A look at the Elephants trunk - PostgreSQL 9.4

Video thumbnail (Frame 0) Video thumbnail (Frame 13869) Video thumbnail (Frame 25021) Video thumbnail (Frame 27332) Video thumbnail (Frame 37847) Video thumbnail (Frame 41191) Video thumbnail (Frame 52635) Video thumbnail (Frame 64079) Video thumbnail (Frame 75522) Video thumbnail (Frame 81172)
Video in TIB AV-Portal: A look at the Elephants trunk - PostgreSQL 9.4

Formal Metadata

A look at the Elephants trunk - PostgreSQL 9.4
Title of Series
CC Attribution 2.0 Belgium:
You are free to use, adapt and copy, distribute and transmit the work or content in adapted or unchanged form for any legal purpose as long as the work is attributed to the author in the manner specified by the author or licensor.
Release Date

Content Metadata

Subject Area
PostgreSQL 9.3 was released in September 2013, but the development of 9.4 is close to reaching beta. This talk will take a look at some of the things that are available in what will eventually become PostgreSQL 9.4
Group action Building State of matter System administrator Multiplication sign Execution unit 1 (number) Sheaf (mathematics) Insertion loss Parameter (computer programming) Mereology Perspective (visual) Information technology consulting Endliche Modelltheorie Series (mathematics) Extension (kinesiology) Error message Social class Covering space Compact space Pattern recognition Email Software developer Electronic mailing list Bit 3 (number) Variable (mathematics) Measurement Connected space Category of being Message passing Arithmetic mean Data management Process (computing) Order (biology) output Cycle (graph theory) Quicksort Procedural programming Arithmetic progression Filter <Stochastik> Point (geometry) Web page Statistics Functional (mathematics) Computer file Open source Patch (Unix) Data recovery Branch (computer science) Field (computer science) Number Element (mathematics) Revision control Natural number Term (mathematics) Authorization Energy level Software testing Traffic reporting Weight Projective plane Generic programming Line (geometry) Cartesian coordinate system Particle system Spring (hydrology) Query language Network topology Blog Video game Table (information)
State observer Presentation of a group Group action Context awareness Code View (database) Range (statistics) 1 (number) Set (mathematics) Insertion loss Parameter (computer programming) Mereology Mathematics Computer configuration Recursion Physical system Area Sampling (statistics) Data storage device Infinity Window function Category of being Arithmetic mean Process (computing) Order (biology) Right angle Quicksort Procedural programming Resultant Row (database) Point (geometry) Level of measurement Divisor Valuation (algebra) Rule of inference Number Goodness of fit Zuschneideproblem Natural number Energy level Selectivity (electronic) Standard deviation Projective plane Line (geometry) Multilateration Cartesian coordinate system System call Personal digital assistant Query language Table (information) Window
Presentation of a group Dynamical system Concurrency (computer science) Code Confidence interval System administrator View (database) Multiplication sign Materialization (paranormal) 1 (number) Insertion loss Function (mathematics) Replication (computing) Front and back ends Mathematics Velocity Semiconductor memory Endliche Modelltheorie Extension (kinesiology) Physical system Theory of relativity Cross-platform Shared memory Sound effect Message passing Process (computing) Buffer solution Configuration space Text editor Quicksort Reading (process) Reverse engineering Spacetime Trail Server (computing) Functional (mathematics) Student's t-test Rule of inference Number Product (business) Workload Internetworking Operator (mathematics) Queue (abstract data type) Operating system Utility software Data structure Computing platform Form (programming) Stapeldatei Physical law Database Subject indexing Cache (computing) Word Integrated development environment Logic Personal digital assistant Object (grammar) Table (information)
Axiom of choice Group action Context awareness Water vapor Shape (magazine) Different (Kate Ryan album) Hypermedia Data compression Physical system Structural load Electronic mailing list Database transaction Message passing Process (computing) Befehlsprozessor Hash function Order (biology) Quicksort Figurate number Arithmetic progression Reading (process) Spacetime Point (geometry) Vacuum Computer file Sequel Syntaxbaum Regular graph Event horizon Rule of inference Scalability Number Crash (computing) Computer programming Operating system Queue (abstract data type) Traffic reporting Computing platform Form (programming) Standard deviation Graph (mathematics) Information Interface (computing) Limit (category theory) Cartesian coordinate system System call Software Integrated development environment Query language Personal digital assistant Statement (computer science) Table (information) Library (computing) State of matter Code Multiplication sign View (database) Sheaf (mathematics) Combinational logic Set (mathematics) Insertion loss Parameter (computer programming) Replication (computing) Mereology Prime number Food energy Bit rate Semiconductor memory Chromosomal crossover Square number 9 (number) Position operator Partition (number theory) Area Data storage device Variable (mathematics) Connected space Type theory Configuration space Right angle Resultant Functional (mathematics) Backup Server (computing) Overhead (computing) Service (economics) Divisor Data recovery Field (computer science) 2 (number) Revision control Workload Operator (mathematics) Software testing Plug-in (computing) Condition number Installation art Validity (statistics) Inheritance (object-oriented programming) Planning Database Software maintenance Object (grammar)
