Bestand wählen

9.4 On The Floor

Zitierlink des Filmsegments
Embed Code

Automatisierte Medienanalyse

Erkannte Entitäten
we believe that we we use the we we are we really need to be use of the information the only thing you need to so to I think so finally maybe that's maybe why would have done some cocaine 1st and we have to do is to we need to the a galaxy here because we can start to think that way the OK I guess we then can start living every day and I know the economy's will you can be a close by all the detail a cave as that comes the middle fingers you putting up and with down again and becomes medium again said that the a little bit little bit of its own 9 4 on the floor this is the title as suggested by just burgers there's a talk about 9 4 features in case it wasn't clear that would women also mycolic 9 4 out the door but that would require actually having it done and that's not done yet so that we are doing that so quick poll having people actually use PostgreSQL again I see you guys are at the right conference that's a great start and how they have maybe you use prosperous 9 3 I'm sorry about that so that's a 9 4 9 was really exciting again going how use of PostgreSQL older than 8 3 OK laughter and more than 1 and this is what I call a just where you has my 1 of the operating and as it turns out there's release coming up that you should maybe do so I
don't want to do so I am your humble those Robert tree you can find me on the Twitter or on my website if you if you want a copy the slides I will post the most likely there or send the of the tweet and I will send them to you or something along those lines and all my background
becomes the 1st President features and all that stuff I've been doing press press for a long time been involved in a number of different things I guess I still work under stress occasionally and I coming and harass other people in their pressing me at the start but also a lot about root beer and crops informants so if you're into those things come and find me afterwards and will have a good time can vary from 0 and but my
more technical background when it comes to pose pressing why care about features of the company I work at a meeting I with you when when consulting and helping build and run large websites that we don't always get to pick the database we work with but when we do get to pick it we always because stressed because of that over the years we've got involved in a lot of press press consulting so every time there's a new release it comes out you know we study and break it down try to figure out like do we want to upgrade to this or not but if it's like 9 3 then we say no and if it's like 9 for think we say yes so so we're pretty excited are actually much more say about this release that was the last release so all the good things it's said about 9 3 if you saw that talk last year I really mean them this year so so with that let's talk
about the stress up the stress as you know we divide the annual development cycles as features a get built of the
basic path of of this from last year in June of 2013 we break the code and and started 9 for development and we do a series of coniferous something people familiar with the idea commit 1st event not not huge amounts of basically the way the process works is that people do not whole bunch of patches of they plan for them to be in a commit 1st which is scheduled throughout the year during the 1st people who normally a writing code writing features focus on reviewing the existing patches that are out there right and the goal is in theory all patches of accumulated at that time will be reviewed and committed or possibly we sent back with feedback as here's why your patches and getting committed right so every other month they go through that process and then the hope is that at the end of January but when we get into the 4th Committee everything will get reviewed and committed and put into the new release but what usually happens is that last and 1st in January stretches on forever I guess about 4 months to were in a while everything story gets cleaned up and people really decide are we have with these features that are not is this you know worthy and ready and that doesn't need more development and the user takes for 5 months for the next for the 1st dated come out which actually has happened so next slide says this happened just last week but I think to the right that was the 15 so 9 beta 1 has come out hopefully most of you knew that if you didn't you know what you do know now so encourage you to go out and download and testing and I got the beta and even on Mac courts which I'm perhaps maybe 1 of 3 people use backwards but even map courts at the data and so you have no excuse to not be testing the dataset from there's actually if you're wondering how do
I tested data presented a wiki page on this on the press with the of how the test of of beta of how to beta test I would encourage you to go check that out and basically it's you know don't look at the new features that you're interested in chapters out when existing applications against the new version of PostgreSQL look for 1st semester things maybe sort of replication see if you can break it many people to go and try to do that but we certainly encourage people if you want help raise questions like you have about 48 hours to define the experts and track them down on the features so as to get as good feedback and you get that's bit is not out with data comes what should probably be the final release 1st feature set goes it's possible some of the things will not make it in there and I will also admit that it's possible so things in the slides are actually not in there but I think that they are I did try to call out features that have not removed so if if you're a hacker you see a feature that is not actually in the release please let me know and I will strike that from the slides but I think all this stuff is in there I think all this stuff will actually make it in there and and so hopefully this is a pretty good overview of what you see in for but where do they
stop so that the folks to do them we have a huge list of features that we make up arbitrary groups and then we 3 features into those groups so many teachers can go in multiple groups but these are the ones that I think is a single hopefully you will enjoy them as well so on on with the feature on slot so also with performance stuff in
