Merken

Planning for the worst

Zitierlink des Filmsegments
Embed Code

Automatisierte Medienanalyse

Beta
Erkannte Entitäten
Sprachtranskript
so I guess I'll have to ask someone I know pass around the the might because as you may know or remember this is an interactive talk so we prepared sound step 2 4 around but really looking to discussion from and sharing standards so the only thing that is going on the other thing that has to be performed at the beginning and and this is the only 1 who can be I guess which was beaten to start or sorry I guess you agree that we now proceed OK so the topic of of this because they just over a year is and an interactive talk so we're really hoping to share experience and why did you do we want to make this this is run a new Java and we work memory and that's where you can find as few on later now discuss things with and but to get back on the title of this talk and it's about what happens when shipments and the main thing in our job Tadeja we ran some pretty maybe through but and web services and that gather data for all customers and we can never be down downtime is not acceptable and using that as which is another story is not exceed acceptable even and so we have developed over the years some kind of and practical reactions and we have around to develop and design our infrastructure and a bit differently and we're still learning that's why it's an interactive book because we don't claim we have the answer for every use case of so we wanted to start with the the basic steps which we lead maybe I hope to move the conversation will have this take a simple example of a which young will introduce you to the presidency of this is like a very basic application like what's going when you start a company or anything so you have an Jennings was service the regressor and that's your of flask applications want those all of those except and you put all your that kind of movie reset the that for example but really that could be like any of these so the 1st
example is what what happens when you databases that so In all cases we have multiple
solutions so when for example the so it could be better but but if you have that you can have just repeat guesses of databases so if 1 burns well still to mostly over the because I can take the lead in OK you continue to serve as a request something else that could happen is that you miss some resources for example you don't have around anymore so if you don't have running all well you could trigger some automatic and kills supplied with arms you with the for example do that you can just saying you with OK if that process data I don't know 1 G divides K culates on you could you see groups are like with their careers or anything just to say OK this process just that amount of memory and if you don't have energy is getting more like that these burn off if you have a big failures and what could help is a red 1 with tend to think of basically never-ending prediction the weather weather conditions on something that doesn't have a really on that as our groups seen you could use a distributed file system but like interface for Richard Wheeler and anything there is a lot of things you could have this is regulated some this case for some of the cases that I get up some there was a risk but that's 1 of the church stick our if you're over several slides database comfortable anymore request because it's already like that it's false was was our while there's not much you can do it and monitoring it so you know where output with when it up and and skating recently so yeah just most of us you can you can handle requests so if you have like some some of the ideas are a summary much about but is there to doesn't like like say it is really an interative things and where you get the microphone at at like the race that you to raise your hand if your back and that server already crashed Europe Web Service applications or you have gone from there alright so I guess you all are experiencing this year like I said we re prepare stuff like this and we get deeper and deeper and in between the 2 so yeah this high so I think we all shared experience of my question would be why don't you will use or did the induced on any of the standard tools or solutions for those kind of problems for example of for deletion here and is also seems to be a good solution I can't answer that and for complexity see stage and who doesn't know me so we must already a lot of this of the of the audience so just to get back about what it is it and correct me saying it wrong it's said Christopher service-oriented um clustering service-oriented solutions with resource management so that again following resource somewhere and when it's somewhere else if the given several which was writing accidentally dies right but setting up method and managing it is an overhead that you may own may not want to to to have to burn it is is also the same kind of thing by google google platform runs on communities and it's so so maybe you could solutions and it depends on the architecture here yeah we took a basic example was Nolde automation whatsoever and because of so we believe that and sometimes simplicity is and and built-in features of the technology we use our best the response to making a bigger infrastructure and adding again complexity maybe maybe you can save complexity by using right technologies or technologies failure in the right way also we want to recover as communities and so but this is really the 1st example in the next example will go like on bigger out stitches so yeah I'm still my experience that I heard a lot of tumor responses from different games and so and the can use oscillatory later on day and we do