Bestand wählen
Merken

Postgres Performance for Humans

Zitierlink des Filmsegments
Embed Code

Automatisierte Medienanalyse

Beta
Erkannte Entitäten
Sprachtranskript
8 the the title and here the right place following really quickly show of hands present here that doesn't use but stressed the OK and a half of them don't right and so on a few this completing the sports 1st and so I do a good bit of writing on that was best among other things on my personal blog and hopefully the helpful resource and exterior something called post weekly and so this is a little different from the 1 it's the hostess rationalist on it's more end user content and how to use it's as what's going on and on the actors it's word curated induces on highlighted the integrity of resource and post status and other things on ground and then of the of the apple hopefully initialized much better have here then no homebrew or my words and then I it reviews are your review get and we run an essentially the largest fleet of but that's in the world the and so the only deal with all the administration the chromatic for you so I shall be sending in speaking a part problems and and what stress I usually put up a slide that like this online in the beginning and the end as some of the Y-scores press free of if many of you are in the romances talk they're not nearly enough hands go up for some of these things and extensions of particularly aged or arising key-value store directly into the database and McManus this conference yesterday was talking about the signature database has just like you know the story kinds of perfectly good 1 I will get into a little bit at a hell of this index is akin to work pretty well and common table expressions of our or some other window functions inducing data surprising right and downstairs form you should probably know about maybe everyone these features if using perspective just as it can make your life better if not look this I'm sitting in the standard units you general were close friend database and LTP in them FIL CP is essentially 1 that's so this is mostly where focus on the while on a lab is a perfectly valid place and you know needs performance considerations as well on search getting much more varied in and deeper much quicker and and so on until after world you're looking at the i and reporting on obviously generals for it but it is essentially a whole of that we can begin to I'm really quickly here people more on like the web applications I here for early doing more or less have a website effect and people doing data were having that things that the sorry I disappointed new units 1 the latter half of leaders of the new so useful on so really quickly on post that something can fail the 98 to much and this this is some nice basically on you know what to do and basically copying follow along and don't worry about it and so here on Amazon and shamelessly and user review or with the result of who post SQL inside your data and basically said about 50 configurations and then you go and then you don't have to worry about it on the other planets what talk on and if you're on real hardware so and you know good performance machines and crystal soccer media starting point on but to really optimise it high-performance Perspex feel this is a book by Greg Smith is pretty much the defacto on her tweaking invigoration of but it's based on 9 . 0 but it's still pretty relevant overall and there's a link in the all the sides of the environment as well laughter and they're a about things like there's a way to solve the the so generally when we're looking at a database and every worker like customers and say you know I'm having performance problems in it and the 1st thing to do is to take a look from you know about the 30 thousand foot view like that nice far away how healthy is the sum of all and I think that the number of things you to pay attention to here is is the catch on so suppose that is really good about keeping frequently accessed data in memory of urine 9 system before this you mentioned warm which is huge when you FIL over to way read slides that has the same factory your performance is basically in fact concept rewarming at free fell over every so often is really nice and post best is that you have added out of this thing then you are and you have to think about it it's pretty intelligence and keep your frequently accessed by data in cash and then you're looking at a thousand 1 performance difference if you didn't catch 1st having C to disk from what we've seen is generally did a poll is 80 20 rule so you don't need as much memory as you know the amount of data on a lot of your data it's very pretty frequently access found Mandela that this kind of rough but generally 80 per cent of your data should be in cash the on and so the cache hit rate was there's a nice simple query to get it and you can google you can find this and we have a job growth we have in of yet shows you can give you the source and that this query from there and it's going to give you something nice and simple that that looks like this at the end of it and then what you're looking for is somewhere around 99 per cent or higher but as soon as you start to get below that the see pretty significant performance degradation and queries and the other 1 is and they said right so how using indexes versus sequentially scanning time data on so this one's a little bit per year and and is something that looks like this and so on this 1 index is I would say below a little bit rule all the right here I would probably say on anything that's over this top 3 tables and their performing horribly I'm I'm doing a sequential scan over a few hundred thousand rows on and this is actually a real world example from after we had where 4 is particular 3 seconds patterns indexes on I got it down to you I think around 10 milliseconds onsets we typically at there and get into kind a ruffled over in the 2nd on so middle-aged room and I say rough guidelines but Kasuri you're going to be at 99 per cent or higher on Tony time looking at a database of and get it 99 % by just from more memory a minority of the and it's a great looking at 95 % higher me and at the this weird thing doesn't rose they're kind of in grade this is a hard and fast on doing a sequential scan across the you know a hundred roses to that and sometimes you know you got 10 rows in a table this with scans the faster than you that's and but it's not heard heavily to have an index on on every single row if you frequently using it as well we so i'm it some those queries a cascade indexes the but generally recommend just saving these and using them all the time but a nice thing to do but coming he customize the PC glossy here but full of himself is news on basically setting like agreement and I can distill select from that and so the bottom 1 right there I'm really nice to say the so if you're on your own from mobile machine in the things that cascade index hit of slow queries will to this 1 a little bit of but super Indians can save yourself so much time versus you know going for everything time what 1 really quick on this scale it is