management then so that everybody likes more performance I think most people don't want less performance of the PostgreSQL so the 1st feature of reduced while for updates as as you probably know when you do updates you writing data into the wall stream that has to go all over the place depending on what you're doing with the wall of that connection become a bottleneck because the I O issues in that so the the size of all records that are being generated when you're doing updates and so as they reduce the size of that you can jam or stuff into the wall increases the ability to to do more I O and so that's the tradeoff for the performance right less well means website so should be less resource usage in general helps enable more have have right activity and nice little little update for free
counterpoint to that acts like insert scaling is another thing that's been approved and reducing lock contention for wall inserts but always with the big things that people are looking for for performance improvement and then you can eliminate lock contention and so this is what was once again where they went through clean up a little bit there improves scalability wall system by doing so everything and that it that everything passes through a lot of stress and trying to scale it boils down to you know what what's going on with your right hand logs and then the amount by which have to deal with their arms that can often be a bottleneck so improvements in that's generally how about most use cases that are out there another performance
helper is something called PG pre-warmed and this is what I said to contribute some sources say it's an extension of so this is an extension of the I create extension feature pre-warmed will solid on your system so that the goal PGP 1 is basically a set of utilities in order to bridge pulled the files and in indeed off of this and push them into either your operating systems and catch the fastest cash or you can pull it in a shared buffers right if you can do the index well it will do that and so and then disorders and differences in how you may or may not want to do that usually where people tend to use this type utility in the old way that these to be done but let's say you had to do you know like a kernel operator something like that so you reboot your system you now blown out all file system cache on our shared buffers yellow processes need to get up and running the way that we would tend to do that past if we wanted to try to get those files into memory before you start reading it with queries is try run like a vacuum on the table or something along those lines is easy to pull out into memory the public that obviously is that the vacuum is not necessarily guaranteed to touch all pages of Table certainly larger so this sort of gives you a way to do it from the SQL level really targeted here's the relations that I want but there's also this this more certain you want is to how you can call this but the main things are if you wanted in DOS catches 2 ways to do that you can get a sink which applied in the background so if you're watching up and and trying to pull some stuff in but you need to start something queries right away you might wanna do they think that you think it'll be faster but obviously it is much more ahead on I is more things here right into memory and usually I think you probably want to go into the as opposed into shared buffers but and there may be a few tables that you have a you know like these are real hot spot tables they always a shared buffers oppose dressed in that case you can pull those directly into shared buffers so I sort certainly makes it i think easier especially cross-platform deployment so that everything easy way to get the stuff where you really want to go so that that's a nice extension that's been created
but another thing and then and I tried to track them more information on this actually thought was kind of hard to find this but about a number of gene index improvements that I think are actually fairly significant and that the 2 sort overarching things is a change to the page format for how gene indexes restored right this format has actually been compacted and depending on what your data looks like you'll get some merging 2 times in 10 times reduction in size but I from numbers that I was looking at something I saw 4 times reduction size seemed pretty easy to achieve so that's a pretty significant drop in the size of the indexes that didn't and have some there also Benson improvements and how how close Kristien's those indexes right enables speed-ups there's well certainly via queries that are have you know some frequent items and then somewhere items and you're doing genetic index stance on that you'll see performance improvements there and there's actually another extension called B-tree and I believe it is which sort of mimics what a tree does that uses John indexes for that and normally people would say like well the need stuff to be trees are good at a B-tree indexes that there are also many connected everybody uses and they're actually the benchmarks to show that depending on what you're wearing if you're doing a thing like queries with boolean values sort of a limited set of things that your querying on you may actually get faster performance with the June B-tree indexes than regular the trees so that that was actually surprising to me on the and so this is a you know that stuff really good improvement and and I don't think it's got a whole lot of press but it's certainly worth taking a look at I think they're going to be new cases where you may be more likely to use unit vectors where you think you should in the past but is also a talk on Jan that's going on Friday 415 by I was using the Russians but I think I t is also in the top so it's against the Russians and cultural issue there so that haven't I would certainly encourage you to check that out of the genome programs a pretty pretty good and I think I guess that I haven't seen a lot of information out there about it so stuff we were checking out up performance the