a lot of lot of moving parts and so on and sometimes appear really sometimes it's perhaps stupid to just use something interest like a weaker do instead of having to answer the form it freely and yeah yeah like I said it really depends on your team and the size of your team follow your company yet but I really agree with you I didn't say like please don't call like planar deceitful system that is just like the sun like Gloucester get get nobody yes you're right when I say when we wrote that solidifies system where world we have we have more in in mind that HDFS which we use intensivity in we spoke poetry and in my experience and not very hard so replicate it would have left fearlessly definitely reducing if there's a master slave we're going to the goal and but the about that but it's less active at our political status when they want go there at our database photographed our Monday did not embrace and how it could it with these yeah yeah we'll we'll cover maybe deeper which for example who relate to your to to the problem you your you're talking about him and like I agree with you that's only pure other very any other the other failure experiments hi so I would like to to mention that mostly which is kind of homebrew solutions on a higher note is that some day and up which are much more complicated architecture for example if you would want to somehow of on making all of these technologies that some fail-safe architecture and meaning in my experience genes have been that which mn with the master of highly complex maria that the clusters and what are and you know what the solution is simply use real use blog framework just do yeah we will get some some of of afterwards you right and well let's let's let's continue and I don't know if I will be contributing much was just an anecdote about hardware failures are and this 1 brother that there was only briefly we have this big data data center and Verizon or Amazon or something and but it was in 1 place in the world and I assume that the state will the of others that on of then unintelligible give us then we of year we have to have another 1 on the other coast of the west shore of stuff like that show sure of course we get to that as well you just want see me walking yeah that's because you say you were retired earlier on all actually on the rebels I or your own no actually it was a little bit late because we were I was stuck in DPS meeting so that these are all server things but yet if you have service just a of reachable so the network is missing and network is painting from as well I think it's also hard work and yeah there that's the yeah that's another possibility and
unreachable by indeed and that maybe what appears in March data and a server burning vernix are actually the 1st thing that comes to my mind was unreachable by is this is a guy who tricked over the cables true story sorry but not mean and the 1st thing is you have to make him remember that's human behavior as so maybe find for feet for for it and they look at a keyboard for 1 week whatever you want but you have to make you remember on the hardware side you can handle also switching and switch failure the easy answers to this on Linux for instance but it also works on on Windows used network bonding now when you buy a server that at least 1 network out of these 2 ports uses 2 parts 2 back to N. and spread them to 2 different switches it's pretty easy to do when you have a really really network a that for people and you can do NAC feet which is but more resilient and more robust way to do the same thing aggregating to ports and adding at their boundaries while call tolerance to Europe that's a or do you have any sharing knowledge about which and regional things yeah the hi is anybody using culture and what is not everyone running in the cloud to use lost a lasting and irrelevant yourself yet again just asking so that we do it ourselves so yes we we we buy everything we host everything ourselves and so we have to take care of this kind of problems yeah in relation to production yeah we use gentle indexing production which may be a lot of you haven't heard about well some kind of crazy people when I said were used to shit tumbling maybe it's passed through any other thing to share about network resilience OK now let's get the beats a deeper in this fact having if proof stack can also help and when it's not about only the hardware part on engine next . 2 things I like to use mostly these that you can handle back HTTP URIs by your upstream and get back to you is the 5 and that there what do you do do you as that these 500 error to Europe clients all you try to Hungary nicely actual an example of this if
you don't know about this it's called name location in g Next we use this a lot so when something bad happens you can see on the bottom error page 4 areas it is we we change 0 could 2 2 and to mask for the user West is serving some kind of um pixel because this is a pixel service and we can even under all if there was a red get parameter in the urine and we can steal redirect the user to the correct page even if our back-end didn't oral make something terrible so that's the kind of a little trick location an error page and handling can really save you from facing have 500 and pulls from our clients we use it quite a
