Cloud Native Computing Foundation KubeCon + CloudNativeCon China 2019 (Shanghai), 5 Jul 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Intro: Vitess - Jiten Vaidya, PlanetScale & Toliver Jue, PlanetScale

Description

Join us for Kubernetes Forums Seoul, Sydney, Bengaluru and Delhi - learn more at kubecon.io

Don't miss KubeCon + CloudNativeCon 2020 events in Amsterdam March 30 - April 2, Shanghai July 28-30 and Boston November 17-20! Learn more at kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects

Intro: Vitess - Jiten Vaidya, PlanetScale & Toliver Jue, PlanetScale

Join us for KubeCon + CloudNativeCon in San Diego November 18 - 21. Learn more at https://bit.ly/2XTN3ho. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy and all of the other CNCF-hosted projects.

This session will cover the basic features and functionality of Vitess. Additionally, we'll go into the details of how to design a VSchema, which is the sharding metadata. The use case described will help attendees to map the solutions discussed to their own situations.

https://sched.co/NrrW

A

My name is Tolliver yeah Oliver from planet-scale I'm in software engineer, and this is Japan g10 is the CEO of Planet scale. So if you have all any questions, you just asked you I'm here just to look pretty.

A

So today we're talking we're gonna talk about vid tests and which is- and this is called stateful storage on kubernetes.

A

Pinghua, the testers think all the Rinne yep young, my sequel, my sequel, who uses my sequel, okay good. So maybe this is for you so anyway. Okay, oh.

A

Yeah, okay, plenty of these alright. So what is the tests? The tests? Tsushima right? It's the sharding middleware or my sequel, starting something peon I, think pintia, something and it what's beautiful. Is this massively scalable and it provides high availability like five nines availability and it's cloud native, so kubernetes, etc. How many people use kubernetes?

A

How many don't use kubernetes? No Coubertin Oh everyone's good burn, eighties, Oh, even better great so who uses Vitesse Shay the test? You know so this cash app is a square cash. That's like we seen the pay, but for us also YouTube. The test came from YouTube originally, so they use it, not Burger King and not Instagram, yes, but those like the top apps in the US and so they're using it.

A

So here are some of the key adopters.

A

How many heard of JD ding dong, yeah, so ding dong uses a lot of the tests they're the one of the biggest users in the world of the test and they have 3,000 or more databases and they have 18,000 or more tablets or my sequel instances. So you know more than just one. So many, and maybe you guys don't use slack as much but slack even them. At least a 1/4 25% have changed to Vitesse. So all the chatting is saved in the tests again before square cache.

A

Also in Pinterest I, don't know if you know what Pinterest is so Pinterest also uses it for a lot.

B

A

Advertising so yeah, so these are some of the many adopters and, as there are adopters, there's also a lot of people contributing you.

C

A

Pete pr2 and so includes lots of these companies, and they all, of course, are using the tests in production.

A

Am I speaking too fast, too slow? Okay me with you back all right, so yeah I want you to think from now on. If you think kubernetes and you want to store state, stateful storage I want you to think the tests. Okay, why is that there's?

A

Actually someone some famous an Internet person who says if you, if you want to run stateful applications on kubernetes, you're, pretty stupid, it's too hard and- and we agree it's not easy but- and the reason is one does not simply move my sequel to kubernetes, because if you just try to do that, if you're gonna run into many problems, you know your pods going down you're whatever going down. Everything goes down right. Who has a service? That's perfect, never goes down.

A

Anyone go down at all. Okay people who go down raise your hand. Only three people have rungs. Okay, then you don't need anything.

A

So just a few things about the test: the test started in 2010 and that's when YouTube YouTube, who knows what YouTube is it's like you, cool or or not, China, okay, so yeah so YouTube had started running into scaling problems with my sequel. Like many of you, they had just one super big, my sequel instance. Of course they had some backups and some replicas, but you know they could had too much traffic and probably many of you know web sites in China Vietnam.