stuff I guess this is performance cycles performs because if you don't performance from different with list but it 9 4 Australian I for knowledge separated out planning and execution time when you do explain analyzed this 1 that just happened to get this really good explain analyze like those related to you probably won't ever see this when you normally looking at which analyzes of Basilica also we have basically 2 ones at the bottom we used to get was was total execution time right when you run explain also tell you how long the thing took to run with they didn't tell you was well how much time is spent in planning versus how much time actually executing the query right analysis 2 things are broken out so the 1st was planning time seconds execution time in most cases you're playing time should be insignificant in your execution time should be what you care about but it just so happened like this is a fairly fresh database that nothing had really been accessed apparently so when I ran this query I just happened it was like monstrously large planning time was like 0 that's the slide away from a talk then you can see like sometimes query that actually makes a difference right like when you have 10 explaining time if you try to optimize the query thinking like any to execute faster execution was of problems after I ran I secure 3 times and then after that like many times basically minuscule and execution time you know not not terribly largely there it's pretty simple worry about but this is always the largest italic how much time I spent in there and if you're doing you know what workloads where you really looking like subsecond worries and that type of thing it's certainly possible that spending a lot of time in planner and tonight pretty easy way to see that in the the another
explaining change by adding the grouping columns into explained and so here I'm doing query it's pretty straightforward right selecting a codes and counts from a table in your group by and the new piece that was not there before this this group the right of that area but I'll tell you like a what you actually grouping on and if you're doing grouping unlike functions or something like that will put the whole piece of information in there the thing that used to be in there that was similar to this like you would see that for certain index gains will tell you is him on the 2nd information but for some of the aggregate stuff that wasn't in there so that's been added in there now so if you look at really complex plans of obviously but straightforward examples good 4 slides and the more complex you have where multiple hatched area trying to figure out like which thing is it actually aggregating step this actually gives you know that pretty clearly that certainly a nice little bonus change another 1 sort
of performance the related media because issues like block weights and Convention against what would apply to really but you cannot so that the amount of information solvable and I think uh in previous versions where you would see you see the top line and then the detail line would be the right word tell you like this process on a lot and the other 1 not even at you know you get you get the top 1 that's probably about it as a lot more context has been added including like here's the actual query that waiting right and that's always been a really tricky properly you can get the process ID and then go looking like the selectivity and do a little looking and may maybe figure out what why it's you know which part of the relation is a hanging on within that but this I think I'd like you to where it was run and even now like the top all right so that's the the disk format there they sell you on the relation in doing so certainly what more information there than what we had in the past few million much easier to debug block way if you having those issues and so on and I'm really happy to see that the the sort of 1 4 for a number of years but finally having no such school so as performance improvements but again like what's nice about performance improvements of like you just get the new version going in like it's all magically there so what else is
new so number of administration related things that happen in your body being or would consider some your the guy has run the database organelles of the amulet like developers who like to find out the tragedy of a somewhat appear as well as the government away have to touch the database again to put a good mix 0 pretty good makes graph on to
some administration functions hopefully to make life easier for you run in the press alter system is a new command has been added in the and the goal Walter system is to provide a way to actually change parameters within the process the Conf from the SQL level and so this is what I would say the 1st incarnation of this and here you can see sort of a simple right of assistance of the workmen the 42 you realize 42 is obviously the right answer for all queries in the universe and everything around so you said that 42 but they have to sort of know and understand and I think will sort of season best practices of all here people get into 9 for what this basically does behind the scenes is that there are now essentially to PostgreSQL other cost right there's 1 that you hand and then there's another 1 will will be included did is machine-generated so we update commands like this the machine generated 1 gets updated and then that will override whatever was in sort of a hand movement of a hand edited once so did commit like this the problem is that you depending on what you're setting you either still need to reload the convict file restart server Fraser doesn't sort of magic solve that problem and so you probably won't Wikipedia we after you do this can change you may have to actually you restart if you have a good function that can like crashed back and like that's a really good way to restart the server from the signal level there's only going to control over extension for that so that's the is that the deal that's a hardware restart the database but this only slight change some of things in and set them from the sequel level the mice change of peace
