Merken

Implementing High Availability with PostgreSQL

Zitierlink des Filmsegments
Embed Code

Automatisierte Medienanalyse

Beta
Erkannte Entitäten
Sprachtranskript
who was a party yesterday everybody was the result OK so sorry for starting a little later and we're going to talk about implementing your idea availability with the progression of so what have been
doing some liability implementations of already in the past and I have been contributing to pose molecules so that that's a list of what they've been doing and what I'm doing now so far more than the threat extension framework and no working on even triggers so that we can get uh GDL support in the next in anncorra bidirectional replication and I intended to work on partitioning and that's kind of a running joke because each and every year the grunt work is expanding 1 year more so partitioning is I'm going to show
you some proven architecture about every diagram of an architecture was saying that has already been running for some years and prediction OK so the the main goal of it is not a theoretical review of to do things but uh but from the field experience about things that we work in prediction and most of them live implementing them at the media which is
a which public here on the competing France and they mainly have free services to offer to the customers that are really different but mostly a collection of stuff it to be so they do a micropayment internet some to services and advertising the way it's
important to give you that context before diving into the idea of is that is that's mostly I availability is the compromise you need to make in between different things and so understanding what the business is about really out doing the equally the good compromises so as you can be as a DBA if you're asked to implement I availability mainly my advice is that your 1st the 1st question is what's the business needs what do you need what why do you think you need i availability and what does it mean to you so and the reason why I mean has been happy with the PostgreSQL up to know is that the the really good capability of progress will have to adapt to changes we will see and that's what that means in the in in the
next slides a little so usually what the business sponsors reliability stability performance is all the jazz so they don't want to cut off on performance is just because you want some more and that's where you begin to talk about compromises and the question on the top of the slides is the 1 of the most important uh when we talk about high availability and PostgreSQL we really need to think of the you want the availability of the data of the service is already on both because the way to implement that is different from those of the set of compromises she will make are not the same and we will see that in some legal duties and what the business wants to usually is after customers happy about the business you provide to them and be able to welcome as many new customers as possible so you need to prepare an architecture which will be able to help you which individuals so
what do you do that the the 1st step there
will be a too easily so fostering unconscious so let's talk about what is availability what is
durability everybody heard about the ACT property of the model on relational databases that have been having for the last 30 years so the durability is that the in a it means that as soon as the transaction commits the system is not too low to forget about it in any some consensus so the usual tests for durability is that you run the tests in you unplug the poor on the server you really get back and you look at any collection but you know as being committed must be on the server OK so that's the reliability so you're not supposed to be losing any data time availability means you have access to the data that is durable at any time and then so the the French title of the book was a little complex and involve all those forwards so that's the glossary will we will be using architectures that's alter organize things that that so that you can reach availability and the ability and what replication exists in a way to implement so if someone asks you to implement replication that's the wrong question replication is just 0 to do things is not what you want to do nobody wants to to repay cation they want availability 1 something I get to you and say I want replication the question is what what they're talking about what do you think will get with the replication and will see you have lots of solution with the rest each with a different set of compromises for different use cases and so again the radioactive thing do you want availability of services or data for both the thing about it so for
this talk will going to start with a single project that we will make you from site to site and we'll adding requirements to it and so will begin with a simple it's a really simple up project your started users beginning you don't have users yet and then you will add some user as you will want the availability of data and services here we look at some of you we want to have some load and right balancing so let's start with
a simple it's all the the same books but it's often the case you start a new project you don't know where you're going yet so if you put it all on the same box have read era APG as an example of an application server but you can in your mind and replace a bitch retained his seat with whatever you're using it it would be wrote regular erase it would be some day framework it could be whatever you want to within a PG because it's easier everybody knows about it it is there to talk about it but we don't really care we care about the PostgreSQL side of things so we start with the 1st question the
world No you get new users is you're single lashing able to cope with many new user usually quite quickly it's not enough so that the next that is to
care about service availability what happens if you reach out of max connections then you you can't anymore except new users because you don't have connections to serve them so you need a way to look to leverage to solve that problem if you're using persistent connection do that perennially works so so the solution what we know about the ones already was really needed prediction of some of you good you know when look at it it's really great it there is what