A

All you guys have lots of web traffic, so YouTube ran into some problem and then so they needed to deal with it and they have a big apps right. They don't want to rewrite so the test since it in 2010, they're, kubernetes kubernetes was just is five years old now and the test is older than five years old so, but it was designed for Borg.

A

So Borg is the precursor the Google precursor to kubernetes, so Vitesse I think, oh, so the test was designed really for really for kubernetes as well, because it's all the same ideas so yeah production workloads. So besides YouTube, of course, stitch labs is the oldest one, so they started using in 2016 a few years ago and then HubSpot has many key spaces, or it may be databases, as you may call them, and.

B

A

Has 3000 databases with, as I said before, 18,000 ëone bhatia tablet whoa right and they have I think they said 67 shards. Now the ocean of teeth or something on their largest. They have many databases, but.

B

That's just their.

A

Largest one and then there's another company here, nozzle and they migrate even between the clouds. So you know here you know: hey I'll Union and you have Huawei Union or something and you and you want to migrate from one to another. So they do that. So you don't have to be tied to one cloud.

A

So- and this was what I thought it was coming after so my sequel before right- it's a old software, very good, but it was. They merely adopted the cloud, whereas the test was born in the cloud right, it was made for Borg, which is before kubernetes. So so the whole point is it is cloud native like it does not. It did not exist without the cloud, so here.

D

A

The the concrete architecture of the tests, how many of you guys are running apps that are connecting to my sequel, everyone yeah.

B

A

I'm gonna stand over here is that okay, yeah, okay, so the general architecture. The test tries not to break anything. You have big apps they're running lots of traffic. You don't want to break anything so before you have your app servers here right and they used to talk to a big, my sequel database right and they use my sequel binary protocol to connect right. But the tests is all of these blue purple color. So they're this part and these parts.

A

So it's a proxy to all of my sequels and under the covers for or saving the data is actually my sequel. So you guys you we can actually migrate. Your my sequel workloads into the tests actually pretty easily. So what's important here is between the app.

B

A

And what is these BT gate? These are the proxy. They speak. My sequel, binary protocol right here, okay and there's a custom protocol here.

C

A

These blue parts, and then all of these, this software, the tablet, connect to actual my sequel.

B

Instances under the covers and.

A

They only talk this way. Does that make sense so far.

B

A

So, but why would you want to use the tests? I mean some? Not everyone has you know dingdong size, workloads or YouTube size, their yoke, yuko or clothes right. So why would you want to use it I mean if you have a small workload like, so what are the benefits of using the tests?

A

So this was your DV right now, right works just fine, it's all good, but you can actually already put the tests in between your main DB and the app servers, and when you do that it provides automatic connection pooling it does deadlines like if you have very long running queries, you know if you have an analysis, it also provides hot row protection.

A

So if you have many apps trying to get exactly one row and it will stop in and pause that, while giving other people quality of service, good quality of service and also yeah enforces row count limits, so not too large queries and it can also blacklist tables like if you don't want certain applications to access certain things you can in the because it's a proxy, it can also say: oh I, don't want you to access this and you can cut off usage so even for smaller workloads.

A

It's okay to use this, because it provides benefits for your application, your application developers, sometimes they make mistakes, have bugs and then it won't take down the database for everyone else. Does that make sense? That's where smaller workloads, but then once you've put it in. You now have some bet more benefit.

A

Well, as you start growing your workload so again before you had only the single database, and maybe you had a few backups or replicas, but with your growing workload, you can you had your previous main database, the master, but but with the tests, since it can, it actually can parse all of the my sequel sequel.

A

It can actually decide to be smarter, it can do replica routing, it can decide. Oh all you're doing our select statements and decide. Oh, you don't need to access everything, so you can go straight to our replicas to actually do the selects and thus it can also do load balancing. You know we have more more connections and whatnot and you can then use Orchestrator as well. If you need high availability h8 to do failover or our promotion, so you can choose one replica. Oh I want this one to become the master and you can swap.

