GDIs in der Cloud

yes welcome to the third

it is amazing so again pds and that and one shoves us now in addition a little something about micro tell service and how the whole works together and when it makes sense and maybe not you are welcome thank you very much yes i'm a shovel I am part of the jaguar development teams i'm there my trainer we do

some frontend above all things so far, for example, integration of supplies in Angola but also have more and more quasi micro services to build and this also almost directly to finish the cloud

[Music] and oh yes the funds that you have today

who can see at the conference we have, for example, too created there are also quite a lot Experiences I with the infrastructure behind it yes Also in this lecture with incorporated

yes first of all first what is the cloud I almost now I believe something shorter because I believe in the lectures have been clarified before we yes, on the one hand business service quasi we computing or space us easy to rent right easy but there is also something like that application software I am in end use and that I can use them me on my desktop pc or similar that could also be installed as But the cloud is about this cloud

not really there now the different service model software as a service platform master service or infrastructure is it now have infrastructure-as-a-service sometimes highlighted because that's what it's about here this lecture now mainly then There are different delivery model which public cloud the private cloud public cloud is simply a big network where I just book more resources can the private cloud will be me actually have my own rake network but these cloud functionalities, then that I just like one service billing or can book space book in my own rake net can and of course the advantage that the whole thing is still my corner is safer and is next to the virtual private cloud that's like here virtual private network so we have a public in and of themselves network we encrypt everything in between so that's almost in the end effect like a virtual like a private cloud will work later too Here is the main focus of the presentation above all things on the public cloud

as well as the benefits of the cloud Of course they are very flexible

the size can be quite yes as I want determine the performance is the term above all things was also flexible here I do not need me only one somewhere renting a computer also takes a while probably depending on that until that he is set up that's why we have here also a very small one furnishing period the settlements is in most cases actually accurate to the second and the small instances are typically cheaper in the same way Relationship as the big authorities So it gets us from the cloud providers also usually imposed a bit makes small instances fardi fast high was down again quickly if you do not need them anymore and not somewhere in the computer to stand to have the whole time goes through the disadvantages

The cloud is first time when I am now can not scale or wants relatively expensive in continuous operation at least in comparison to one not cloud solution we have no static machines that means we have many instances on the a page and they also change constantly for example the ip-address and The accessibility is not so just like a non- cloud solution just like that moreover, persistent memory is still A different typically treated the is that I need me one more time extra book the cost again extra usually the instances are quasi when I drive her up and when I do she is down driver everything is again away what is stored in it and for some use cases as appropriate interesting it is being virtualized beautiful if I am now somewhere again wants to virtualize as I have often problems and that's not worth it then yes micro

services now if i get the cloud Anguish as I said the big advantage yes that I instantly up and up can drive down again that's actually perfect with one micro service pattern the whole implement this now ask what are actually micro services that is actually an application software the independent in individual processes can be composed and talked about communicate with independent interfaces can at least that's what it's about wikipedia also what are the advantages

the microsoft we have individual small services Of course, in the overall context, they are relative well manageable in contrast if I am the whole like monolith or I have to look at something similar these clearly defined interfaces but have also a big advantage this gives me a clear idea communication they are individually scalable that's always the problem Something consumes an awful lot computing power when i put that in microserver I can pack it easy synonymous better watch which of these micro services just has a lot of power and I could do that simply start up several times or one larger instance or similar all things but also an independent deployment that means we are now not for example that they say okay we bring in the end of the month a new version out and that may take three four days until the new version outside is but we can each other individual service at any time actually the little ones and we can do that a lot bring out new features faster and faster just more independent we have to not us with great others like that arrange but we can also for example in service easier to exchange so if us now somewhere this service makes you stand out again and again only problems then could you go all that there is comes here before all things on the independent programming languages interfaces on it and they have to remain But that's why I can do it without it go again and only one rewrite small micro service or just stop also expand and change and And it's definitely a lot easier in micro services microsoft it is also independent of programming language and technologies that means when in a microsoft service let's say annoying search necessarily be used would or would solve many problems then this micro service can be without further with a different technology work as the other micro services but all the same with the so i could speak microserver write ingo and the other yes yes yes javascript or yes if all that is fine then we are up also in principle fail-safe us drive just typically micro service not only once but at least twice high if a micro service fails is still the others there and if you 're good a good one orchestration behind his micro services should then have the quickly back from other instances can be raised But what do I find the biggest advantage of microserver cez is that one is his Software according to its organization in the structure a company this is often related to microsofts called the so-called conway and here I think is also the really with biggest advantage in that's just because of the microsofts individual teams independently with each other but you can still work as a whole service behind it