that come people have ever used piece of so a force recognize most of you have your hands up so I know something like this that is like a really under use feature and I actually think that that this particular feature which is has not been there is 1 of the reasons why nobody really uses the set so the basic idea now is is so for those who don't know peace that in SQL you have a way to change the formatting output at the SQL right and use a piece I command access the set in order to change things like that so and so what you didn't have before which is we had for most of the other commands that is now just that such peace and hit Enter you'll get a list of all the different things you can actually change right and I didn't have before he did that that just would give you anything back so now you can see like here's what I'm actually using and different options and change but also they so that a number of linked open this up so you can go experiments the like what what type of things my 1 change from all displays like a common 1 if you know about this right so you can change and also not chosen all like have a view toward all of you not sure what you're actually looking at and a couple other ones record separators 1 I think but today so you can even play with that again output that you that you'd like to see the sign change for that but also in our should have completion so if you start type 1 and you can detect completion and get the right option and so is like what is like it's a really little thing but I think I will really help people because people are pretty familiar with such commands and of stuff in SQL well I think they don't use piece that most most people don't because the only way to know what the options are was like the look docks or maybe the source code infidels out so this is certainly help people they're trying to get up to speed on the but another 1
sitting tablespace options create time and so against was like this sort of active actually that there we probably should have this for a while used to be if you wanna do some like this so here making a table space and my location is magic brand this because I don't care about the data and so the option of passing it is I'm putting in a specific random page cost right answer I'm claiming that pomegranates I will get better than than the quality for random vs. sequential rights which is probably not true but I'm putting my data was on a random so clearly enough to concern about these things that you could do this in the past year after year create statement for the table space and then do an altar statement in order to set the different options so this is just sort of making that little bit simpler right you can do it all 1 command and set up so that you have a nice feature certainly makes it a little bit easier if you're dealing with programmatic building of databases to be able to do that all the 1 step know the tablespace
improvement has announced much easier to shuffle objects around not that we don't like writing scripts to do all this stuff but now having a simple command to do forces is nice but now you can move any object from want another and you can break it down different ways like if you want to all the tables all indexes are just everything in this table space over there process will take care that behind the scenes and here moving all my indexes onto my friend right out of the PG defaulting space you can rebuild the indexes right the primary corrupt anyway so why not the time so again this sort of makes it easier for those that are out there trying to manage and and shuffled things around like that a new view this in
the back end is PD that archiver and this view tries to show you what's going on with sort of the the wall archival mechanisms that are there so it's fairly straightforward it's scared that if if you know the more complex you start building out of you know replication set ups and then having archive files all over the place in multiple databases are within the sizes your post press instances of the more you need this type of information and a lot of this is actually fairly complicated to get at like you needed to have command access and and get access to certain utilities are really tracked in control log files to see what information was there so this makes it a lot easier to get that kind of information here no it's pretty archive count like the number of times that the archivist off from the last archived wall price it's giving you the actual file name of that but most of the furniture failed count if you break for some reason or for some reason the the wallet had communism actually working right that count will start to accumulate and so you can see like suddenly that that's going to make it really easy to to monitor that and we learn a tool that that we wrote a column on the data and likely will model like to try to monitor the stuff because it was complicated to monitor so now you can use a simple queries like a my field counters increasing like obviously something broke and I have to check that out so that that's the type of thing where it's like analysis of is like when you make it exposed the super level suddenly it's much easier to do all kinds of things as those of the tools that were used to working with them and also tell you the last 1 failed that kind of thing so yes that's certainly really nice if you're and monitoring which I hope you are into monitoring should be and then the noun it that much easier for you I guess this
is improvement but it's probably worth knowing at least so we removed rose 5 authentication having people out there used a rose 5 really really meant it OK you gasses are trouble so that have been deprecated since 8 3 which is like by my math like 11 versions ago or something and say 0 yeah you should use the GSS API instead and where is Magnus actually in here I would 2009 a supernumerary was I think he went to the other and then you may want to a query him and say however migrate and and make that happen so that 8 3 was a long time ago that I believe that 3 is not supported anymore is not so yeah that eventually deprecated things you your removed it turns out that have at a similar sort of in