Point (geometry) Dynamical system State of matter Euler angles Patch (Unix) Multiplication sign Range (statistics) Execution unit Set (mathematics) Online help Distance Mereology Perspective (visual) Field (computer science) Order of magnitude Number Revision control Workload Frequency Negative number Software testing Endliche Modelltheorie Position operator Computing platform Area Key (cryptography) File format Digitizing Interface (computing) Data storage device Sampling (statistics) Line (geometry) Limit (category theory) Benchmark Subject indexing Personal digital assistant Factory (trading post) Universe (mathematics) Quicksort Table (information) Local ring Window Library (computing)
Figurate number
also to close the doors to get some level of error in conclusion that we have to some so of holding you have been here the whole day until all the talks but that's illegal and so i minus sign a I'm here today to give a talk about what those was known for look like 3 days ago but things have happened in those 3 days on my part and what they did everything that it would look at them approximately as would be so so let's hang on the political 14 member and the members who has not done as much work as he should have on 1st model for other people have told moral weight this summer but that's to be expected sometimes around those working with Europe which is why you'll find a look over our table in the AW building effective there yet and look at our merchandise and of us questions such things the working on variables 1st coworker comical rental involved in Stockholm Sweden where we do of generic open-source consulting work FIL can obviously my folks suppose 1st the and when they're still name start by actually going the to 1st line states of his model for coming here are already running was wonderful that is surprisingly many minds race that's nice noted that when 1 of 9 0 8 4 and helping the units of work to do the things they will for this and the light in June this year you need to upgrade by the page 3 is already spread and 7 for the purpose of a human OK that's some point there so that in particular those to be wrong Pose procedure for some work on it might start working on it now because come June you will no longer get about fixes for any support on the part the there are a lot of people were still using it so definitely that but today we're going to look at the final 4 which is some of you in fact it's not there yet as most of you probably know this money for the release of the upcoming worship not to the quiet because of it PTD compact plant of authorities because this blog aggregator under a couple people there who have done a lot of talking about the new features of both the from very technical perspective but they basically take the commit message read it and then translated into something that other technical people and understand that in reticular and that the question that affect packing written they each built around series of the the gold gold waiting for lined up for which is every time a new major feature goes and gets there so that i stole the things from the because that makes life easier than that of course not accidental actually wrote all these features and persistent you things coming and where we actually or with this model for today but some of you may be familiar with the terms of the cycle so we suppose this development which is again for a release once every year that we promised all those with 103 being on June 14th last year and that's the master branch became books now don't turn into 1 of 4 and we open it up for the development of new features that it was as we work with a cycle basal equipment fests which is basically we take it is the original ideas would take all of the development and I would take a month to act like I'm not actually managers think that gives you never finished in 1 month and they overlap and cycle that sort of the basic cycle but we're working so where you rolling for commit for typically without 1 gene 1 September 1 November and all that sort Syngenta I think which started all the commit 1st on time and will finishing of on of this talk about that's very informal and articular the idea January connection which is where we are right now even lower now if ever as you'll learn when you got your mail reminders this morning based January 1 is expected to take more than 1 because I I I I if you if you're past the small get really needed to be in the generated 1st you are not going into 1 of 4 so the status right now is we or in this final in test so we started on January 14 so we've been going on for a bit over 2 weeks uh were actually making good progress when the number of patches here I think almost a 3rd of them have been handled so of course we handle the garden that's easy and fast and many of the big ones there are still pending but the idea of this 1 is it continues until the end so if you have some spare cycles in particularly if there's stuff that you're interested in the AI there is when application commit fast was or we get a list of pending patches new granted to ballot applied test just applying it make sure it builds still helps just close the report the main is thing or try this and I have this passionate tried to run a query example and people will know that that helps everybody with the development so that we can actually gets 94 out of but was that the the statistics you because you know you can get interesting stuff out of it a recurrently at approximately 2 thousand files changed in order for the 1000 insertions and 55 thousand deletions of this is actually significantly lower than monetary which also a little bit lower than 1 or 2 but we'll low-lying town is the best possible way to measure developer project in light of support to that there somebody made using some also we're less than halfway into the final that that's there some very large patches waiting so volatility there's a lot of work already gone into demand for the course will be more of it so try I always try this for for every version and I'll try to figure out what's categorize all the things that money into groups and I think this is pretty hard to actually try to find groups for them and things in the 1 