you will do with it you want uh so pink boxes those are application server green 1 that's your database usually the 1st step you insulted you will besides the database server because it will underlie many got burned connection from the clients and it will manage to handle them with a few several connections I will but don't go into details about it does that because that's out of the scope of this distal comes only have 1 hour that's really the conditions that really nifty trick and sometimes you prefer to to other polluting happen on the pink boxes and the application servers and if you do the pulling only and the application servers you risk reaching a gain them X connections at each time you add a new books OK so you of traffic growth free books is not enough you add a 4th 1 so you add a force the connection pool and so that's another 20 connection to the server and if you have too many of them you will gain which makes connection so often enough you have to do once they're both on the application side and on the server side of early yesterday I use case where they are for people of the differences 1 after the other yes yes I won't have time to address that question and it'll come to me after it so the question was about the what was said to be due bones and the uh it's an important question and I happy to answer it but just not in that all hi sir there
is an example of the the legal of flexibility rich with the ones you have to watch for the this company there that's the number of clients in average connected to the server and then there is some magic here so that's the
active connections on the server and that's of any connection the servicing so we have more than 400 skies connecting to the server at and and prosperous grows on missing 4 of them so you can handle some growth and why it's important if you get back to that
growing is you want service availability so you want to if you have that many new customer using your service you want to be able to add some pink boxes as many as you need to for the application server without maxing out your database server
so it was a really nice so
that's opinion to durability
was doing backups and so for the end of that book I would consider that the people didn't raise their arms on that yet turning was was good in production I guess that's the only sensible and so if you're running in prediction your backups that well uh automated nature the are weekly backups all automated who as automated restaurant OK so you don't like it OK that that's as simple as that and talking about the availability of data and architectures extra extroverted if you not testing you will your dams well you don't have the dead the ability to begin with so there is no point continuing with this talk suffix of QuickTime waiting for you know living but that's the spiritually so that's the 1st thing that durability if you want to have to be sure to be able to argue data around dump them regularly assessed restaurants and when you do that
we talk about data retention usually so it is a review that the classical is that you do nightly backups you keep them prolapsed 7 days and each week you keep 1 of those for survival and h bonds you keep 1 of those for several years and if you're a offended business banking maybe you're asked to after 2 years of backups in which case if that's business requirements and you wanted to derive ability you really need to think about the storage medium you're going to be used because Turkey is from node the artist that is spinning and the server will not may be able to spin again so that's question that it to think about it when implementing a high availability architectures that's part of the game so if you use
didn't use the jurist or have attitudes to work to address the 1st parts of the problem but the restoring time in didn't die that's what in in case of a crash you want to restore the last backup right so you concerned with the time it will take to restore data because
With that technique you you are able to ensure data that do way ability to really important but what about availability restaurant is going on nobody can use the data you have to wait for the restored to be done so it's not wait for availability of data that's why
we have other techniques so was
family are with the idea of a point-in-time recovery the right and that's new that has been you impose was 9 1 in 8 1 that's like 10 years ago so you can really close that it's working uh we added 1099 and know will daily it in the next diagram eating images it's easier to explain that with the basics are that you are using archive command and restore a command to ship the right to log files from 1 1 1 0 to the other so what happens with both risk model is that any change you make to the data is 1st of located in the right a log because that of course to survive for that to simplify things that thing has names so you will find or the BGX log file with those are the same thing that just means that in case of a crash we can read from that the information was stored the data we had committed and allow you to use the several days if you do that and use leadership away those to a 2nd server then you you implement wants and that that was the way of doing it at the time and the in in does proven architecture as I want to talk about the something using both a woman monitor the which is included in sky tools in the sky to speak age it's just a solution for availability by Skype people and have been using a will uh which is from who and allows you to archive and restore to and from a reason 3 so if you're running in the cloud of control it's really and then you have to link to solve the problems school PG bomb and some at be divine and the only reason why it the steps written in there is because I don't have it's like 1 year of prediction of a story to tell about it but is really good and stories