sort of security and intellect and stuff which I think we of unsecured this point right now with the cylinder then and still clinging to the hope of those events and SSL TLS improvements new versions being supported numeraire always is going out and also changing so that the change in the fall for status has been done but from from what I can tell this should be pretty transparent for just about everybody in the world unless you're actually doing something specific and you ban in like a process company modify the sorts of things but it should be more secure but the general transparent all users it if you normally get with the default is the voltage better defaults but hopefully you won't see too many troubles but again membranous test and beta and beta testing you that hackable they can think
parameter showed up I think will will be of help them out of acting work members is I think the self explanatory but I will attempt to explain in many ways but also of vacuum work works like many of the other order vacuum settings that they suggest mimics the server settings and the default on this is is negative 1 like with the other ones and basic just means use whatever the regular work set to you but if you want to set your out that workmen something different you know actually have the option so pretty straightforward but but this was under like way we only have that feature and pretty sure that was there only 2 versions ago you acknowledge analogous them the session preload libraries and so if you're familiar with the preload libraries there's different options that you can use to like started back and when you start expressed on this now allows you to do it at the session level and colors all this is a foreshadowing of some things will talk about later on and while hints how they are you familiar heard about him that's so you this this painful smiles at the front of infants as without getting into outer died had been hint bits are little sort of pieces of helper information that prosperous has for doing queries against a database that has a passes around it in order for you to get a decent performance all phrase it that way normally associated not sent to the 2 rubble right they're not sent through the wall but in certain future replications areas of people working on again a little foreshadowing that lecture me something that you need and similarly useful so there's no option to turn that on it increases while along with the metadata is in a wall of which is why it's not on by default but but against and I have option to turn it on if you wanna use that of these 3 I think the 1 you most likely to to run into the art of acting work so just be aware that 1 and the other 1 you probably had a specific times but it's all good so I assess
performance administration prevents and then we get to my favorite part believe it or not is SQL but I guess you probably all people here like SQL right like you the movie like SQL is not dead yet the some of the like so if you like SQL and still use I think you'll be pleased with some of the features that are here but I was so
and so 1st thing I was a hard time making time for things right 1 has so many things during the day it's difficult to make time for all the things you want another new ways to make time inside of stress because that's the kind of helping data suppress response to be that what these are a series of functions that enable enable you to build sort of date and time interval strings so have example the bottom right so you select make interval and what's nice about these functions is that that she's named parameters so you pass in the parts that you need prices have make interval that 2 hours 1 minute and 120 seconds and what you can only see down here is 2 hours 1 minute and 120 seconds was way easier than like all the manipulation that we have to go through for this concept so there's 1 for animal is 1 for timestamp and ending and 4 times sometimes and certainly from a problematic aspect this will make your life much much easier if and when it comes to doing this stuff nice little will things much bigger
feature file is moving aggregates so this is sort of optimization for aggregates they're called within a moving window so I have beer people from window functions of the while I'm actually really surprises and and that's pretty often but if you're familiar with linear function you understand the idea of as you're working on a results the idea of the window that can move along with the results of right and look at like 2 rows header 2 rows behind that kind of thing out so there are certain aggregates that can be used within those windows right if you have a moving window this actually Bruce support for those areas within the moving window of sort of walking you through a very detailed example this would be extremely painful take a long time so I will say that if you're not reading defenses blog you should be because he has obviously very often modeled on how this works in goes into the details of this also that that that the sort of main thing is if you're doing this type of stuff using these areas of moving windows that optimization has been done in the grammar is I and wait for other are agrees to be done that way so and that's a really nice bonus nominalist there are things
like this amendment everything lead and lag those functions I believe are there if I yeah so this is not really adding those functions in its optimizing those in the moving window condition in del
examine average others obviously there so this is just optimizing
an ordered set areas is is sort of a new thing has been added and really new feature in that sense so it's sort of a new class of aggregates it you can use that and in the basics are ideas a term that you hear people talk about this already there I also in a group or within group is sort of a mental model right and then this'll make a little more so some examples of this this is based on the sequel 2008 standard so I guess it is more or less interest compliance which will have to comply with standards so scared of 1 the few examples of sort of show how
this works so let's say I have some data in here basically of doing right this is just some random data that have which are dollar amounts and the number of people who have done this is like a payment table right and so I've got data source dispersed among and what they want to figure out like well again here's my data and all these different amounts of couldn't figure out like what is the the median amount right and you actually can do this in sequel now I would say it's fairly painful and so this makes it easy to use so here during a