year elements function groups most performance and other things so Canada and so on until we can go through a laundry list all some of the things the and obviously we can't cover everything just 45 minutes but some major features some the minor features that may make a major impact of all of the sort of little or no skill features on the course that's developers building against phosphorus we don't cover too much about and we get infrastructure which is for developers for building process often building extensions into best so sort of a more core level of the the administration saying it's all the glory of back to doing some recognition recovery of a kind of 3 was very unusual version that actually have any major reclamation features inputs is 1 of 3 they said no we added some more stuff well let's let's start at the beginning with develop natural features actually let's start right so how many you would consider yourself a developer using prospects and the more of the DDA administering processes 1 has to be here what's left in the field developing and not administering that marketing but from OK so so develop this is like the SQL level of new things we have some interesting use of a smaller things that a bunch of work around added it's that will help a lot of people that we allow variadic aggregates we don't ship with any would have the same discussion about other functions were upset enough where the conditional them because it's really easy to get confused but don't you want confuse yourself go ahead so variadic means for those of you that basically aggregates with a variable number of arguments right now you have things like the spring i which takes 2 arguments in the field of whatever separated you want but you can also do variable number or it's true that is an interesting improvement of the explained that section very useful properties to get to keep in your explained last which you do before Seattle look you explain from the old that looks like it came from which we went through buys the death of that will actually tell you very simple things that can help you review process for something to help you particle class is that we now have support for filters in aggregates tree will
build solutions that is the built this way you build solutions will also will select something in a case when something and 0 or 1 and so that I think just say something like this so called store filter where the greater than 5 I will then count only the rows where he's going to of course we can combine multiple these aggregates within a single point so those features that seems reasonable I'm not sure what did not it before but I'm sure there's good reason somewhere deep in the code the a very useful in particular simplifying the categories right because it's quite often this that you actually want to you know count number all approved versus of inventories they thought and filter will make that a lot easier no we also very big change when it comes to aggregate when you're actually know the ordered sets aggregators so 2 of you and you write notes on this stuff and you review the mutation this project so here we have an idea of the sort of the same I remember when I was this presentation of lateral last year did no lateral loss I knew what I knew it was sitting knows collateral and so the same thing with is likely would also we wanted but noted for it is in the SQL standard yes the point of the of is a while which is why do we want it polynomial support so this is basically a classifier if you look things that also inside of you the he's syntax is within groups and we also love something will hypothetical aggregates so these native fantastic hypothetical items that lets you aggregate on rows as that's what they would look like if they had no the act that the 1st and the last part of the reported so that it was a selected as a mold of within group order by the but we always have to have known about this gives us the most common value in the group the this'll stay for each day the most common getting quite useful actually doing that today it's much harder you have to what is a window heard Minnesotan CTD and things like that the other thing you get is you get access to percentiles so in this case it's a select means percentile call them 1st of all this is the continuous and discrete percentile 2 goods which rose is 30 per cent into this group so from the beginning move 30 presenting to the group which I was there again this is something you couldn't do before I will I'm going to get the whole set back in and try to reduce the the hypothetical ones are a possibly the most important ones sitting around for In this case as if there was a role that was exactly this foreign what 1 1 per cent right for presenting what would it look like even wrote was in there the so today's you take a range of values and you say well all the exact several that you get that even if there's no role there what does need you rotated about you that you would have in that whole for that role the results shown this value to the group you don't have to provide you can do it over the whole table that's the same as a group I nothing original have multiple levels of about was you can combine the most of these can also be used together with a window function I'm sure that's also what happened from good example yet the but but opens up a new way of accessing the without having to read back the whole groups of data you and your application or maybe just for procedure and process it means that you can just let the data the character is usually the thing you move the problem to somebody else's problem and just thing that's the whole idea of a scholar you tell what you want and someone else figures out of it but hopefully fast and then we live in is the imperfect world where we have to start tuning things but in the original to do that so we have significantly improved the handling all updateable views that we support partially available use now which someone makes a much more useful so this is the