lot you can also server from guassian so in in excess cash caching capabilities you can say OK if I get that never codes from my back and I will just serve as they catch response it's pretty and as well on your flask application usually you can also use their caching which can be handy if your database not right there is known as well you can have someone so actions from state caches there to answer something that and approach the and then you can have which effectively and the techniques to not use data this is more focused on that using data on protein and dust differ role in in in the best way is the is the way that you get some data from your http coal and this data is very important to you you don't want to be asking your client to send is that that twice more when and in our case navigation data so it's not brother and users browsing a website data and we can't have these data back and sporting it means that whenever we have it we are not forced to immediately instead that we can take these data right it's somewhere on this and have another process be fitted with this data and inserting in a safe way so if your bike and is down to just just just can try and try and over and over setting this data while it was a long time ago since responded to the client that's just the feral Darussalam messaging technologies such as may be worried about it already year 0 and q radii Q which is more resilient and and stuff like that that can help you get data and making into a task that's what so the salary philosophy which a which is a using right and you as a as an estate broker the important thing here to make and to us it is don't send back codes to your clients even if them unless you really have to depends on what you're doing that you can handle them even on higher levels of your infrastructure and don't use that don't ask your clients to send again data you have and ways and means of choose and all this kind of failures as well and to not ask for it do any of you use any of this techniques to traits for what what is the used a work for a Wordpress hosting company and a lot of what we did was basically land with the reverse the cash to a lot of the content being served is actually just so static static content in a way I can't think of a lot of people running basically websites single those glorified logic basically just you static content and for a while and then the back could fail all the time and customers would never notice if you served from cash everyone's happy the front pages of the main articles Europe all other things are available especially when your website is basically a content publishing platform because if content doesn't actually change that much of the dynamic it works very well and you know I have to wake up every 5 minutes and that of the wondering out you can see through it everything's fine no 1 will notice is that the people trying to publish an article if something really urgent and against the war from any other users will want to share their expand so what you should be using it for yes to compute so this seems even of on website like become 0 so of you can use similar techniques and if you need that that is a actually to In so that all those loss is because like 95 cent of the content is static so you can have something they vanish self static and then use some tiny JavaScript people just get you to think about specific to the user made the user name the name in the basket and that's it and I've I've seen it used tools like light the lot to the charge on the on the Beckhams and it gets really effective and even if you have like 1 or 2 minutes of downtime for you but can you use of consisted of an art navigate websites products on maybe maybe better than the other 2 counts of the in back upon you want resentment in yeah anyone else yeah I guess the conclusion here is is there around even a degraded version of your website or whatever of services through your on that having it 40 down depends on the use cases yes it can be argued that when do it come on we're here for I wanna hear the kind of for example if you just money from gland is better to say i dont les take finding after Mr. allowance I guess it and that can be a real lot so the next thing you can do is of course clustering application so and you 1 of the few but can use them are
monitored at base is done well it's still
so the bad thing is even with low balances is still a single point of failure so you can always go you can always get more redundancy even if you all you have to load balancers and 2 of his and is world that the and can go down so we have to get on that at the center of so it's kind of funny and I think you but yet redundancies could OK so now we can get through your
points where yeah that passenger interference yeah this would look pretty bad I don't know if it was shown shop or it's a if it's an actual but I was the light I don't want this is to be this is so coming back after the fire in the in the lets us in the room on the other side actually it's pretty simple have multiple datacenters if you run them yourself if you use the cloud and like
it's been subjected to suggested that in Amazonian you have to the sermon the notion of availability zones is that you should use make sure you do remote backups whatever you do and this then you France we had arisen story where and that the company lost its customers data and they found that they felt that they had backups because they were using directives and remote backups and when they try to get them back to the yeah I and can be said and it's face there again I don't want to be this is of or there and you don't want to I guess on the IP routine and connectivity step you you have BGP any test set for adding a single IP address accessible over all the role the world and something I'd appreciate also is DNS have checking for the we used with 50 training and they but US I don't know who we who knows about fruit 53 working not so much it's said dinner services from a US where basically you can have Joe distribution and based on dinners responses and other to do so and the