what it it's looks like so remember the previews diagram still have the pink and green boxes and we just added blue well the blue 1 is the warm standby what happens is as the master server is working he will it will thanks to archive and restore commands implemented by will manager for example it will send the wolf eyes to another box and but rule books will be able to recover from those fires allowing you to have up-to-date data so if you implement that you have the better of data availability because you don't need to wait for the next digit 2 of fresh states that it's it's happening in real time and for a while but real-time for will weeks way better than with it is operated and so that was the 1st at 1 time of the so you know they will see just later we have alt standby instead and I will tell you why qualities altered and warm so mostly you want to implement that days but it's important nonetheless to understand that part because it's really the basics that all the replication stuff including progress well have been built around so it's going to little creaking in your today if if you're not familiar with the techniques like I guess I've left many gaps open but if you want to understand the political replication and implemented as a look at that technique it's the basics and when you understand that you will easily understand the rest so now
what about service availability because if we
look at that diagram and we have a pressure on the green box it's still with you have to go from that to the blue box redirect all the being books connections to what's the the blue 1 and we have to be careful when doing this which of the 2 have been using data so either you compromise on the service availability and you do the switchover as fast as possible and maybe you lost some transactions or you want to as a compromise and the data availability so when Kroschel curves you want to read with the green box of test us the service will get it to be sure that it's lost and only when there is so most of you will be able to switch because you really want the data and it's not a problem if you're not reopening the service yet you you want the date OK so with that object Europe that's a compromise have been talking about in the openings In case of failure you have to choose between data and services With that technique there is no order answer so we're trying to to
bring in the tables and new techniques right so should care about service availability the best answer to
that is to talk about it with the guys making the application because in that example we were able to split in between uh the part of the application that's dealing with setting up the service and the part of the application that C is used to get money from the customer was a click on it you can do that you see bread that and 2 different systems that are autonomous and you get a different set of compromises on each 1 the way to do that with was most well techniques is how to implement it with 1 of those magical replication things the solutions are socalled so slowly is the the first one that existed in PostgreSQL and then now we have to understand the problem and my professional might search personal preferences to was this but the flea arguing about the same thing got was a little difference what they do is install triggers when your database and the reserve each and any changes you're making and then the yellow you to replace those under changes to the database the good thing about that is that the the 2 database servers the provider of the subscriber there are different systems and their willie loosely connected they don't have to be the same at all so you can have different tables different indexes can be all the front end user just ship the data from 1 to the other so it's really loosely connected it's if you
use long list then you can even leverage the the transport mechanisms using called particular and you can implement your own specific batch processing of the data when you want to ship it from 1 0 to the other so you can have a processing in the middle of the application process and whatever we wanted and it's still transaction also have some leaders in the next slide that's particular or it looks like that when
implemented so this time they do the meaning something different from uh this and we have to production services uh boxing you running for the worst all the pink 1 is back and the blue 1 is the prediction service so that the 1 that you really depend on to your customers when for example so in the case of media where are implemented that kind of solutions it's really really the blue box is really getting the money from the customers directly this internet micropayments index on the phone payments extract so you want that books to be available all the time availability is really really important on the back end you set up the the command you want people to pay the AccessForAll also if the backends goes down to customers will be unhappy about it because they cannot do the set up they want to do but the money is still incoming and when you expand that to them it's easy to get that through to go with that OK it's a good fair scenario and the demon books and the bottom is running some because and in that case it's written in PHP you can do something else if you have more taste but with and that so it's earning that computes some really bad statistics it's basically the dashboard of the customers so what data into the back end system and they can see what did happen just now and the already be system so they can see how much money is flowing into their items and that's thanks to particular on the bottom of the drawing so with that case the replication in between the PostgreSQL boxes is made with the list so it's really to independent system and the application really needs to know about it you cannot implement that without talking to the developer today there is no way you can bring that on the table without involving the application development but this is the simplest way to implement I availability in which in inductive OK so again when asked him to implement a availability it's easy to think that to be a problem and maybe I need to talk a system guys to implement it with the right network or other words set up extract but if you involved application guys you can have a much better of much flex much more flexible solution to implement and BG Q was