automatic update use so you can all you will always be able to do your own obtainable using posters by creating a bunch of roles that in 9 3 you could and you will get the automatic rules if you use or a simple white select promising technique but when you have things like join it rapidly becomes small available because the system doesn't know which row in the table that corresponds to get in line for is you'll still be able to update the rows or the columns that are from sort of the master table even if we don't know where the other ones are because they're affected by this so we can update some columns even in the case where some of them cannot be predicted and therefore not be automatically that will so they less valuable work or if you need to update the ones that you can't define which they are well you have to start by defining which they are there is no existing completed up but we've also about sample with check observation again automatically updateable views that will have you verify value so in 1 of 3 if you have say a view that selects store from my table where group for you can actually insert into that you would a value groups that small for and then you get a throwback from and which may or may not be what you expect if you enable with check option it will let you do that so would you say we check option local it will verify on that specific using well you know you users group equals 5 I want them to see that growth so you don't get to and we have we check option cascade which will look at other user given as news and do the same thing you will no longer be able to insert a row the that you can see but sometimes you actually want things to be able to insert process can see depending on how users the works just don't specify with check option you still have your behavior if you chose above of the um is sort of the the aggregate size area we've added the argument honest working will erase let you on MNIST to erase into a single table the even another completely independent this query will return a table of 3 rows from which is a 80 1 which is medium 1 to see at and obviously does not the actual erase it can be subselects things like that works on the you can also call this with a keyword that's
with ordinality the that means that for each row that's coming back from this problem nest this will just add the row number the the seems like that it would be easy but it's pretty hard to actually do that today have to create the temporary sequence and if the value of it's also quite useful book they should just selecting this in reading about Interapplication because you can count rose from for you not do that in your application is if you joining on something else that the ordering goes away this way you know the original ordering you get or you get generate call that you can then they order your final results something like that yes for upon of you like writing procedures and gesture the outcome of your only about trolled debugging them because you don't know where the call stack your and and now you have to have a stack trace get diagnosed external return status using beauty context and that will be part of it and that would entail the useful but it helps when you do like some of like accidently infinite recursion stuff like you can find out what happened that just simple factors there are some other improvements to appeal against all well that's 1st I found this to be the big 1 that we have so for no it's going to help that's so that's going to few the infrastructure
things that you're probably never gonna use as he knew use of somebody else is going to use them and give you a nice um how many actually built for the people in Europe have built a back-end side extensions prospects in see I know that the 5 or 6 people in the well this is for you for estimating it would not valid we know something we call dynamic background workers 103 3 go back from workers which is just utility processes started by both 1st but they all have their little while you know like we have already things like were we have this this is the celebrities our system back on workers basically they could only be started with POS present would change anything of resources that now you can start and dynamic you convert so you can start from as you all you can have impact workers or other and workers solve all at once it gives him ability to build a more logic that runs basically on the server gets access to a bunch of reasons the alongside with this and also very useful way for these background work things is the without dynamic shared memory the name of the shared memory segment where we keep all of our buffer cache those things is the dynamic you still have to restart the server if you wanna change a cat sorry but we might get the thing that you can old picture number all request in the background workers so that we can put it can the for actually able to communicate with each other in a much better with this this is all about building an infrastructure that we so far have nobody that's using other than example but once it's all there it provides an ability to write things as an extension of the previously have possible post was so hopefully you will find cool things to do that we also have a likely message students builds on top of this so you can have a message queue between your back on workers for example for between a regular bacteria is talking to your back from work something that will effect you more that is really deep down just a code change from the internet was structured now no the snapshot now always sort of posters artifact whereby when we stand and use our system tables we haven't been using C the multi-version recursive involved but we're thinking about a lot so we have some really strange Visibility Rules for some of this is all well being changed and all using CC the same way as the rest the systems and special about doesn't exist anymore see any 1 of you had your hand up has actually build an extension that uses natural your will break that a