industry codes that have taken so we show that a center or whatever happens is down 1 of your IP to your to your Web services down if we know the answers concept from genus queries anymore it's pretty and and she as well and application design and you have to think about Joe distributed applications whole rounds at least 1 due Richards said here OK so I'm not thinking about too much people but still it's a very interesting thing to do as a developer it's a real challenge as and that's iterate challenge even when you want these service of this kind of an isosurface it can be a database editor all around the world it said so a nice a nice thing to to try and achieve and 1 had this kind of problems already whether we're relying on everything in 1 place yeah probability on on the holdout estimator for years so obviously and not the administrator of the fruit of some kind of 90 was the denoising result so what on mainly and services located in 1 that the center and the size of the power of the and it ended up just in 4 hours of outage complete nothing more crucial infrastructural the so we just the valve of our clients is safe sort of and the after the In other words the number and the distribution from what time did it take to distribute your model behaving I would have to ask my administrative him and I know that so that's 39 there just edit it'll be the because of uh just how easy terrible stuff can happen and to that of center especially if it's not like a big company is the multiple center or service provider clearly this hosting again because these looking at kind of the same environment and basically so many things can go wrong we had a story I was I mean I want the company but basically it happened overnight and that there nights she who was monitoring object just everyone Felicity suddenly and the thing is that all the arms and basically when the morning shift game uh lake hold the temperature in the summer of where we've had a lot of our customers hosting this services was like since degrees with open the windows and setting that is just taking the threat to get somewhere this but basically a lot of things can go horribly wrong so choose your uh datacenters carefully and try to really get all of them if it's possible yes contracts with your providers on yeah you but I'm just freaking to do that contract to your providers enough usually and some of them provide a city like 99 . 9 9 9 and cent but not 100 per cent of the yeah this luckily this was a data center that was only used for the development but um we had an air conditioning that was running really hard and it leaked water into and power outlet that was behind the UPS no more interruptible power supply and what the proved that it was indestructible and uh it was down for 2 days where there was a major major problem you you have to call your clients the end Indian it so I guess this is this must be very hard to explain but I don't want you to be in the same department at this time the problem with GO with a surprise separated the distributed locations is not a problem when it goes down yeah when things come up again that's right I've had a few times where services came back up and we had both of them active because they couldn't see each other but the rest of the world could either C 1 or the other yeah and then people started using it and when the when they see each other again then 1 of them has to decide to be slave again and we have things happen that's called the speed brain situations where your brain doesn't know anymore because you had usually choose peers that's why in clustering in general and in everything you should do is always be at an even number and your original about the voting strategy OK if I am in dyskinetic situation who is I am also is my peer down the show only 2 appears be on the way to not at least to have treated as to be able to know if you can't reach any of the video to other peers your but Starting presorted it's not always something that it is pretty solid that is always thinking and even and there's always whatever it is OK so Teresa grapes but some some some real world problems are a bit more complicated and it's not always like that of sort of I can be like really coming straight from the because that's what we are going to see so I 1 day I was working normally doing stuff and 1 of 4 market gaze came and then told me there and that's the claim so you can put on to date on the server and the website something's wrong of all located Geneva logs and that this is something like maybe 10 dentyne those so look at this let's let's see maybe something's wrong so I guess it's estimation based you can log and everything OK so what kinds of what what
is so here's the plan was so
he goes away and I'm happy has something like 1 hour later and I'm still walking and the guy come back and tell me that it's still not working for for the client so I'm exhausted that are alright not say the good maybe something's wrong z I look at my application and I see that doesn't something wrong from so after so this again and that is that the same e-mail function can figure so if the they malfunction favor I will introduce OK pop it works so yeah my conclusion to