However, Microsoft also has synonymous microsoft disadvantages First of all , the decentralized are the first we will also see what's there later all for challenges then come the us bonds that typically clear higher in a cloud so if i am now something time critical from where I so little latency as possible wants to be one micro service probably not that perfect match for us communication is often more difficult so unite what points accessibility and so on I suppose because of the example when I within for example one monoliths and some function want to call is the one thing of nanoseconds or microseconds until that This function is executed skoll within a service micro structure First of all I have to change my mind contact service via the network that will probably be more in the millisecond or still ten milliseconds range and higher just lie and above all things are the whole architecture much more complex we need Specialists the environment is clear more complex but as I said with the Benefits monument you can see me too find the cloud he holds in about the Libra but bring the micro services then in the end the actual advantage behind a cloud environment like that

What do we need for my opinion necessarily a vision we have to be clear what we have in the cloud want to exploit so before some else possibly also in the micro services why do we want to go to the cloud that way and then we should be ours mission as we implement throughout the year want to think about it there should of course also be taken into account and in that we are included should also us strategies of course also for the advantages but above all things for the disadvantages of the cloud and especially the micro service here we'll just take it for granted

a few strategies like us against the disadvantages are heading to go to and what are disadvantages First of all , we have a problem as far as logging is concerned we have a lot of individual micro services on different machines if anyone has any information now I just want something right now I'm running right then I would have now just after the problem I got from my computer as a developer on who knows how many instances and above all things have to be grasped and there One is the loks would have to get that Of course, that would not be really nice and very expensive what you for it

Of course, then use is one Central attracts that means everyone of these individual services sends his his lures water has to one central service of this wide the so to speak , you can do that too browse example or similar I can recommend here above all things

we use the erika stack at rewe

digital for example and that's it a real relief you can see just here in the temporal order then loks too All services are gone and can about it then, for example, quite well shut down what's up now we have expired but also within the central loks a problem when we look at that

just imagine here

here are two computers that make two request at the same time I can now from the outside do not immediately see which gateway has for example the computer turned when there is union all is scaled and multiple is pulled up which web service has then answered and which other services has the web service each addressed so be it database user or something similar but do we have a correlation id or something similar in our blog then we just have the big advantage that we understand these ways can So that means we should not just lure but we should as possible try in ours overall architecture has a correlation with and that should be natural as early as possible within the call can be generated at all so that they are consistent until the last service is recognizable and he must of course also from service service be passed on and also lured otherwise that will of course bring us that Nothing is another point

the ba against in lok if just and just service at one point not correct works it is very handy when to break through some flex credit parameter or a lighter or similar the man sent debugging mode can adjust where then the services even more users can virtually lure because that want We do not typically that either constantly such a detailed lures yes is written with another point is

because the service discovery we now have the problem we have A lot of service is coming up They will also be different depending on the scales and are there several times like So now make it that one Service knows where the other service is at all on the one hand there is the simple one possibility I'm here on the on the right side you have that reverse proxy uses it for for example, dockers form does that as well automatically that means I turn me every time to the loadbalancer and the load balancer will send it to the respective service further the recognizes a bit more elegant I think but the whole thing is still going on example with consul da he offers one domain name service in the end and I can then talk about my services can dissolve and also each one service in turn give the opportunity to resolve this via dns and use dns Either way, for many other things on the Internet That's why I think so actually a much nicer one solution so further big problem is

of course the communication between the services the most important thing for me First of all , that's the interfaces very clear and great by standardized should be

in the end, if we go over again now the individual teams talk about them maintain individual micro services In any case, the team is always there that is responsible for the service also for the interfaces responsible and they have to the possibility have one to be able to bear responsibility and to build this

but then these teams should make sure that the apis no have breaking changes because that automatically ensures always for any problems and in most it is somehow possible to get along without breaking changes then you have to stop just new endpoint build or similar there is of course things do not work there either especially with legal aspects or similar but in these cases should Definitely a great communication between the teams now as possible Again, that's easy is communicated and you should also watch a bit if these interfaces dedicated are still be used or not but there is another one possibility of so-called contracting fact is that for example a whole good example I can do that between the teams such a kind of contract if you will closing a digital contract one says So I use the following interfaces and expect the following data and this data can then be shared with the other team just packs and pack packs that then against the api of itself and then the team would be right away be informed here is a change that the other team that could not affect them more work another very important point is when we are in the public cloud that we secure our communication I think it's always best there the tls proceed right now in the version 1.3 where it is this zero round trip time is the name of the game Communication is also relative here fast and does not produce that much overhead and you can do that too within the public cloud, so to speak ensure private publicity

another point in communication most of the time you will find the rest api it Before that , do it like quest and chris pons back it works very well too But there are certain things when Data rarely change, for example, and each time I would have to query there as well At least a lot of overhead you could but working with caching now does