lower comply that sentential because national doesn't exist but there are other ways to do similar things but we wanted them to break so that you know that they've actually change there it was a simple moral structure it involves form more products specialized there are a lot of code that have very specialized access paths for these systems could relations just to deal with transnational now uh and in the future and that's an ongoing process that velocity resulting required for a number of the operations but less looking for all table today was still mean we don't all the table itself depending on what you're doing but still persist in tables this will allow us to make less loss on the system tables and allow greater concurrency in this idea operations as well a going on into the DDA and administration the suppose this 3 added materialized use how many of you are using them nobody how many of you are not using them because you can't reversal without loss no you can uh you cannot do refresh materials you can grow a refrigerant and as you can concurrently and then you may I was the mineral water myself and of perfect personally say that the material you that we have imposed on the 3 were not very useful particularly because you couldn't repression without the exclusive you could even be done while they were being refreshed referred to earlier you concurrently lets you keep doing stuff you that you read them while they're being refreshed you can only do this if you have a unique index on you the the because that's how emergent changes that happen what is being refreshed but hopefully we made an ensemble foreign due to remove that restriction but fairness the indexes we usually have somewhere and once you do that you can refresh your materialized view without having any law without blocking any of your other operations pop up and irked effect but it thank tt looked at the the top and that a little over here change now you can see supplied when you create of possibly more useful 1 the data actually in an environment where using a low tablespaces is that you cannot move objects in Table space is not just 1 by 1 the job table faced with a honor tables and the thousand yet move them 1 by 1 if you want them to different tables this for now you can just say all the cables based move all over here on the number indexes over here always you can still move the individual tables if you want to but this will let you know all of them as 1 the batch I think is that the James Frazer across a wide but obviously but you will need to be in the database that contains your table and you don't know otherwise we can find out that still exists on the impact that it could top not because you see the ones that are in the fact is that we we don't know that they exist at the end of the day that you will notice when we try to drop the tablespaces as notice of here you need to move this thing my my track reasoning time my it is single threaded little 1 by 1 of to you could probably do it multithreaded on their own by basically doing all available but this 1 will just you sequentially 1 model the about right these are the out of there we get it will the model on the on the very but
with small simple to local pre-war you if you need to pre-war cash other tools that can do similar things like the commonly used PTM think or 400 years but 1st of all this is sort of those that you get a new function 1 is all this will be prewar reconsider this table the leading to the operating system cache for set read ahead for the operating system cache were loaded into the process of catch of little just read the data into memory and and not do anything with that for a simple but something about useful for planning to replication switch over to make sure you don't switch over to cold cash you just bring up a new node in the clustering 1 mature things are in cash before you put your workloads and hideous or workloads to put things in the past 3 quickly but what it's doing that can be slow different simple 1 but it can help you all lowercase cases an online feature the words is platform independent so it works on all platforms because uh except for the read-ahead hinting which only works if you if you apply from hospitals accepted advice which most them to accept the government funding that were not at all of all the others the we had the ability to James configuration the school we don't have editor conveyed in you know of course that confidence you can also older systems that and I will change for a while what it does is it puts it in half basically this Africa they call so we will now have to predict possible because of the home and this suppose the this full by 1 of them so basically just write them there it overrides whatever is in the public house overrides what you have imposed upon Sivaraman all the systems so as you change it in the post the postcritical follow the oldest consent will when go through it does not change it immediately you still have to be loved but in this case the you would
select the jury will call which is the load whatever is in the right now thank you OK so you need to ask the user to do this yes a regular user cannot all the systems that that would be that the in the the reuse of water for example a regular user consumes still give you set for them for itself but it can change the system before this all this is to use no are contexts still apply so parameters there are set for example for the required read all the required we parameters that requires the star still require restart but you can actually do that we start from a skill so if you create examples here share profits said the change that with all the systems that but then you still have to restart of the service from the operating systems and there are some tricky thinks that the big advantage of this 1 is if you for example within an invalid value in many fields you'll know before when you just added to the file you put an invalid value in work on but then you realize you know difficult for us see what