B

A

To the all the apps, it's invisible, so it's just all natural! So remember, my sequel protocol only happens here and then oh, this is only management, DBA access that make sense.

A

Why, as you're growing the workloads, if you have a lot of read traffic, you can add more replicas. No one knows the difference, and now you have even more traffic or too much too much data to fit on one disk or many disk.

A

So you can do a few things, and so with here, with their growing workload, I can split before I had only main DB will call it DV 1 right and you had all of the traffic going there, and so you had to have a disk, that's big enough or all of your data right. But now we want imagine, you have two big tables. You know they're growing, growing, growing, big and then they're gonna blow up in one table.

A

So why don't we split them into two different ones right, so with this you can split table one like.

B

A

In to keep it in database one and then you have another big table table to moving into database to so. Yes, you have now. You know, table 1 in database 1 in table 2 in database 2 to your application server. Because again it's proxying. You can have a unified view. You can see both table 1 and table 2 like they were all in one big, my sequel so and again, just like before each of the database, one you can have many replicas and database 2 has different numbers of replicas make sense.

A

So also you can with this proxy, you can do the space sub setting you can decide. Your application can only see table 1 and table 2 and different application can only see table 3 and table 4. So you can split that way to.

A

Alright and as you grow even bigger, remember a table 2 was here, but table runs growing, really really fast right and maybe you're a big ecommerce website like Jing dome and they decide. Oh, we. Actually. We have too much right traffic we're. We have too many orders happening all the time and so table table 1. It can be horizontally charted.

A

So in this example, here we see 3 different shards, so just table 1 we can split into 1/3 1/3, 1/3 and, and it will go to separate instances, so you can have now three times the right traffic and you can still have many more times, also the read traffic and you can control that quality of service there and whereas table 2 didn't grow very much. It's pretty constant, but still big, and it.

B

A

A different quality of service right here but again to the app server. It will all still look like one super big database make sense so far good so and then you get really really big right. um So this is again a ding dong, style 1 and they have and not just you don't but many other company all so have before it was all in one big data center or one availability zone. And now, when you have the massive workloads you don't want, you know one bomb to break the data center.

A

You you want to be able to handle the failure right, so we can flip even with a single shard in two different availability zones. So this gives you other high availability options right before it was only for the reading reason. But now, in case one goes down, we can split and have more high availability. You know here, one.

C

A

And three: yes:.

A

The test will handle a sorry, yeah yeah, my yeah, my sequel connects the actual.

D

Duplicate to the data from which, through some icicle binding right, yeah.

A

Other questions is that no answer your question: okay, cool! All right! Thank you to ten for answering so yeah, so, even for as you go, really really big, your go, you're going to need extra extra, high availability and that's the benefits here. Okay, so uh some demo time, this demo is very simple, very stupid, but everyone can run it on their own laptops and I'm going to demonstrate by running on my laptop. So this is a FEMA marketplace schema it doesn't really matter. It's got four tables.

A

My actual dev only has three tables, but whatever same idea, this was all one big main DB and we have a product table customer orders and merchants. But uh in my demo we only have less okay, so I am going to demonstrate. This I have already brought up a.

A

So this is actually my sequel. Let.

C

B

A

It bigger, can you see who's? Okay, yeah.

B

A

I, just I just brought up a my sequel. I actually brought up three instances of my sequel and behind the tests, and if you see databases, we see only a commerce database and inside commerce database, we actually see three tables here. Something called order called see order, but order customer and product okay make sense so far and.

A

And I will show you there's some data in this in each of the tables. It just selected showed all the tables okay, but uh alright. So that's just the data, the current one in the main, dB okay. So.

A

Next, you know maybe again I said table 1 and table 2 got too big and maybe want to split into two databases: 1 DB, 1 and db2. So here when we shard, we can decide. Oh all of the products go into one one database and we'll make we'll make another customer database yeah customer and orders. So will me with two of the tables from the original main DB into what a new one will call customer DB. Does that make sense? Alright? So how do we do that?