period on the payment stable all right so gravity coupled I them account the minimax many this new function percentile pond in passing like point 5 so 50 per cent right and then using keywords within group right from looking at within the group of data that I'm looking at ordered by the amount of prize as part of the way that the work if you order your data in some way could you tell me where I am a 50th percentile right for the query so then means is 399 so that is my median amount everyone quickly goes through Wikipedia Google a median average mode to remember this slide I did it when I was renaissance and I've probably confuse this 1 so otherwise if you want compute median and it's now very easy to do that but within you think about what is also sort of special in the it's about this that's see 1 final multiple pieces of information data so here I'm like can I get 50th percentile maybe 90th percentile of that data right can pass right into that function with the different amounts at you want and you will get array that with the different amounts as the calculated so that a lot of sort of the machinations behind the scenes in order to make this stuff work that is pretty interesting but but that's early logic and the statistics data much easier if you go back and look so few Members of 399 is where where we say the median is 6 and 9 is the 90th percentile which back
really quick so 6 ionized here so I have most of my this look like you some rough Matt really about 3 thousand tied up looking at like 12 14 15 thousand of all my rows so that the problem right so that's the quick math related to these as a direct anyone here probably had an outbreak of muscle and will check but I trust them so that was
much easier if you decide like well that's all well and good but
maybe idea of what the mode of the density of that as well so that was not the only function electric motor function so using that exactly the modem owners for 99 and the median will 399 so anybody wanna bravely say what does the mode mean nobody's ever erasure share most common we a lot of did a lot of people rather than so most common for 99 was correctly
data of 4 9 9 2 3 thousand 89 that they're actually has the highest count so it turns out it is the most common means it's you you know you should chasing appeal Lawrence but also to success of here like like modes in these narratives like you to look like all right again did
did we obsolete similar work that
made this event like you could figure this out somehow right equal and you can see that the things the if I can figure out what is the highest number in like water by the Countess anyone that 1 or something figure out the value you probably could gotten that value but
they only this way much easier much nicer and so I think that's the open up a lot of a lot of like the easier we can make it to do statistical analysis on our data like the the better for everybody so that a bunch of other
functions in this area as well so I would encourage you to go check out the docks and look at those right because rank is particularly interesting but you can pass in a value that does not exist within the set and say where would this rank within that set of that type of thing and I find to be kind interesting so again orders narrative thing really to open up some things and make it easier for people do more advanced equals properties about that stuff and the other 1 then and I was talking
to Richard last nite and he said well this is completely useless and sequel but in post-crisis totally not completely useless and I'm so excited I was really excited about this feature and it seems reasonable that is assume gold filter aggregates and so this is adding support for the filter plausibly people have ever heard of the filter calls for the work on other databases and so so what this solves is the problem of let's say it like here's my current trying account bullets also account where the amount for the far right so member half of my my if I go back to my data
right so if the if happened the people were 99 so I wanna say like well you may everybody in an account number of people who have spent more than like for 99 right so the
only have to do this in the city of protein and members of government so you would you probably so the counselor filter where the amount greater than 5 which actually looks like what you want you to have go done some of the operand case when an amount greater than 5 1 else 0 and clustering right and like it's not that doesn't work it is that it's really ugly an offensive and you know to use sensibilities so this is actually much nicer way to do that I think people can look at that and so even ask my friends especially like the more complex things get right the more people look at this case statements of there's like where why are we even doing something like this doesn't make sense and then when you like add Knowles into the Mexicans even more messy so there is a much easier on nicer way to do that and and certainly so actually has a little bit of portability for process as well as support cause is another feature I think I'm probably overly excited about the future but like will make it makes me feel better about this equal that after a set of such that the variance as well so the sets of
sequel stuff that that I'm really sorry about the sun about back and things but I've seen a
lot of POP just go code that many of you in this room written and I think is a really good feature for most of you and so now you can get stack traces from European OBD-SQL and I find pretty much everything about debugging PLP just you know it is difficult but really good and sorta like a caveman esoteric way but but beyond that it's it's doubly pretty tough so this logic stack traces and see what's going on with your POV just other