story is that you have to know you got lot infrastructure is great but called and if you don't like the guy who wrote the code even if you don't understand the good you know it's it's it's if your mantener of something you have to understand what you're doing and you have to reflect a resonance there are also other possibility that from the Zen of Python will and you don't with waves come undergoing sometimes it's easy like OK that's not my fault it might be a another servicing so that's why the posting as well so you can like really understand what's happening on a server and if you just the developer and duration and use of those awareness streets so do any if you had similar situations or did you hear what what kind of really we seen that from OK now is going to be brave for developers to raise their hand on it by 1 1 of the city's situation where a similar thing where the signal the cut this isn't working for a client she's trying to like do all these things like she had a really old workflows sourcing denounces working with Tessa passing by going to the site and on the invading this is all always working fine and around is what and what in really realizing the like a week to realize West have going back to the point where like all use no script so I have been using in back into their joint find within realized was a key enabler gel scripts Jozef uses different API and that's the thing causing a problem so make sure to eat your own dogfood and use or an API last life is that not call that yes it looks like it's working but I don't want to it's like this 1 out yet but in the end you're is considered as ad and that's it a rise in Python we get used to the libraries using reason exceptions a really common 1 that doesn't is non-cash pretty much every man library will return 0 so the reason is that and and so the rattled isn't about there's 4 5 for a 2nd think of different projects so we want to know where we trace back something to do likewise and working it's because we think non-gaseous looking when it's not yeah I tend to like the memcached library because of this but sometimes can be a nightmare yeah so you have always to chat about the it's like the goal 2nd of all you have to check the he retired and and that the Persians into any race or any other brave developer pleasure about this for years so we have 1 here this is this is true yeah so my example is not related to vitamin really but to beat the the yeah I know that whole thing see it that was more in PHP that's great developers of move for my defense and the role the cold uh but there is a very to think well you and you try to all some final and you have a syntax error then if you don't have to this properly then beat dice we dance and the web server returns the blank slate with the 200 OK and OK no other way to there right for that's a nice 1 we we ended up the workers who could where work we ended up writing some code in the reverse proxy that would detect the stress situations the white pages and the weirdest thing just because it's such a stupid default why would you return a 201 something's wrong and yeah it's horrible to monitor for that all what I have to mention the fact that the rate of the body itself this room for example as other science attitudes exceptions French sentences don't always research things to think switch but and the 1 thing that's that's not related to poison or any programming language really um I was having a so there would be pretty large disk in it um and there were very very very many files on that and then suddenly a developer called in and said hey I think that this so I gotta look doing the answers now it's not only 10 % used these as well I can't write involves anymore as OK locate such file and digital yet it ran out of our notes here something yeah yeah that's and that's the 1 that's and that's the 1 we we we often over here is absolutely right and 5 systems we don't rely on line so when you know that your obligation mice from another Phys think about them indeed I have another story that I forgot to put in the presentation that have something to tell you right now you're uh basically in my life told company like it was a small set of and if I walk animal and so we were trying to to get fast so basically or web server was running inside a the max some and sometimes when we looked at loads uh we were like just calling him too much and 1 day we were like 0 my god this is a web service fronting anymore and actually it was just to make when
you scroll it send out oppose the samples to the obligations so this application was down just because to with trapped what we were trying to achieve with him conference annexing production patterns and above that of philosophy I don't know how what kind of objective we can add who works as a devil or in the DevOps minded I see I don't know if you're waiting to see him or but this OK when think with company you just wait come the microphone is because we can understand you in the back the now it we should have thought of all the little probably liberally to do the DevOps question is hard because when you're managers and everybody talks a lot about DevOps hired a guy who is develops a theory as of DevOps position that that it gets you know so you get back to this decide so we develop and the devil so yeah back to developers and the seventies I was actually a developers who had to run back to our admins to check of why the fact is Dr. not working again all this elastic Search from being class classes with each other were interesting so in that sense I was developed because I need I needed to or about the code and about infrastructures messed up so it's not to keywords and yet it has a different there the depending on when you step where you stand which leads me to question who runs below current collection and can you show some experiments with it I'm interested and more and more interested in when it's failing obviously just 1 thing we were actually doing it's just this new products so is it was more of a proof of concept but we've already started to get it out to customers and the B word working together with this current consultancy who don't told us have to do talker and come