before where you could say, run the rate limiting plugin before you run authentication. You won't be able to let me based on consumer anymore, but you will be able to limit based on IP and protect um your database from all that traffic. Whilst people are trying to brute force your authentication methods.

B

Now there are examples in the the condocs.

B

If you go to dogs.com hq.com click on Gateway, Kong, Enterprise, Dynamic, plugin ordering you'll see a concrete example there of how to use it or you can head to conchhq.com forward, slash blog and look for a blog post on Dynamic plugin ordering from Victor one of our developer advocates that walks you through exactly that use case where you have to put their limiting before the Authentication.

B

Jay, do we suggest to upgrade from 2.03 to 3 directly I'll go from 2.0.3 to 2.801 to 3.0?

B

um The answer is neither we're about to release 2.8.2 and con. Gateway 2.8.2 enables blue green deployments for those migrating to concrete with 3.0. So my recommendation is to go from 2.0.3 to 2.8.2 when it's released. It should be in the next week or so, and then to contact with 3.0.

B

Final question: where can I read more about websocket support for Cong? If you head over to docs.com khq.com click on get way, websockets.

B

uh You can learn more there, alternatively, head to the plugin Hub, and there are two websocket plugins there um request validator and frame size limit, so the request limiter runs on every websocket for them, I can't remember which one they said. uh Let's do uh frame size limit. First, it runs on every websocket for them and if the frame is too big, it will refuse to send that frame to your upstream or to your client. It works in both directions.

B

You can configure the size differently and when whether it's going Upstream to client or client to Upstream, so you can protect your consumers from oversized rooms. In addition, you can use the the websockets validation plugin. It will run every frame and if the message does not meet the Json schema that you provided.

B

The frame will be rejected again, Works in both directions, client server and server to client.

B

um Richards merch, we showed almost 10 new features of these running this components, so I can remove the one that I don't need from Congo, where 3.0 or I can get one and use it on con Gateway 2.. They are not um everything that I showed today requires con Gateway 3.0.

B

uh Some of them are running as plugins, for example, up in telemetry. uh You could remove that from your build if you wanted to build custom images, but it doesn't make much sense to Because unless you enable open telemetry, um there is zero performance impact, but yeah to use anything that I've shown. Today. You have to be running con Gateway 3.0.

B

That is all the questions.

A

um And I see one more in the chat from manzat he's asking about Cassandra.

B

I see yep. uh Thank you for your question. Manzat uh Cassandra is deprecated in con Gateway 3.0 and will be removed in com Gateway 4.0. uh The alternative is to switch to postgres.

B

Most of our, our customers are using postgres for a very high scale deployments and you can set it up with active active deployments so that you have a highly highly available cluster um yeah. If you search for ha postgres active active that will give you all the information you need.

A

All right, thank you, so much Michael I think that we're oh wait, one more! Oh, no! That's Apple, explaining I! Think! That's all the questions! Let's give it a few more seconds, any last ones.

B

Excellent well, thank you for your attention. Everyone uh I've enjoyed it. Thank you for all the questions. There were lots of great questions in there um Dahlia back to you.

A

Yeah, thank you so much Michael for the great presentations and for answering so many questions thanks everyone for attending um the recording will be available tomorrow on our YouTube channel. If you want to check it out again and yeah, thank you again. We hope to see you at our next event have a good evening good day. Bye-Bye.