using already OK really a few people and that surprised that k uh we use that and the prosperous Camilla org infrastructural that's a good it is it's a really nice if you ever as soon as you want to upload some of the processing out of the interactive session with the user think about the Q it's exactly made for solving that problem and it's still transactional so if something bad once and the transaction gets a black people want to happen in the on the scene in the queuing so you don't have to care about such problems so if you're doing several different things in the same transaction GQ grantees you would have all of that will happen in the same batch when you're processing when your batch processing of transactions so you still have a transaction when there is about what's happening in the prediction and you have a really clean needle API using it by default it's written in Python so you use it in Python I have written letter lead to use it in PHP because in a media rights and it's a PHP shot and so they have all you know the demon classes and they have loads of things implemented already in PHP they want to reuse the batch processing and having done that I can tell you that if you need to do that in Geneva or whatever it's really easy to support the JQ to any language you want to because they have been really smart and implement it a complex the API implementation happens on the post was real side so all you need to implement you in there for your own batch processing is to be able to connect for SQL and cold store prosody and that's it that's the EPA so whatever the language or system you're using you can use and it's solving a lot of problems for you so that get and again this nowadays you can even have cooperative workers which means that you can have several workers splitting of the event stream in each of them is a processing unit of batch so it can leverage that if you if you need to uh will not insist so much more than that of look at particular for for me it's like the missing piece in possible scholar offering but it's interesting actually the so just a little so we are continuing
but the thing that the prettiest architecture we add some more flexibility in the availability compromises we want to be able to make but we still want to be able to do both the data and service availability at some point so let's see there some more techniques to get there so in 9 1 we don't want stand in I know what actually but missing some little options so I would advise to consider 9 1 to prefer 9 1 so white hot standby this autumn warm standby only war the difference here is in weights what's the matter is that you the Primera server understand you can use standard as a read-only server and when you do that in case of a promotion of the of the sun that really means that the master crushed right so you have someone with only load understand by and when you promote no query is effective they continue running you really load is unaffected by standby by promotion so it's really it's hot standby because you want to have 1 service interruptions when you promoted OK 1 means that you really mean it in a way and when you promote it you need to restart it or something so you putting all the connection you something that is happening and then you can use it as a new master with all its and by there is no service interruption so that so that's a way to to get a better service availability with the 9 1 you can implement either synchronous replication or asynchronous applications where conditions were so the difference is that little septal it's the difference that what does it mean as a as the application to note that the comet has been done if you're asynchronous it means that the Committee has been done on the primary so and sit and later on it will be done on the understanding that it's asynchronous if you're in synchronous mode you have the client application only knows about the comet having been done when it has been done already on both the master understand them so if you have a really high availability requirement you can implement it's with what in a way that any transaction you make when you committed when you know it's been committed it's being committed and to self so you can use any server you want you still have the data online and ready to go so that's that's what it's about this have use and the problem is that of course you need no wait for the standby to have that information related to do your transaction that you just committed so it's so it's slower I get back to that the other thing is or do you send the information to stand and what's really good about it that is that is it's using the usual connection facility that you have with PostgreSQL and that's called the PQ that's the implementation of the PostgreSQL cortical and it's what we use for the uh standby connection which means that you have all the usual Apigee HPA set up and you see the future of the it's not exactly meaningless it's an open door with all your data are all the bits you having to server thrown away from the dual so that anybody can read them so if you're a little concerned about security that's not what you want but with the use using the acute means that you can implement SSL to encrypt the communication and you can even go as far as the finger in a certificate of indication when that you want to do replication so you can do that obligation that streaming replication if implemented in a very secure way if that's what you need if you're on the network and how do trust the way to implement that is using would cover that come and they have a look at the political documentation if you want to know more about that as an aside if you enter data into any of the the legal of details about anything you you are you hear about PostgreSQL well the 1st thing to do is to look at the official documentation for the major version of you could go and use Google search but then what you will find some requests and if the block post is like 5 years old and the set in the way the best way to set things up as changing between you won't