when refreshing materialized views concurrently this is a big thing if you remember 9 3 we got materialized views the big problem with that is if if you want a refresher even if was a lock on materialized view while you're refreshing at that basically meant it was useful for almost nobody but this now makes it useful for most people as you can refresh a behind the scenes at I think the thing to be aware of obvious is like still if it's a query will take a while for that refresh the happen but the good news is that concurrently so you don't have to hold a lot while you're doing it yes they get to use the
old version whilst building a new and then it will swap that almost like how you want to work so but it does a lot swap for the most infinitesimal amount of time so if you think about it like when you do a concurrent index build it's sort of the same mechanism as the
other 1 with check option for other updateable views a bit old he's really interesting and in the current incarnation because you can update things in a view that you can actually see I think that's interesting in sort of ways in which I don't use those features that's fine with me for the 1st who don't like that sort of side effect you now have the ability to make good use the user with check was and you can specify what you wanted to check is the where the amount of data actually visible in this view or is it in the visible in this view in any other apparent reason might be looking at this prices some more flexibility there as well so that's good that's not a complete sizes so use a phrase completes the abstraction of using tables and that now where you can throw areas if like the rows and actually in the view throwing here but to the so it's much more difficult to tell the urine of of a table but I'll
hipsters really excited about this allegation be functionality on how many people have not heard of jason than this function OK so everyone has heard of this is this was a place and a lot of time so that's also so basically suppose the movie j somebody from regular J. Sun said no need to replace the timing of there might be overselling there's less need to report data as your swapping it around on w proved indexing support by wide margins and get things like a quality checking and and and sort of better occurring within the Jason data structures that you have but I think there are some
constants white space munging I think is perhaps more problems than people think it is but if you use JASigning your current applications I would strongly encourage you to test the Jason stuff while still in beta from before they get the world and causes trouble of as it as there are some J. Saunderson be sort of quirkiness issues where like if you cast invalid J. onto adjacent be you'll get an error that it's not allergies and so like there's some quirkiness things like that but also found like it takes more disk space if you have small adjacent objects it'll probably take more displaced to use choose on the than it would otherwise as you get larger adjacent objects and more rows that trade are to go back the other way is actually a whole bunch of talks related to Jason being a different storage formats and working with that data tomorrow afternoon in more 256 I would strongly suggest checking that out of using Jason some pretty heavily you absolutely should get if you think about using it then you may want to can choose still worth taking a look
because Congress has also and did some benchmarking I decide throw that into the slide that those out of time series and so just if you want look at a quick comparison of like Jason residues on B and and he threw Monday because you have reminded you be so as you can see the table size of the data he was using was a little bit bigger which is on the smallest among the index size systematically smaller which is the thing and then the 1st index size like indexing the entire some documents and the 2nd 1 single key indexes if you just go index a specific piece of adjacent document there's a smaller as well and you'll note that all the numbers do you have be monitored the set but we don't quite your data is easily quantities that's not exactly and part of but speed also is dramatically faster using Jason regular old j song so again tomorrow afternoon check it out if you use and J. Sonner interested Jason the self and and start beta testing and to make sure I that really needs a good sort of ones over to to make sure that work for folks a couple of things
here I'm I'm recovery target immediate as is sort of a straightforward 1 on the recovery target when you're setting up 1st question unit you know to nodes like process replications of every target tells stress when it should want the 2nd instance and actually make it be a widget database so we had a bunch of options for that you could do it on the timer like a certain transaction ID or you know named boyer like when all the wall is normally played out in that finishes that was sort of the the normal 1 from the median allowed you to tell it as soon as you get a consistent state and so you know you have a working database watch that thing you know up into 2 this is a real database mode and certainly by making thing in some cases where like you know when you're setting up like reporting systems are like that of instances in that type of thing where you really don't care that you have every wall process just like this is I get something up and running please get up and running and that will make that an option but we're almost
out of of time so a couple of quick
things about future developments dynamic background workers and 3 added the ability to have background workers were traditionally processes that respond to express started and and they could do then background work and I 3 is only a start up 9 4 gives the ability to start there is dynamically either from like sequel or or sort of the machinations of and press press so that's certainly a nice thing in and that is laying some groundwork for some future use and a parallel career as the flea 1 take a look at that and user of another
1 dynamic shared memory which I have to say is not dynamic shared buffers as a 2 entirely different things these do not confuse them so but they never shared memory is not what it does is it
allows the server to create new shared memory segments on the fly so for example if you created a new background worker from equal and you wanted to have access to some shared memory at Conakry shared memory set segments and use that other losses a message passing infrastructures put into place so this allowed back ends that want to spawn other and to be able to talk to the back in the response so again it's sort of laying more infrastructure for that type of stuff