happened here the all resistance actually you forgive yeah but in particular when it comes to parameters in restart context it Michael no so you can for example set it to invent invalid size but it's valid number but it just doesn't work or its dependent on something else so we can see it the you restart and it's still also doesn't come back up so particularly when doing all the system set all postmaster context variables you still have to be very careful does the character of a group pop up behind the last act like were job offers a terrible exonerated of gene they face a lot rules on events in the code is used as a test for this the fighting completely unrelated parts of the code on my job offers of a terrorist this where that is in the literature as well so right there still right next to the other 1 assumes that the form in which case it will still be the data Dr. where is your main 1 will be in the EC director because you know everything else would be way too confusing if had everything in 1 place but that can be the same everywhere for so couple new important or less important depending on the environment configuration parameters we have a separate from an Minolta backing workmen In previous versions of the documents used maintenance work on the problem being that all the vacuum can be used in multiple parallel sessions which makes this somewhat tricky thing to configure in large installations so we have to down and that wasn't high enough for other operations things like that it's just is the same kind of functionality is the same as maintenance work used to do for all of our human previous versions in fact if you set it to minus 1 which happens to be told then the the system still uses data for what it gives you the choice to control the or back in memory Chris separately from the other maintenance work so the only parameter called session the networks to locate its and we previously loaded plug-ins either which share preload libraries which low when you restart the whole cluster there's always this local programs which loads every time you connect the the little local Polaris is they can only be loaded from a specific subdirectory because any user can load them so they can only be trusted code so this is sort of a crossing between the so it allows you as the user in the config to specify a library to be loaded for every connection at connections stockpile and it also loaded before and from anywhere in the post this library directors they just sort of crossover between these 2 parameters the so the primary goal while it hits it was not really loss hint that changes throughout did not help anyone that's the that's the commit message so is useful for to face is useful for what call remindful of which are tools that will now probably be safe to use he was back a replica of a couple of transactions backwards in order to be able to connect to a new master but it has some still some limitations but of withdrawal dual was also gives you gives you a hint about what's your excellent overhead going to be if you enable checksums you know as as a like 3 we can detect some of the victims of course would you need to do that in time and depending on your workload it can come with a large overhead or small overhead this is a good way to actually measure the transactional part because this parameter you can change it as a result of this and it has the same amount of data from the transactional but the checksums would do but you will also have some more CPU overhead if you need the actual checksums but that's usually not a problem of scalability from some form of problem tends to be related to our and this gives you a hint all all time packed it is PG start statements that we the resting look at you should be using it's great and wonderful making better uh the 4 expose what we call the query ID which is the actual hash value all the query there is a tie act that look exactly the same piece that statements but there are actually different they can be different because they have different answer fast they can be different peoples and object pose from recreate but you can actually tell that from the information contained use that statements that the piece that statement could tell because the Newton different it just expose the information to you if the user that no doubt so this internal hash value in a field very ID which will uniquely identify the square nodes about no this is not stable across versions it's based only on the parse tree as called this based on internal state oppose 1st which is not the same in the online for for a lot of these things so you can't rely on this to be a long-term stable not it will also be different for the same query on a different platform on my i ty things like that the socio modification that will make it unstable as well but in general it gives you a much better view the reasoning as to why these various looks at and we're tracking varies over time using somewhere gives you visibility into what is actually the wasn't on the replication recovery features network that actually doing something in this area after having in the lower 9th where we focus on other things that get any recovery for it but I don't think so it's becoming refusing to figure I think we need to fix their face that's of the right now you have recovery you can set recover target XI because that without required time to recover target name and now you can also that recovery of equals immediate which is the only value that is at the very top the point of recurrent obviously the