know about it that PostgreSQL condition is really great so that that should always be your 1st stop go look at the documentation where it will fit we we are expanding it would deal for forking to maintaining the documentation and so really of credit who would be happy you have done it and so back to the performance problems and flexibility viewpoint you can switch in between synchronous or asynchronous those replication mode where transactions so that's unique there is no of database in the world that I know of that is able to do that what that means is that you can have a trigger for example that looks at the balance difference for example you're you're you're putting some more money on your I come and that's a is OK if that's worth more than 100 2 hours then I'm switching to sequence I don't want that to be lost whatever happens if it's less than 100 are the cost of waiting for the standby to uncertainty is not what so it's asynchronous that a compromise and ready to make for the business OK so you guys it's always the same compromised talking about since the beginning it's data availability against the service availability and what data is available if you need to skip service continuing so when you have the time to look at after the date of what the you if you're in synchronous replication anything that's been committed is already understand but it's costly to do so and you can switch that in a transaction which means in which we go OK that's a level of flexibility to implement your own compromise that seems as far as I know and problems so that's really important and that those used to reach good performance legal when implementing even Synchronous replication so it's really known that Synchronous replication means it's run slower because each time you come at you have to wait for the standby but you can do that only when it's important most of the time it's running as fast as usual and when it's important it's waiting more but the overall focal it is not that much affected OK so you can have this makes in the same server in the same cluster set up of synchronous and asynchronous so that you can reach a good compromise on the availability of data with keeping good performance even so you can't have it both can it all with of was and if
we mix and match everything was set up to know if there is what it looks like so at 1st sight it looks so literary little too complex the set up of influence that if the set up is too complex by definition you will not have high availability because it's too complex to have a availability what you have to remember when the reading that it's that it's so you have to spit it in the media the the the middle of the rows are willing loosely connected that's the techniques we've been talking about before its long-distance GQ so it's really loosely connected so what we have here is to autonomous systems with different set of compromises foreign availability loosely connected with some logical replication stuff in between so if you're concerned about the complexity of the set of well just look at all of the screen at a time because it's something that happened to this part of the square and the other 1 is not concerned it will continue running did you have to say quite resilient to do that for or you know the guy doesn't you bring it will just spit out some or single log in every 20 seconds and when the other guys back the replication of is flowing again and will catch of that I have to that of an production more than once and it's really just walks so that the failure cases you interested in in that case is when you lose 1 of the green boxes and the solution is to switch to the blue book so it's a failover that you can do you promote the standby and off you go and if you're using old standby you can use it as a read-only server and if you're using will streaming you can you have the option of goal of doing asynchronous replication in which case the usual lack is in the ballpark of like yeah uh 200 or 400 ms and if you're seeing for those in the data this is that has been committed in the green box already made its way into the blue looks so at the time of the promotion of the standard you already have data so that's the data availability part OK but we have some more
that this diagram you do just so before is not implementing right scalability so if I show you so we are going to talk about charting but if you look at that if
you're if you're thinking about uh reads scaling you so the scaling is important but for the data availability but for the service availability right if you want a service to remain available until sought to remain available even if you have a high load in to spread of or the service is not available anymore and so in this talk was really about trying to reach data and service availability so that's where the exponent that so here you can spread some reading on the blue boxes it's redundant but you can spread the reload but any rights that happened on the on the green ones is going to happen in the blue ones too any single right is made on both boxes so you didn't uh help yourself out all with that your capacity to understand writing transactions
so what do you do that the way of implemented to do their it is the approximate their proxy comes from the sky to so they are they have already developed a full set of solutions for all their their availability and scalability problems and their solutions to spread the White Lotus via proxy what it does is offer a distraught procedure API that allows you to run the function on a remote server so when I was about to foreign data records that many of you I wanted to say it's it's a function that for another recurrent but that's that been many of you so it's really when you call a store procedure on the proxy server it will connect to a shot and run the procedure over there and get you back to the results so with that you can distribute the load and there is what it looks like so