from if someone knows CloudFront so our whole infrastructure in there we would provide services for Cloud the basal dockets unexposed reduces elastic stuff like that but the doctor clusters was actually 1 machine with all the containers for any for all services so don't do that for OK thank you so I can show a funny story where only to book a demon interest on the ice is still you can imagine that had implied 15 on super you made the 1st quarter just there and it was also found to the blood because you know who would have thought of it yet but do whom the projects to the previous points on how much effort would it take to implement something like home supervisor or whatever process that would more he draws the demon role in that much yeah you're right you're in some ways some thought sometimes where I on on various and we get a really agree with just about the DevOps saying yeah it's kind of of the world except it's especially for recruitment and from that's always see of normally is really like not a single person being interviewed but really change where you have here people will develops people who make up the the same here on but working together and understanding what the user is doing it's just a getting time so to a developer to acquire and in In understand what is not used to do and it's and the real world problems the hidden and so on we are all statistical graph and so it's a very nice for how and where I was looking at a graph and am I so that a statistics so it it wasn't really important because it was really just maximum processing time of 1 over service basically user as edge processing time was still very low so and we didn't really investigated z that thing that's on it stays there for like I don't know maybe to Austria Reed-Solomon more may being different but we never understood me what was obtained with what what is the maximum processing them was so I and then 1 day data but like what in so my 1st idea when i so that the graph was going so is 0 my god the services don't actually know it was still running have but what was happening is so i search for like 1 or words understand what what what upon them and then talking with 1 of the most POPs guy in the team include me and that's strange because at the moment I deployed instability will become 1 of the kinds of servers and so we looked at it they work what was the difference and is the only difference was in the BTC boasts 5 have basically Apple Denisova
was like uh all the time query at each that the reason session so sometimes it was just all of so just putting uh there is and the ideas of about a base indeed feels of each motion fixed the tree so yeah that's as pretty weird power of here sometimes some of the stuff like and it's not just uh resolving the the database server you might be surprised to see how many codes is of reversing the and of the inner so you have to not just for but also reversed is still happening in quite a lot on and that looks like this problem to be honest we felt pretty stupid with this 1 as well and this was pretty interesting because 2 days ago to make the presentation of what using consul yet for the sounds very contradictory from yeah and the thing that was have to try to a local the knee excursion and then all of my 9 of and that deal around 50 seconds of the life that would make the you yes absolutely what's what's embarrassing with this is that we also lack consistency in what we do you on another type of infrastructure we have local cache locality in a sketch but there we didn't have it and when uh Gilliams's the of people or we're working on the uncivil playbook is also true stop the normalizing all of these so show maybe we feel that we have something in production and it's running for so long that nothing can happen to it and we tend maybe sometimes 2 4 got about its resilience your performance or just applying the latest of your knowledge and just for the sake that it's running I don't care how long I don't need to bother so much and something where it happens in this case it was news you know were satisfied with this she Tiraspol processing time but on other types from on another type of application you might have been so I think that's 1 good trick is to always profile your applications of these slides and I was I recently used the improv from by by guys and it actually just slow down my service of our up around 5 per cent this is actually viable to this course which 1 instance and check what actually your code is going to actually in that situation I profile the code and I didn't have the same results as of the governors of subsets where was like what the fuck right it's not working as a chilled and that's where I like it wasn't really important so can we just played into video well it was a bit surprised when the 1st we came upon the ways embarrassing examples so you feel more confortable sharing this they have a common regarding the performance monitoring tools and you really have tool configured properly we had the situation that the response time was the other 2 response time was between 30 and 60 seconds and it'll schools by the uploading files for example by the gigabytes files sampled by comparison it through to people and the the need to increase the average response that in the same kind of problem sometimes or metrics ever goes down ends in which income I got my application is done it's running the show which using ways not using the metric system wouldn't do metrics on their application but doesn't have this kind of graph nobody you understand when the Isaianic leaving people question on have yes all you have to have could Western I was precisely going to last the station of this kind of questions