media is to still restoring your back the the back of the Bayes factor you go through the you have all these parameters to tell you how far to go this primal will tell you to go as short as possible you just get the system opportunity can can that the typical use case for this is actually being able to use related back as a step back the it will recover to just midnight or whatever you take it cost exactly as far as it needs to go to be consistent is you just a story based factor and nothing else the database is not consistent we don't like that you prefer to be consistent so this will take you as short as possible get the system up and running as soon as possible maybe building a reporting slave off your back to something like us will get your system back up so quickly as possible who started lobbying Our transactions to the while at regular interface that enables you to keep susceptible to attempt that's what what what the pop a is to know about 1st section snapshots from the master previously would only ship based upon the kind of text what PCA I bet at all well and the problem is if you had a very long running transactions only for someone you master years later when it's going to get stuck in an inconsistent state and will actually start up it would start up of wouldn't let anybody connected the so what we're doing now is basically were just log in the list of running transactions every 15 seconds because if you actually were doing anything other system blocking that every 15 seconds it's prominent in any measurable and if you're not doing anything at all were not you if you're graph is completely unaware of the so make it more miles it could also more directly useful feature of we get time delay stand bias you can configure your applications like intentionally be behind the master normally will try to be as close to the best possible in asynchronous and look in the cases you can provide milliseconds after the sovereignty of 1 of the more than our half an hour something like this why would you want that well it's useful when you do something really stupid on the last of the fact that the important table normally what you have to use that you have a thing is it apart and so on and a the specific point just before you got without having to go back and restore completely from back get a large cities in restoring beautiful from back even the restoring with that independent of recovery rate is very fast if at all but a couple terrorized it still takes time this will be much faster so it's a way to do so 3 the lawn Dong for doing stupid things so you want you probably wanna use this in combination with an actual replica for the case when the server crashes and you don't want prevented from doing anything but it's also worth noting that it keeps replaying over more than 4 is actually just I all transactions you can get a very spiky load it's going to get reasonably even at whatever delay the use for the time delay so those are the main things that was going through this always more of cost there is a huge number of small values of the order of the thanked at or if I talked about the fact that it took a list of things that are in progress some of them were actually looking in progress but I was at a conference yesterday on a plane 2 days ago it was it's updated asshole my mean leaving the stock America was up today that what we're in the middle of that 1st of things happen the theft of the energy is being rejected the null that's a just a node that we get people working on triggers before in Tables and inheritance before tables and who knows for most useful because use inheritance for partitioning so would like you to combine partitioning of on tables of the problem you have a lot of interesting things that we get some fairly large performance things they're still in the queue we're looking at some of these reduced copies for all table that to reduce while value for updates just decreasing the amount of information in a while that for certain types of objects using intelligent compression and things like that the partial sorting but we don't need sort everything uh you get some people didn't expressed down because of geneticists can currently faster than their habitat available how it does that but no no no I thought I think several these are problems in very good shape forgetting applied the living working on my while but in the back up space were looking at things like the fulfilling being able to make a backup go slower to have less impact on the system was running and being able to relocate tablespaces base back right now if you use the duties that would will fill spaces it's a really 1st that you might want tables this mean exactly the same position on every node but you think that that I know the last 1 has been applied previous that are higher gives you you into your log archiving tells you how far along the system is what it does uh 1 of those that we might get you don't get your hopes up too high it's been debated heavily for a long time can be held insert into all duplicate for updates or some people would like to quote up search and the other the longer end because you know the single standard lost it I don't think it's a sequel started the long name I think that's just here of the of the signals that have March threats that this is the Irish it's not all of but this is intended to cover the most common use case which is exactly you should be able to insert into a table you can do this for most of the existing table uh but the problem is there's a race condition you insert into the table and say 0 that's a duplicate the updated them that somebody has already made it and that's definitely a race condition that's basically what this thing I so that you can combine this with a writable CT get up search so we will have a statement because that's where instead of you will the combined it with credible cities and the nice the so all duplicate people for object we might get a shorter version of that and of course the feature that everybody's waiting for but not everybody but a lot of people were hoping to see a sort of to happen in your using a sort of a of how many the know of a star but just learned using it for the rest of you need to go read up on a store them they on today it's just a key values for a single call