yeah in that there aren't as many application server as as before and we are assuming into the first one OK if you see the 1st of the year of the rows I didn't growing arose from the 2nd and 3rd front-ends server that just because it would be it would make this thing is really complex to read so let's zoom in on the first one what happens on the is not the main difference in the in in these books in pink color is that no we have PostgreSQL on proxy rolling so you you install a program scrolling stands on the application server file you can do it in a 2nd layer if you want to run and it directly on the applications are but this instance is nuts running in the data for you it's just a storage procedure proxy so that the application would call a store procedure on the so-called instance of progress well but this procedure is going to be ready for we'll on 1 of the green box is depending on the shopping so there is a hash on some arguments and depending on the action you will run on 1 of those petitions so that's what I did level then P 1 P 2 P 3 that's the partitions order is a little bit here if you're using a proxy the first one is called G 0 for each and every time I wouldn't it and if you want so that helps you spread the right because now when you have a transaction that want to write data on the server which will write on 1 of the 3 I drove in that schema not the free of them so that that you spread the right load and so you you can reach a better service availability with that and if you want to data availability you still need standby for each of the the postmistress service 0 and by the way if you done by UCLA backups and stress that out but even in that case you still need backups
sometimes you don't have the budget with the money to run 1 of standby server for each partition for charge but with was worse quality of of great flexibility even there so that you know you can actually implement implemented that way which means that you have a single server a scene that is running as many for clusters in parallel as you have prediction charts and it will keep all the data sets for you the compromise you're making it the so the compromise your making here is that in in case of a crush of 1 of the main chance partitions the green box then mostly I bet that you won't be able to directly use the arts as a new prediction box because it is a link the load of doing the replication for all you the rocks in the prediction system so it's a budget compromise if you don't have if your project is not worth spending that much money so as to have a 2nd but standby server for each partition you can do that the compromise is that in case of a failure of or in books you have to bring a new 1 in the table and work we get back from the standby to this 1 and then use it so you can have a huge um time window but that's a budget compromise so I implemented that way already in a project where we well you it wouldn't be walls whether the management team it's a and that spending the money to off
that school diagrams and finding the money to on that and I
understand that we have a bigger than window OK but that's a management and budget decisions the good thing about it is that we have posed was well it's easy to implement it that way if you need to and so we've been
speaking about services availability data availability uh with uh the spreading the read load on the spreading the white clothes with proxy on some charts what about having both and data availability well if you want
both you need to stand and there is no way around that the only way around that is that you have a compromise the leading you to is the 1 you the data already the service availability but if you want both it's good know the 1st way to read that is the classic availability things the reason why we we have thrown free pink boxes over there if that if you lose 1 application server you you you want to still have 2 of them because as soon as you have only 1 of them you cannot speak about availability anymore so it's the basics you want with servers with what was going to be the same thing if if you lose the main the master server the green box you will want to promote 1 of standbys and still understand so the other 1 will have to switch over to the new mass 1 of those will be a new master and the other 1 will will have to be a standby of the new master OK so it can used a domain that you have to implement for archiving in that case so both streaming replication and archiving but thanks to a key over there and 9 free you will be able to do that really streaming replication and thanks to some other areas of development that we have in 9 3 a much reduced uh pertaining to promote the several so you can really fast what 1 of the standby is about you the 1 followed 4 foot on the new master another reason why you need to standardize is the queries conflict resolution and I but I'm not sure I will have enough time to go over that in little so to to make it shorter just remember that if you want to both I availability of the services and and data it's more complex than it sounds you have loads of different techniques and software to implement it imposed well lots of solutions most of them that I've been thinking about our including progressed well itself some of them like to understand predict you will monitor of or Whalley and bomb on our external projects but if you want the force you will need to look at that OK and if you want the rule picture if you really want to be sure you need to stand by for each master so uh did