A

Good question? I have no idea. Okay, so we have some scripts here and these are actually on the Vitesse website tutorial. So you guys can get the code from Vitesse and actually just run these tutorials yourself. It's exactly the same I'm just showing you okay. So let me show you what we're doing so.

A

Sorry, if this is too small or can't read, there's.

B

These are actual.

A

Commands from the tests I'm just showing you what's in the script, it creates a key space. Key space is a database like database and named name Commerce, and it says to route all of the traffic for it. Alright, sorry, sorry, sorry, customer I'm, making a new customer database and then but all of the traffic for the.

B

A

Database actually goes to a place called commerce, the old database, the main DB, okay, so I'm only just gonna run this command here and it's done and then there's a few other things. This would the next the next command I'm running just brings up more instances of my sequel, three more instances for the tablets.

A

So here we go they're really starting. My sequel, I will now talk to wait for the time, so it's just bringing up more but of course, while I'm accessing this again, I'm I can still see all of the what's called all of the data from the original main database. Nothing's changed right as I. Wait for that, and the point is in in these steps we're bringing up new databases and then we're going to slowly move the traffic from seeing all of the data in main database to a different one.

A

But all is it's invisible to the app server in this case the my sequel, client, so.

B

A

You can see here it's bringing up my sequel and also bringing up the the little tablet that that's unattached to each my sequel and yeah. Okay great. So it also has that.

B

A

Master in the new database, and so we'll see this working. Hopefully, let's see show databases, oh look. We now have two databases before is commerce only and now we have customer database and so in a second okay before I. Do that so I'm going to go to the customer databases right and show what tables we have?

A

We show we have some tables, but so all the traffic for if we wanted to select from the customer database, it will look like it came from the other database or it will behave like it'll come from the other database, but it looks like it's also available in this up in database to the customer database.

A

Okay, sorry so from there. Oh and I can still access as before all the data, let's see.

A

Okay, so, actually now before I was getting all the data from the database, one the main database, but now my new script just gets the data from the second database, and it shows that it's available there as well, okay and now one of these scripts here is a vertical split. So a vertical split is the moving of tables from database. One. The database I can show you the code here, it's literally just one command. The rest is environment, variable that is splitting off the data.

A

So it's doing some stuff. What did it do? Who knows?

A

Let's see so what it actually did. There is in again we had new empty databases and it copied all the data from one to another and then from there continuously replicates from even if you write the to table table to in old database, it will start you will start seeing it in table two in new database, so it sets up it. Does the copy and then replication from then on as well. So that is what this command just did.

A

So still nothing different here. Actually again, we haven't moved any pointers for what data were reading, we're still reading from the old database. Even if you now, as you change your app go to read from the new database.

A

So the next step here is it visible: okay, we're going to migrate where we read the data so kind of to go back here. Yeah.

B

A

Of here right before it was pointing originally all of the queries were going here, but now we're going to actually say just for the replicas, the read-only or the reading traffic- hey. Let's, let's go read from here and not here anymore, so we can they're independent pointers to go instead of pointing from here point.

B

A

A

So the inside the shell script, it's just saying move both the read-only.

C

A

The replicas over say, but not the master yet so this is all the read traffic, so alright great select customer zero. Still look the same. Remember if it all looks the same. That's good! It's not the same, then something's wrong. So yeah still selecting no problem and then our last one is migrate. The master.

A

So my great master same command only instead of saying read, only a replica, it says master so master is you've decided to change all of your writing traffic over to your new place, ok, and by the way before we even do this. You hear I mean I right now, I have moved my read: traffic from replicas read-only right, I've moved it.

B

A

Maybe I have a problem. Actually you can still reverse, because this is a pointer, so you can decide. Oh I want to reverse and I can go back to reading this, so in case you're worried you can try it and if it has problems, just move it back, you know and then you could debug try it again, no problem so, but once you move the master to the new one. Well, now all of your stuff is there. So that's what we're about to do right now migrate, the master up pretty fast, even in production.