I also recommend problems here all things event streams such as kafka to use that is, if the change in a service is created deuce the as message knows that and then other services consumption in the end effect so that can on it say the say we say times a new user is created and there are other services that say if a new user is created then inform me about it then you can get this information I've got it right away obviously less unnecessary communication then another point

The metrics metrics are awesome problems in advance to recognize but also for example the most important order to be able to scale reasonably because in the end it brings nothing is the scale but where i am Not at all, where is it ? the bottleneck is or something like that

my personal recommendation At least yes is celebrity teams or similar things in particular Use connection with cover they make a very nice dashboards where to get this information wonderfully then read but then another

Another problem we have now insanely many services have to can also be started up first managed to shut down again and and that can quickly become a shirt to work on that's why you should be in the cloud too Be sure that you are in does some form of castration you can only be with hubertus mine personal recommendations or if you are only something small may have to work with dockers form as well

said the biggest value after mine opinion is in conway that is the micro service structure reflects or ours organization structure Problem is of course here Responsibilities must be very clear be defined I would be domain driven design of mine opinion an architecture and recommend Of course the teams have to do that responsible can also wear so

Now I wanted a short conclusion for the guests what do they bring me then my work are quasi noticed and the questions are almost ready for the open source for microserver at metrics

I do not really have anything yet really can find that is, they typically become not provided with the scalability it is mostly a little better but they are not really optimized to be scaled to that means you have to be yourself Again, take care that too really that logging is working actually you can do it relatively well do not do much wrong because you here the loks only need to collect however, they too typically have no correlation aldi therefore one small vermouth is there However, the communication through it that most oecd standards use is in any case already generated a very good thing so far as the lecture

if that still exists , yes [Applause] yes thank you very much there are questions that have all understood so straight away Thank you very much for the presentation a question is there typically the case with you that services also with each other communicate and if they do it then directly or is somehow in the mean values ​​in between the services mostly communicate directly but it is also very common before actually typically only just let's have a rest api if that's easy

just so a query history is but sometimes you look like I said watching this event streaming is one really practical thing that if you have the possibility you should have a look seriously consider whether you do not use wants to use simple rest calls instead of Thank you yes I have another question this is shown earlier the ex day for the logging was evaluated with cuban and then back at the metrics there was graff anatomy prince where would you like that the demarcations do not look like that now easy to make even with celebrities and there to integrate something yes so the metrics that are very small deposit and so on with those i only work so i count to For example, the number of requests is high I have no further information because logging, for example, says so More there is a mistake in line like that and so happens or I'm coming to something not really clear that it has just then lures what goes out Of course it can also be purely informative its like new requests came I could also use metrics on that of course Of course, they are not like that performant like correct metrics about politics you have to think of everything You do not want to go everywhere always write out loks that is yes also avenge intensive and grad metrics they are very narrow that is in the the end effect is just a number I get somewhere in my program and I have high count with each request then something I counted up the number of requests and then becomes 56 or something like that Of course, hardly anything in mine in mine Performance makes you attract Write with any stick create and much more complicated others may ask bit killed I still have the problem with me the size of a single micro services Imagine if a task now also distributed several micro services Of course it could eventually become there are too many problems micro services will be because I'm so fine granular the whole on split that the latencies in the end the term of the very high up that was what you mean Regarding a monolith can within less seconds back let's say we ever deliver microserver but try bring in and stay modular and yet stay warm the higher the latency problematic is there any best practice to say what is classically a micro service or From when do I have to split in several books that's always that's one very big problem I admit that too we have now, for example, at rewe digital the problem that we certain services have only separated and then only noticed they are always writing things back and forth and actually make she is a logical blog that is We are now considering this part too merge again you can see that, but I think so in fact, I believe the metrics practically we would have that without the metrics also not found so if you just just stop with the metrics then man sees here is the response time too big or sg messages on the Lost away what you just stop in a small network so barely mitkriegt but if you of course who knows how many requests then go also like lost packages that is not allowed to forget So I would do it as small as possible cut at the beginning and possibly through monitor monitoring via metrics and if necessary merge again but actually it is from the structure out in my opinion most of the time better to cut the first small and we did not have one yet real problem that the latencies so have gone up or theirs post it that for a long time Has lasted it is already quite fast So you better work bottom up than out his monoliths try smaller ones to build packages it is not quite understood prefer to work bottom up than his monoliths then try the fragments of too many work so thanks if are asking other is the suspect also intended for coffee outside disposal