this will be the capability of course was grounded and just after it did happen over like 8 years no 1 in more it began with the implementation of a point-in-time recovery in 8 1 0 years and years ago now by the way it was running at 8 1 in prediction OK good don't do that so 98 to anyone in prediction something you shouldn't what is called so in a in a 2 we got warm sunlight so it's the ability to have this done by books do online point-in-time recovery if you want to call it that way so that it's ready to serve in case of failure then we got it just done by 2 all q implement that is and that was duplicated quick enough but by streaming replication of old standbys technicality of somebody's not covering streaming replication while we can mix and match the terms some of and then we got the Synchronous replication that the princess to another time it's really unique in the way it's implemented in PostgreSQL uh in that you really can release switch that virtual selection which means for example in a trigger that's really unique and that allows you to all of the the best data protection you can imagine while at the same time having an allele of performances that is not compromised by the data security in 9 to read cascading replication to that if there was not a single diagram here with cascading replication architectures that's because the angle that both really was to show you things that have been running in prediction for some time and that has been proven and you can trust those ideas something that works in the field for real and I don't have that the role of experience with mine yeah OK but is there you can use it with cascading replication all you to have a standby I connects to a standby that connects to the master and you can I think in some tests automata did of like 125 128 standbys change 1 after the other so you can go crazy that way with the you shouldn't that if that's what you need you can have a try the next step
of the ink or replication is the direction of replication it's called direction replication it's because everybody wants to master but nobody agrees on what it is so if we call that multi-master everybody in the water will be disappointed that we're not solving the problem they had in mind when reading with the master OK so to make it shorter the master or does not exist facilities able to agree on what it is so it's it's a great but the direction replication means that the flexibility we have seen with along this in the slides before where you have the to several and the 2 of them are the primary provider server for some tables and the subscriber several for some models and you have crossed replication that way that's the directional and that's a way to implement the master OK so we're getting we are working on getting to be directly logical replication and go and in 9 3 that meant some of the corner stones that we needed some of its predicates a refactoring uh some newer internal stuff do that that even triggers for to be able to later implemented DL changes so it's on its way have written their names for but nobody knows what will get 9 4 gates interventions project so if you are really the key we will we will be able to talk about the direction of replication glory 9 4 but nobody knows OK that's it any
questions no photograph of taking was it if you do a Google image search on the course that's what you find and I found it cool so it's of course forward-looking within gore replication solutions and that support of what and saying it's it's cool over there was a question yes I'm sorry I'm not hearing you you know when you hear the 1 come to me out the talk and not to renew and so on OK against thank you
Resultante
Partitionsfunktion
Implementierung
Mailing-Liste
Datenreplikation
Partitionsfunktion
Framework <Informatik>
Computeranimation
Arithmetische Folge
Datenreplikation
Richtung
Booten
Hill-Differentialgleichung
Maßerweiterung
Ereignishorizont
Kartesische Koordinaten
Architektur <Informatik>
Telekommunikation
Dienst <Informatik>
Computeranimation
Internetworking
W3C-Standard
Diagramm
Dienst <Informatik>
Prognoseverfahren
Datenfeld
Maßstab
Hypermedia
Computerarchitektur
Airy-Funktion
Stabilitätstheorie <Logik>
Kanalkapazität
Architektur <Informatik>
Prozess <Informatik>
Güte der Anpassung
Mathematisierung
Hochverfügbarkeit
Systemplattform
Dienst <Informatik>
Kontextbezogenes System
Computeranimation
Stabilitätstheorie <Logik>
W3C-Standard
Rechenschieber
Stetige Abbildung
Dienst <Informatik>
Maßstab
Arithmetische Folge
Menge
Computerarchitektur
Softwaretest
Relationale Datenbank
Subtraktion
Architektur <Informatik>
Kategorie <Mathematik>
Varianz
Dienst <Informatik>
Physikalisches System
Computeranimation
Metropolitan area network
Transaktionsverwaltung
Dienst <Informatik>
Informationsmodellierung
Menge
Datenreplikation
Server
Computerarchitektur
Web Site
Quader
Dreiecksfreier Graph
Kartesische Koordinaten
Dienst <Informatik>
Ranking
Framework <Informatik>
Computeranimation
W3C-Standard
Summengleichung
Dienst <Informatik>
Last
Last
Rechter Winkel
Server
Projektive Ebene
Einfach zusammenhängender Raum
Server
Dienst <Informatik>
Prognoseverfahren
Extrempunkt
Hill-Differentialgleichung
Dienst <Informatik>
Computeranimation
Eins
Inklusion <Mathematik>
Einfach zusammenhängender Raum
Server
Subtraktion
URN
Freeware
Quader
Division
Datenhaltung
Kanal <Bildverarbeitung>
Zahlenbereich
Kartesische Koordinaten
Computeranimation
Einfache Genauigkeit
Eins
Metropolitan area network
Client
Forcing
Konditionszahl
Client
Server
Eins
Einfach zusammenhängender Raum
Server
Datenhaltung
Kanal <Bildverarbeitung>
Kartesische Koordinaten
Speicherbereichsnetzwerk
Computeranimation
Einfache Genauigkeit
Dienst <Informatik>
Server
Client
Hill-Differentialgleichung
Personal Area Network
Punkt
Prognoseverfahren
Natürliche Zahl
Speicherabzug
Dienst <Informatik>
Computerarchitektur
Biprodukt
Datensicherung
Datensicherung
Computeranimation
Strategisches Spiel
Ereignisdatenanalyse
Fehlermeldung
Hochverfügbarkeit
Systemzusammenbruch
Euler-Winkel
Datensicherung
Datensicherung
Computeranimation
Strategisches Spiel
Knotenmenge
Nichtlineares Zuordnungsproblem
Spieltheorie
Rechter Winkel
Mereologie
Server
Speicherabzug
Speicher <Informatik>
Normalvektor
Mathematisierung
Familie <Mathematik>
Systemzusammenbruch
Sprachsynthese
Dienst <Informatik>
Datensicherung
Computeranimation
Wiederherstellung <Informatik>
Informationsmodellierung
Prognoseverfahren
Punkt
Bildgebendes Verfahren
Fehlermeldung
Datenlogger
Elektronische Publikation
Packprogramm
Kreisbogen
Diagramm
Rechter Winkel
Gamecontroller
Server
Wiederherstellung <Informatik>
Hill-Differentialgleichung
Computerarchitektur
Information
Streuungsdiagramm
Quader
Konvexe Hülle
Schlussregel
Dienst <Informatik>
Packprogramm
Computeranimation
Metropolitan area network
Dienst <Informatik>
Diagramm
Datenmanagement
Echtzeitsystem
Arithmetische Folge
Digitalisierer
Datenreplikation
Mereologie
Server
Weitverkehrsnetz
Aggregatzustand
Softwaretest
Einfach zusammenhängender Raum
Quader
Konvexe Hülle