A

This should be a very fast switchover. You know maybe a 1 second or less delay just to propagate all of the configuration information, because, again all of the replication, all the copies have been happening. All of this time, ok, yeah Beppe, so I've migrated the master. I can still see that everything still looks. Ok, that is the whole demo by the way, not very much so so we went from one big database to splitting it in this example of splitting off tables.

A

It can do horizontal scaling and split that way to just don't have time to actually show in this example. So how does all of this work under the covers? This is actually one of the keys of the tests. We call the replication or Vitesse replication. It is filtering all of the replication. So, as you see here originally we had some tables here and maybe and you we can set up different types of composable replication whatever, and we can have, for example, the order information propagate over this way. We can. This is a horizontal chart.

A

Example right here. The this is for what's called the different shard names like half of the traffic go over here or half of the traffic go over here, so because this might have two shards split. You.

B

A

One some of the raw data and split the other, the other half, so so it's all doing filtered amounts of replication. So it's not doing everything. So all of this is composable. Oh yeah.

A

When you have multiple shards, how do you handle Joe good question? So the question is whether you have actually do you mind if I answer that at the right at the end, we're really close. If this is going too long, this generally make sense. So with with this V replication, these are the benefits you can have materialized views. You can have real-time roll-ups re sharding again. This is that before was actual examples of recharging in real time under the covers back up back filling of look of indexes.

A

So if index is a the test index, it it's how you decide which shard go to when you're looking up data or inserting data, and sometimes your primary index, it's not what you want like.

B

A

Authentication table you have a or user ID, maybe have user email. If you just want to look up where you're given email is you need to also figure out what shard so the secondary lookup index? The VIN Dex, is what that's for also. It allows for different schema deployments and data migration and yeah change notifications in general, so the the the beauty of all how all of the tests works, especially is with this V replication to do all this charting and to keep everything in sync. Does that generally make sense all right all right?

A

This is the end of the talk, I'm, just gonna, say some things and then we'll open up. The questions is that okay, all.

B

A

Yeah we have a flak channel that at Vitesse IO the TCO is where the open source project is hosted. We have a wee chat group. We you should come to us only because we chat has problems with more than a hundred users at a time or some other limits so come to us. We'll we'll add you, the we chat groups. Also, this is again was an introduction. Talk and juh 10 over here will be giving another talk a more advanced one on when we actually bring down for disaster.

C

Recovery bring.

A

Down you know, shards or bring down masters and etc. What happened.

A

Yeah I can edit this to show you that and then also we're going to have another talk about translation of documents for Vitesse right now. Unfortunately, all the documents are only in English, so today, June 1 show the boot Ahava. So it would be great if we had some translation stuff, and so, if you would come by with that, we can talk about how we were going to be doing that.

A

Also just like I'm running on my laptop, we would love for you to try bringing up a cluster on your own laptops and we can help you out with that read the code and you can read some blog posts from square cache about their deployments of the tests, all the problems and actually even Jing, don't like they just gave a talk on other problems and how they solve them with the tests.

A

So you can and heat this person here can talk about them if you're curious, I'm on slack on the Vitesse black channel, as at Tolliver and there's I, know as Twitter not accessible, but if you can access it with VPNs or whatever you can use. These I think access does that make sense, alright, so open to questions now. Sorry, whoever you are sorry, can you repeat your question?

A

C

Points when you have chartered tables vertical and horizontal right.

A

C

A

So the ideal case is you: don't do anything wrong, but you have single shard stuff like if you have a user table authentication table, you can co-locate like based on user ID, like all the user ID data for both tables, be in one jar, that's the best option, but of course, once you go across what how do you handle that? So we have what's called three modes or two modes here for cross shard?

A

Was it called cross card transaction modes? We have a best-effort where you know you where you can, where you can have each one independently transact and of course you have the default. Are the not default the two-phase commit option, so it can two-phase commit across that's for inserts, but for joins as well.