it's fully dynamic however you don't need to create index for a key you can search for anything you want they start to allows us to nest them so we can have hierarchical storage we have achieved that points to a value that is another of store endocasts directly Jason and we were looking at adding a new data sample Jason B which is binary Jason but uh but the thing is not directly on top of each store so that would give you dynamic Jason indexing where you don't actually have to declare today if you wanna index adjacent field have to declare which he want to index all 1 and that's for each of them we Jason they you don't have to do that you just say here's where Jason just index that new fast access but we don't know how much of it we're going to get we don't actually know what we're going to get any of its but we were lucky we get all that that's going to expand some really really the good on the capabilities of storing the schema-less lesser dynamic schema where you wanna call it that sort of data document storage key-value storage at 2 point this is the features some of you have probably seen it's some given to residents among Dublin of where the guys went in the evening when everybody else 1 for beer solid shown number were among the indexing was faster imposed by so there was success is the sound and the fact is that this 1 in their benchmarks obviously this is 1 that works but it was significantly faster than what would you the for case and the will see exactly where the big difference between adjacent adjacent be really from an interface perspective is going to be used for the format adjacent adjacent today impose present comes back in exactly the same format because we actually just possible restore your test Jason V. will be deported to pieces and then put back together when you read that it will not have same indentation or whatever that you're using but it's syntactically also the same days time-invariant always try to find those it's like this microscopic patch of 4 lines that dates make some people so happy that I would like loading is no debug 1 you anyone is used local preload libraries knows that every time someone connects your data is it lost every library lots of and there install post this on Windows for example knows that appeal divided us justice all that stuff is not debug logging attitude so that's a little less factory problem help if you have a problem because of the and the is of just because repose trust that the the novel also to support year numbers greater than 5 digits the normalized so hormones that would be supported of courts but if you if you like the we're backwards American date formats or something that we actually only supported for years states well now you can have 1 4 year dates they attempted i ridiculous limits but there was no reason for the linear to be there so now there is no and I kind of feel positive and negative years it needs to be within the range of times which is no going back to what a Member it's 405 years before Christ I think an 800 thousand years in future of America and so I don't think it might actually applied to negative because plants that have sort of part of and those were the features of obviously there's many more that we are currently in the middle of CS 4 when that is finished will put on the 1st day version if you wanna spend some time building it please go ahead and all the get ahead and try it out let us know now what about because there are lots of the earlier than you let us know that we can get the text if you don't go through that the universe is out there will be packages of the packages for all major platforms with your own Linux or Windows all of the other please download it right with your workloads let us know if things don't work let us know things were slower and will let us know they're better to that and test on the new features to make sure we didn't miss a quarter of and again if you can get this testing done before the release will that once we get the release you can just avoid production behind but they don't show up hi I'm going to look at this in a way that will go over the the the the units and that land and we use the same all the time and that this means that user in the model and a lot of the abilities of modern science settings of the 1 of the things that you need know the what the the the the eye of the MS in the war that you have you that have to do with time and the magnitude of the the the the name of a really what you have to you know that you time you on the on each of the 3 and a half years of the time you got the in the reality of the matter is happening on the handout that I gave you the only way you have I think that was know the the the knowledge in the in the in the time for all of the things that have the original in the the army and the and the and then you have to have kind of things that I don't know if we want to be in this whole thing I think the it user so that we don't think that there is a lot in the in the in the in the in the in the end of the you think it's in the areas where the fact that the that the period of the of the of next best thing that you will have to the case and that's the end of the of the of the of the of the of the in each of those who was you know can also be to the model so what you do with those that means that the probability of it is you're going to be in the distances in a
the gene and this is what you do but what about all of the tables for will be that would be more of the of the of the of the of the of the and the word and and and and and the end of the day and the and the and the end of the year and you and in the end of the day of the year and the and the and the and the the the half of the and the the we and you around a OK and you you you you you at the
you know this of the the the world and and so on and so and and then the the it and the and the and the the and and the and in the end the and the the end of the the it the the the the that what 1st and and the and the this could have the and the and and and the and the kind and the the the and the the and we and the the and the the and the the and and the and the the and and and the and very Ireland I figure since I work in a review of