Green-Funktion
Dienst <Informatik>
Computeranimation
Objekt <Kategorie>
Dienst <Informatik>
Transaktionsverwaltung
Diagramm
Druckverlauf
Gewicht <Mathematik>
Offene Menge
Ordnung <Mathematik>
Innerer Punkt
Tabelle <Informatik>
Subtraktion
Prozess <Physik>
Mathematisierung
Stapelverarbeitung
Ablöseblase
Kartesische Koordinaten
Service provider
Computeranimation
Datenreplikation
Mehragentensystem
Umwandlungsenthalpie
Kraftfahrzeugmechatroniker
Prozess <Informatik>
Datenhaltung
Mailing-Liste
Physikalisches System
Datenreplikation
Rechenschieber
Dienst <Informatik>
Transaktionsverwaltung
Menge
Automatische Indexierung
Mereologie
Debugging
Server
Stapelverarbeitung
Tabelle <Informatik>
Subtraktion
Prozess <Physik>
Quader
Formale Sprache
Klasse <Mathematik>
Implementierung
Kartesische Koordinaten
Extrempunkt
Computeranimation
Internetworking
Demoszene <Programmierung>
Streaming <Kommunikationstechnik>
Prognoseverfahren
Front-End <Software>
Minimum
Coprozessor
Speicher <Informatik>
Softwareentwickler
Default
Statistik
Datennetz
Applet
Mailing-Liste
Physikalisches System
Zeiger <Informatik>
Biprodukt
Gerade
Ereignishorizont
Unabhängigkeitssystem
Arithmetisches Mittel
Dienst <Informatik>
Transaktionsverwaltung
Last
Automatische Indexierung
Rechter Winkel
Heegaard-Zerlegung
Hypermedia
Client
Ablöseblase
Wort <Informatik>
Hill-Differentialgleichung
Stapelverarbeitung
Dämon <Informatik>
Tabelle <Informatik>
Bit
Punkt
Versionsverwaltung
Kartesische Koordinaten
Benutzerfreundlichkeit
Komplex <Algebra>
Synchronisierung
Computeranimation
Übergang
Metropolitan area network
Stetige Abbildung
Client
Datenreplikation
LES
ATM
URN
Datennetz
Fuzzy-Logik
Datenhaltung
Computersicherheit
Hochverfügbarkeit
Abfrage
p-Block
Zeiger <Informatik>
Biprodukt
Konfiguration <Informatik>
Arithmetisches Mittel
Transaktionsverwaltung
Dienst <Informatik>
Menge
Einheit <Mathematik>
Rechter Winkel
Konditionszahl
Server
Information
Software Engineering
Standardabweichung
Lesen <Datenverarbeitung>
Telekommunikation
Subtraktion
Folge <Mathematik>
Gewicht <Mathematik>
Kontrollstruktur
Quader
Implementierung
Autonomes System
Dienst <Informatik>
Mathematische Logik
Interrupt <Informatik>
Wiederherstellung <Informatik>
Datensatz
Gewicht <Mathematik>
Bus <Informatik>
Indexberechnung
Störungstheorie
Transaktionsverwaltung
Touchscreen
Einfach zusammenhängender Raum
Digitales Zertifikat
Zwei
Datenreplikation
Summengleichung
Quadratzahl
Last
Mereologie
Hypermedia
Hill-Differentialgleichung
Computerarchitektur
Zentrische Streckung
Quader
Exponent
Kanalkapazität
Dienst <Informatik>
Computeranimation
Wiederherstellung <Informatik>
Eins
Stetige Abbildung
Diagramm
Dienst <Informatik>
Transaktionsverwaltung
Skalierbarkeit
Rechter Winkel
Last
Eins
Störungstheorie
Personal Area Network
Software Engineering
Lesen <Datenverarbeitung>
MUD
Resultante
Proxy Server
Bit
Subtraktion
RPC
Quader
Gruppenoperation
Kartesische Koordinaten
Datensicherung
Computeranimation
Datenhaltung
Übergang
Differenzengleichung
Metropolitan area network
Datensatz
Skalierbarkeit
Maßstab
Arithmetische Folge
Proxy Server
Front-End <Software>
Hash-Algorithmus
Speicher <Informatik>
Optimierung
Transaktionsverwaltung
Lineares Funktional
Parametersystem
Datenlogger
Speicher <Informatik>
Zoom
Elektronische Publikation
Partitionsfunktion
Algorithmische Programmiersprache
Transaktionsverwaltung
Dienst <Informatik>
Menge
COM
Last
Rechter Winkel
Server
Kantenfärbung
Ordnung <Mathematik>
Normalspannung
Instantiierung
Quader
Varianz
Physikalisches System
Binder <Informatik>
Partitionsfunktion
Computeranimation
Demoszene <Programmierung>
Diagramm
Dedekind-Schnitt
Prognoseverfahren
Datenmanagement
Menge
Last
Bildschirmfenster
Datenreplikation
Server
Projektive Ebene
Cluster <Rechnernetz>
Tabelle <Informatik>
Proxy Server
Subtraktion
Quader
Gewichtete Summe
Kartesische Koordinaten
Dienst <Informatik>
Computeranimation
Fluss <Mathematik>
Unendlichkeit
Domain-Name
Dedekind-Schnitt
Datenmanagement
Software
Primzahlzwillinge
Datenreplikation
Weitverkehrsnetz
Softwareentwickler
Bildauflösung
Konvexe Hülle
Rohdaten
Klassische Physik
Abfrage
Ruhmasse
Schlussregel
Zeiger <Informatik>
Satellitensystem
Entscheidungstheorie
Dienst <Informatik>
Flächeninhalt
Forcing
Last
Server
Hill-Differentialgleichung
Schlüsselverwaltung
Lesen <Datenverarbeitung>
Virtualisierung
Gewichtete Summe
Implementierung
Dienst <Informatik>
Extrempunkt
Term
Computeranimation
Data Mining
RFID
Prognoseverfahren
Reelle Zahl
Trennschärfe <Statistik>
Datenreplikation
Störungstheorie
Softwaretest
Datenmissbrauch
Computersicherheit
Winkel
Güte der Anpassung
Automat <Automatentheorie>
Datenreplikation
Diagramm
Wiederherstellung <Informatik>
Computerarchitektur
Hill-Differentialgleichung
Wasserdampftafel
Mathematisierung
Gewichtete Summe
Service provider
Computeranimation
Richtung
RFID
Informationsmodellierung
Digitale Photographie
Datenreplikation
Störungstheorie
Ereignishorizont
Bildgebendes Verfahren
Binärdaten
Varianz
Datenreplikation
Rechenschieber
Prädikat <Logik>
Verknüpfungsglied
Ablöseblase
Server
Projektive Ebene
Richtung
Hill-Differentialgleichung
Refactoring
Schlüsselverwaltung
Tabelle <Informatik>

Metadaten

Formale Metadaten

Titel Implementing High Availability with PostgreSQL
Serientitel PGCon 2013
Anzahl der Teile 25
Autor Fontaine, Dimitri
Mitwirkende Heroku (Sponsor)
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 10.5446/19050
Herausgeber PGCon - PostgreSQL Conference for Users and Developers, Andrea Ross
Erscheinungsjahr 2013
Sprache Englisch
Produktionsort Ottawa, Canada

Inhaltliche Metadaten

Fachgebiet Informatik
Abstract Availability, Durability, Architecture, Replication... explained How to implement PostgreSQL in a demanding project, what are the different technical offerings good for? All you wanted to know about replication and never dared to ask. PostgreSQL includes several High Availability solution, some replication solutions, and some external Open Source projects complement the offering. When to use which project and what for? This talk will present the usual needs you want to address in a medium size project and how to use several replication solutions to implement them.

Ähnliche Filme

Loading...