A

All of the they're only go back to the architecture diagram here here we go yeah, so you see actually that VT gate the proxy actually does talk to all of the tablets and remember it does parse all of the sequel and it actually also collects all the results from all of the different places and can and can do all your joins at that place.

A

If necessary, does that answer your question yeah, so yeah cuz, each gate goes to all the different tablets so and remember: there's not just one tablet here you can scale this I think jingdong. You guys have like a thousand VT gates or some yeah, something like that. So it's not just one. You can do many more than one. So does that answer your question? Okay, great! Any other question! Yes, I.

E

Have a month ink master cluster rabbit cap from Amazon to Alibaba, so I want to support I want to use bit ice. So can I my cry to this without harm g10.

A

Do you want to answer this? One.

F

You have a master in Alibaba and a replica in.

F

Aws and that one is running in is easy, to instance, okay, and what is it that you want to achieve.

E

F

So witness does not support multi master, because multi master is not a great architecture in our opinion, because it can lead to issues when you have rights going to both the Masters and then if there are conflicting rights. So for a given child we always send the rights to a single master. So, even if you have a multi master architecture, your rights would go to one or the other, and we then we would allow you to failover to start sending right say from AWS to the other one or vice versa. Yes,.

E

The guys we use the multi master because we have a user in China and we have a user in our China.

B

E

You know reply some, you, your China user, a Chinese user, so they want to have set our web application. Sorry I just inserted a data into the by in China.

F

E

F

E

We still for the manager, so they have to outside I need to see data above so that's.

F

Why we have, which is that you can run your masters in one cloud replicas in other cloud. You need to make sure that you have an IPSec turn out to them so that you're not sending packets unencrypted across public network yeah.

E

F

You can failover witness, will allow you to fail over from us from from one cloud provider to another and will also allow you to spin up clusters across cloud provider.

A

Yeah understand, actually we can actually talk about the sharding right. I mean you can actually shard based on your tenant. You know you are saying that you know there are some Chinese provide of users right, and so you can have them in one shard and others in different shards and remember each of the tablets each of the masters they can. You can individually select where they are. So again you can have the unified view, but again, based on this one, you can actually have a custom sharding scheme.

A

You can write your own charting mechanism, so you, if you know that user 1 is this China user user, not China user. You can shard in that way. Does that make sense? Okay, yeah I.

G

Want to ask about because I see that do you moving out moving out table table to the database, so I want to understand so what what they need to be chained in the application to the. If I say we have the version one collected either by with typos, but we have the version two collected attaboy correct the.

F

Typo asked the beauty of it is right as long as you're in the name of the table, he is hiding. If the name of the table does not collide between ki spaces, your application doesn't need to know your application. Just can continue to write to VT gates and litigate will send those rights to the correct database where that table exists. So.

B

F

As both ki spaces do not have the same table, then, if there is the same table name then you have to, then you have to specify it in the table.

F

So if DB one had table 1 table 2 and you moved it to db2 after that split happens, your.

G

App doesn't need.

F

To change at all.

D

If I'm very concerned about the load latency, so why how we we test can help you I won't comes of the change, a key latency, so.

F

Transaction latency at high right map volumes is what you're worried about right. Yeah.

B

F

Business itself adds a little bit of latency to every query or every transaction, because you have to do if you go back to the architecture diagram. There are two network hops that it has to do right. Your application is connecting to VT gate from litigate it's connecting to VT tablet and from way to tablet it's connecting to my C quality.

F

So it adds about four five milliseconds of latency to every query, but because you can horizontally charge, you're right QPS on every master is a lot less and that's why your latency ends up being a lot less than it would be. If you were writing on a single master.

A

B

A

It, it may add a little bit of latency to any single query, but it will increase your throughput by quite a lot. Is that answer your question? Okay, any other questions.

A

All right, if not thank you very much for coming again I'm Tolliver, and this is g10 and we're from Planet scale. We do the tests things.

F

A

May have a few, but not too many, unfortunately, anyway, if you have any other questions, you'd like to talk in private feel free to come over, probably outside I think the next talk is coming. Okay. Thank you very much.