have to manage need to put in direct in find from from the person feels graphing so you can all a 90 % to 90 % 5 consistently in the 99 per cent of all the the response time so you all know what that kind of problems can have in trying to having heart problems with these in the end we have deployed from me the use lost their final in using bad with the last assertion fully having a lot of trouble to really calculate the 99 % in the all the 95 % you might have you manage to basically we just show how much we use in general is a comparison between the Citroen data and is the basis of the days ago so it gives a good idea to it is normal is it where and in using carbon and and are for the for the visualizations you also have the and annotation um feature which is good where you can have a bar on your graph single OK and you can plug it to your deployment or continues in is stuff so you can have a bar on your resting OK from this point on this is version 2 . 1 and then you can do so metrics comparisons and related to quite cut deployment it's it's pretty it's pretty good in disaster recovery itself so but which thing to have uh so when you know that you broke something and you can do the same with service provisioning and deployment these at this time I added new server maybe it has some weird uh effects side
effects that for the percentiles if you mention you've already got Elastic Search if you don't have added this but have the actual requests there you can just use combine because it has a really nice visualization of 1st point of gives you the percentages the percentiles as well I think you should get for free from combined all the answer is the problem is to combine it with reference for the remote audience did anyone come with the question I guess we have for the formula for it's less so open discussion that question this 0 I'm looking anointed my has some experience with so trying to little deployment of version of years back and and only deployed to that's the 5 per cent of users tried out see how we handles and then go go from 1 especially mention next thing jungian instance with the so book progressive deployment anyone the thank you just for the same company was thinking about so a bit a little but recommenced we're always wearing of the traffic but it was eventually virtually in front of in I think it was expected to exceed that on demand any extended in that order and I like the new samples lowering make them present traffic on the new version then we had the software or monitoring and and on the metric on this little checking that's like the response time was not just the doubling etc. and I thought you know well if you most of the training and at the end of the day of the traffic was wonderful that was the answer to question depending on your stack I can relate we do and the lower level of this and we we run our using you was gangs and new was gay you had this feature where you have the it's called Dutch chain reload where your workers reloaded 1 by 1 after all and you was see we make sure that the 1 that is reloaded reloaded correctly before reloading the others so it's a good sensationally lurid and deployment and trick and I knowledge if you are really really committed to trying to kind of releases which is usually the main cannery released when you put the bottom and try the 1 of these questions to units that result this problem in a very reliable way but always very long the may be too much complicated for the case when because xactly this kind of procedure when you say I have a role in deploy strategy when I want to keep in the number of parts which is your application deployed and you and how long would it began to increase number 2 realities of the sky this future encourages but could it is still the 1 of the strict right is given given it is still doesn't have illustrate right yeah halfspaces yet how many of on you look at the various but the thing is so the pain was 1 of the really bad thing that I us to to document and you were and all that kind readiness that I think about yes so I want to especially thank you so the book was this is like an interactive format and the experiments but not the only ones that fosters will the original of thank you for a like a ticking people faith from the 1st you from this thing you think you've got through the few thank you very much yeah that yeah also that by you thank you thank you
Dienst <Informatik>
Umsetzung <Informatik>
Bit
Web Services
Prozess <Informatik>
Festspeicher
Endogene Variable
Applet
Regulärer Ausdruck
Kartesische Koordinaten
Standardabweichung
Prozess <Physik>
Web log
Familie <Mathematik>
Gruppenkeim
Kartesische Koordinaten
Komplex <Algebra>
Computeranimation
Rechenzentrum
Metropolitan area network
Datenmanagement
Web Services
Prognoseverfahren
Dateiverwaltung
Vorlesung/Konferenz
Funktion <Mathematik>
Serviceorientierte Architektur
Schnittstelle
Hardware
Datennetz
Datenhaltung
Systemaufruf
Ebener Graph
Arithmetisches Mittel
Rechenschieber
Dienst <Informatik>
Diskrete-Elemente-Methode
Verbandstheorie
Rechter Winkel
Festspeicher
Server
Ablöseblase
Disk-Array
Overhead <Kommunikationstechnik>
Aggregatzustand
Standardabweichung
Subtraktion
Server
Gefrieren
Systemplattform
Datensicherung
Framework <Informatik>
Überlagerung <Mathematik>
Bildschirmmaske
Reelle Zahl
Spieltheorie
Endogene Variable
Cluster <Rechnernetz>
NP-hartes Problem
Physikalisches System
Energiedichte
Mereologie
Computerarchitektur
Gemeinsamer Speicher
Extrempunkt
Computeranimation
Homepage
Metropolitan area network
Client
Schwebung
Front-End <Software>
Code
Datennetz
Minimum
Bildschirmfenster
Vorlesung/Konferenz
Parametersystem
Schnelltaste
Hardware
Pixel
Schlüsselverwaltung
Datennetz