but extensible toast support so if you been waffling about the type of to supporting needed of was now gives you some options that category you have added infrastructure it so the infrastructure is out there you can support external or alternative compression schemes and I think everything else lives the right and notice discovery with the rate at which the amount of LC again by which I mean so it works for most people the interested objects with with sort a large database for that way on the other side of the Julian compression algorithms so this should later groundwork for people to be able to use other compression algorithms for data has early the type of data that you're trying to impress me make a difference as to what you want to use
and last but certainly not least replication improvements and there's another 1 where there's actually like a whole slew of changes that have gone and and and I think probably the way that I would sum this up if you heard a lot about replication changes in like a sink is coming in multinational itself I was here for the average user probably not a lot has changed in a world with 9 4 when it comes to replication there are certainly a lot more features and if you're the kind of person that likes to build replication systems you now have all the pieces to build the systems and but that the systems themselves and then the thing as sort of being easy to use like you know very post-residency dashed able molten mastery Synchronous replication distributed global hash table like that flags not there yet so so I'll stuff that you read about that I think it is worth reading about it understand what's going on behind the scenes to that those systems are definitely coming but as far as I'm forced concerned this is really about studying that the groundwork so that 3rd party tool authors people want to build their systems have the capabilities to build them but if you're into that type of thing and or if you are like this is the 1 feature really needs so I wanna study and that lead to model it for 15 hundred talk unchanged at extraction from that I think is probably 1 of the more complicated pieces and more user-visible pieces of this suite of changes that they put in all stuff were gonna taking a look at so that that the heads up on replication improvements in there's a Roger thinking that only fewer bugs in replication and 1 for on promising so so that is replication of improvement step which I got the talk into that and that's mostly that which is
good because runtime again Roger
tumor Twitter is I will post this slide someplace violators as question 1st of all I will not work and the maintenance work men were used to really I feel like they were before I feel like work before you know it just says that I have to believe General
Gerichteter Graph
Vorzeichen <Mathematik>
Fahne <Mathematik>
Statistische Analyse
Kompakter Raum
Skript <Programm>
Ordnung <Mathematik>
Automatische Indexierung
Natürliche Sprache
Dichte <Physik>
Ordnung <Mathematik>
Tabelle <Informatik>
Objekt <Kategorie>
Folge <Mathematik>
Mathematische Logik
Spezifisches Volumen
Demoszene <Programmierung>
Virtuelle Maschine
Reelle Zahl
Endogene Variable
Äußere Algebra eines Moduls
Tabelle <Informatik>
Anwendungsspezifischer Prozessor
Wiederherstellung <Informatik>
Wort <Informatik>
Boolesche Algebra
Prozess <Physik>
Befehl <Informatik>
Fortsetzung <Mathematik>
Computerunterstütztes Verfahren
Komplex <Algebra>
Regulärer Graph
Wurzel <Mathematik>
Lineares Funktional
Physikalischer Effekt
Speicher <Informatik>
Dreiecksfreier Graph
Konfiguration <Informatik>
Arithmetisches Mittel
Automatische Indexierung
Web Site
Total <Mathematik>
ROM <Informatik>
Kontextbezogenes System
Physikalische Theorie
Wiederherstellung <Informatik>
Physikalisches System
Puffer <Netzplantechnik>
Front-End <Software>
Operations Research
Speicher <Informatik>
Einfach zusammenhängender Raum
Physikalisches System
Objekt <Kategorie>
Dreiecksfreier Graph
Formale Grammatik
Streaming <Kommunikationstechnik>
Suite <Programmpaket>
Befehl <Informatik>
Kategorie <Mathematik>
Güte der Anpassung
Gebäude <Mathematik>
Kontextbezogenes System
Suite <Programmpaket>
Wurzel <Mathematik>
Rechter Winkel
Stabilitätstheorie <Logik>
Klasse <Mathematik>
Automatische Handlungsplanung
Abgeschlossene Menge
Dienst <Informatik>
Service provider
Inhalt <Mathematik>
Statistische Analyse
Automatische Handlungsplanung
Elektronische Publikation
Patch <Software>
Konfiguration <Informatik>
Gemeinsamer Speicher
Kartesische Koordinaten
Einheit <Mathematik>
Mixed Reality
Figurierte Zahl
Funktion <Mathematik>
Zentrische Streckung
Prozess <Informatik>
Globale Optimierung
Funktion <Mathematik>
Twitter <Softwareplattform>
Gewicht <Mathematik>
Keller <Informatik>
Ganze Zahl
Kerberos <Kryptologie>
Mobiles Endgerät
Diskretes System
Auflösbare Gruppe
Gemeinsamer Speicher
Migration <Informatik>
Keller <Informatik>
Formale Sprache
Zentrische Streckung
Binäre Relation


Formale Metadaten

Titel 9.4 On The Floor
Untertitel A Talk about Upcoming Features in Postgres 9-4
Serientitel PGCon 2014
Anzahl der Teile 31
Autor Treat, Robert
Mitwirkende Crunchy Data Solutions (Support)
Lizenz CC-Namensnennung 3.0 Unported:
Sie dürfen das Werk bzw. den Inhalt zu jedem legalen 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.
DOI 10.5446/19071
Herausgeber PGCon - PostgreSQL Conference for Users and Developers, Andrea Ross
Erscheinungsjahr 2014
Sprache Englisch
Produktionsort Ottawa, Canada

Inhaltliche Metadaten

Fachgebiet Informatik
Abstract How 9.4 will shake your world Join us for this first look at new and upcoming features in the 9.4 release. We'll discuss some of the what and why of the new features, and help you start planning your upgrades. The Postgres development should be mostly through feature freeze by the time PGCon rolls around, so this talk should be able to give you a good idea of what will be in the next release of Postgres. Improvements in JSON, replication, updateable views, materialized views, and more. Hope to see you there!

Ähnliche Filme