Division
Relativitätstheorie
Applet
Systemaufruf
Biprodukt
Menge
Verdeckungsrechnung
Randwert
Dienst <Informatik>
Emulation
Flächeninhalt
Automatische Indexierung
Beweistheorie
ATM
Mereologie
Server
URL
Baum <Mathematik>
Streuungsdiagramm
Fehlermeldung
Instantiierung
Einfügungsdämpfung
Web Site
Prozess <Physik>
Gruppenoperation
Familie <Mathematik>
Versionsverwaltung
Kartesische Koordinaten
Zählen
Systemplattform
Mathematische Logik
Code
Computeranimation
Übergang
Homepage
Task
Hydrostatik
Metropolitan area network
Message-Passing
Client
Task
Reverse Engineering
Front-End <Software>
Code
Endogene Variable
Inhalt <Mathematik>
Serviceorientierte Architektur
Caching
Tropfen
Radius
Datenhaltung
Anwendungsspezifischer Prozessor
Biprodukt
Arithmetisches Mittel
Dienst <Informatik>
Suite <Programmpaket>
Rechter Winkel
Caching
Baum <Mathematik>
Aggregatzustand
Punkt
Relationentheorie
Einfache Genauigkeit
Datensicherung
Computeranimation
Lastteilung
Summengleichung
Rechenzentrum
Task
Softwaretest
Räumliche Anordnung
Direkte numerische Simulation
Baum <Mathematik>
Resultante
Distributionstheorie
Bit
Abstimmung <Frequenz>
Kartesische Koordinaten
Datensicherung
Login
Service provider
Computeranimation
Videokonferenz
Richtung
Rechenzentrum
Client
Web Services
RPC
Bildschirmfenster
Vorlesung/Konferenz
Verschiebungsoperator
Softwaretest
Vervollständigung <Mathematik>
Datenhaltung
Abfrage
Web Site
Zeitzone
Web log
Texteditor
Dienst <Informatik>
Client
Strategisches Spiel
Server
URL
Cloud Computing
Programmierumgebung
Wellenpaket
Wasserdampftafel
Automatische Handlungsplanung
Zahlenbereich
Unrundheit
Interrupt <Informatik>
Code
Netzadresse
Multiplikation
Informationsmodellierung
Reelle Zahl
Spieltheorie
Endogene Variable
Direkte numerische Simulation
Koroutine
Softwareentwickler
Leistung <Physik>
Schätzwert
Einfach zusammenhängender Raum
Systemverwaltung
Einfache Genauigkeit
Peer-to-Peer-Netz
Quick-Sort
Design by Contract
Arithmetisch-logische Einheit
Objekt <Kategorie>
Wort <Informatik>
Baum <Mathematik>
Resonanz
Punkt
PASS <Programm>
Kartesische Koordinaten
Euler-Winkel
Computeranimation
Homepage
Metropolitan area network
Client
Web Services
Reverse Engineering
Code
Skript <Programm>
Default
Schnitt <Graphentheorie>
Gerade
Metropolitan area network
Lineares Funktional
Ausnahmebehandlung
Bitrate
Kontextbezogenes System
Dienst <Informatik>
Digitalisierer
Grundsätze ordnungsmäßiger Datenverarbeitung
Server
Client
Projektive Ebene
Normalspannung
Schlüsselverwaltung
Fehlermeldung
Proxy Server
Subtraktion
Existenzaussage
Wellenlehre
Ikosaeder
Kombinatorische Gruppentheorie
Code
Benutzerbeteiligung
Mini-Disc
Programmbibliothek
Softwareentwickler
Programmiersprache
Videospiel
Einfach zusammenhängender Raum
Physikalisches System
Elektronische Publikation
Last
Tablet PC
Baum <Mathematik>
Subtraktion
Punkt
Prozess <Physik>
Ortsoperator
Momentenproblem
Extrempunkt
Klasse <Mathematik>
Implementierung
EDV-Beratung
Kartesische Koordinaten
Code
Service provider
Physikalische Theorie
Computeranimation
Virtuelle Maschine
Datenmanagement
Reelle Zahl
Stichprobenumfang
Mustersprache
Vorlesung/Konferenz
Softwareentwickler
Cluster <Rechnernetz>
Statistik
Graph
Systemverwaltung
Ausnahmebehandlung
Biprodukt
Objekt <Kategorie>
Dienst <Informatik>
Beweistheorie
Grundsätze ordnungsmäßiger Datenverarbeitung
Datenerfassung
Server
Wort <Informatik>
Projektive Ebene
Reelle Zahl
Dämon <Informatik>
Baum <Mathematik>
Streuungsdiagramm
Bit
Punkt
Prozess <Physik>
Gemeinsamer Speicher
Annulator
Versionsverwaltung
Kartesische Koordinaten
Videokonferenz
Übergang
Eins
Netzwerktopologie
Einheit <Mathematik>
Code
Visualisierung
Schnitt <Graphentheorie>
Datenhaltung
Stellenring
Abfrage
Biprodukt
Algorithmische Programmiersprache
Domain-Name
Teilmenge
Rechenschieber
Dienst <Informatik>
Verkettung <Informatik>
Rechter Winkel
Server
Strategisches Spiel
Dateiformat
Ordnung <Mathematik>
Instantiierung
Server
Zahlenbereich
Kombinatorische Gruppentheorie
Code
Ausdruck <Logik>
Arithmetische Folge
Software
Arbeitsplatzcomputer
Datentyp
Stichprobenumfang
Elastische Deformation
Response-Zeit
Widerspruchsfreiheit
Leistung <Physik>
Soundverarbeitung
Videospiel
Linienelement
Graph
Zwei
Bildauflösung
Paarvergleich
Elektronische Publikation
Caching
Basisvektor
Mereologie
Wiederherstellung <Informatik>
Direkte numerische Simulation
Baum <Mathematik>
Metrisches System

Metadaten

Formale Metadaten

Titel Planning for the worst
Serientitel EuroPython 2016
Teil 91
Anzahl der Teile 169
Autor Jacob, Alexys
Gelin, Guillaume
Lizenz CC-Namensnennung - keine kommerzielle Nutzung - Weitergabe unter gleichen Bedingungen 3.0 Unported:
Sie dürfen das Werk bzw. den Inhalt zu jedem legalen und nicht-kommerziellen Zweck nutzen, verändern und in unveränderter oder veränderter Form vervielfältigen, verbreiten und öffentlich zugänglich machen, sofern Sie den Namen des Autors/Rechteinhabers in der von ihm festgelegten Weise nennen und das Werk bzw. diesen Inhalt auch in veränderter Form nur unter den Bedingungen dieser Lizenz weitergeben
DOI 10.5446/21092
Herausgeber EuroPython
Erscheinungsjahr 2016
Sprache Englisch

Inhaltliche Metadaten

Fachgebiet Informatik
Abstract Alexys Jacob/Guillaume Gelin - Planning for the worst Sharing our worst production experiences and the tricks, good practices and code we developed to address them. ----- This talk is about sharing our experience about how we handled production problems on all levels of our applications. We'll begin with common problems, errors and failures and dig on to more obscure ones while sharing concrete tips, good practices and code to address them ! This talk will make you feel the warmth of not being alone facing a problem :)

Ähnliche Filme

Loading...