an awesome tool for visualizing this on cascade index it and some other things you'll actually track visualizer rely farm look for things like blood contingent other stuff and still 11 again every repo in solitary pretty straightforward that so from there the obvious thing is on captures the decent of indexes a decent but some things are so slow and so understanding specific query performance the of coming of you understand where explain plan site and posters you and can read it and make sense of it there there was a lot really ensure hands there of flu from so pretty basic query on so for those of you not familiar every query has a core planet and its final show you what a because of the human might explain why show that you know stress on the actual what happened how long it took the step and so this is what 1 looks like essentially on amending showed this slide once before well I'll talk about in a 2nd but young vectors and I haven't analyze the data I so it'll still the cost there and the cost is what it roughly things and some of you and it's actually in some measurement that I don't understand I don't think it's an actual MS like it is I'm explaining relies on its some weird costing setting internal it as best we can generally judge it the same way on so supercomputing trying to simplify down on the initial cost right there is the start of time for all of the middle is the next science and on the right is that the rows returned and I believe in 9th you now have they been made 3 also of rows excluded as well sometimes on explaining analyzing when discards rows which is and very helpful as well and so this is nice and then marginally helpful of the real values is finalized on the difference here is it's actually to maximize the the query so this is where you don't want running on your system is that the CAM production in and well don't do this with it it's going to run it on but it's also going to show the actual time so you got the estimates of the result before I'm essentially the the actual time that so for each step so you can start to say I don't know certain things that right to 195 ms I'd say this is actually pretty bad for what it's doing it's returning 3 rows in total which is pretty poor on that actual time there's where you wanna pay attention to and rough rules on their and for most Web applications of you're leukemia page response and I would say you're looking for a year and 100 milliseconds so common queries under 10 ms where grace under 100 ms and it's entirely changes in the data warehousing world and by time small of result sets so this should be pretty much we're aiming for and it's a common queries even getting down very close to 1 millisecond pretty commonly the but we starts as a kind of micro optimizing read to be solving their problems on as they said that's a pretty bad time overall the and the obvious thing indexes so creating an index of concepts the source of the crime limousine certainly isn't even understand now and is in the ring time down to 1 . 7 ms so their living that's a couple orders magnitude shaved off of which I think is where you want to optimize Sony query I'm looking at you how many orders of magnitude prestigious roster and so man is also mention this in his talk in the Learning from a user makes us so there some people are aware and using it but it's still way too many I don't think you should you have process and solve you should be using this on a probably weekly basis so PD stats statements normalizes every single query is running into database so if I actually have something in there looks like stock meters were analysis is greater of the volume of the cell long but as you see right there it's in recorded a ton of information about it as the core the number times it's it's run the average time that it takes number of I O blocks 38 the written all of these things there are interesting and by the the most interesting thing is that you can with a very very simple query by getting inside of work I did the best cost-saving targets and so on hopefully that's fairly straightforward is basically a take the total time of that every query I'm linda that as minutes and take the average time on that it's a taste so certain tell our own optimize and how many times its run as well the so that's very did something like this I'm assuming the full queries out there so you can see that I've got 295 minutes on the first one thing my database and a 10 ms on average against each time it's run and looking at those 2 things I can pretty much tell where I can give myself a time that you and I are virtually optimize the first one energy stuff down to the 2nd 1 since the total is close enough that I can probably gives you orders of magnitude that the person has anyone in the room with a database like this and taking advantage of is that since this the people of this is super easy way to say no worries many times during the year that the all right so I'm indexes unstressed index is really lot and this quantity types on Speech Ginger's STDs and each here that is coming along as well but I don't know when are more at least limited by the crazy Russians that creates of all the index set and if you're like me you know on the personality this is true and this was kind of my response and but this is great and it was explained to me and I still don't quite follow when to use which and then they have entirely different implications of the year will it actually of
performance size on this and other stuff on many crosoft talked a little bit about the storage in ingest earlier on and performance on I'm a boil it down hopefully much simpler and this isn't perfect rule of thumb the trained usually what you what we say create index this is what you're gonna get on in most cases this is what you want to use and not worry about it the Jan so for those using a race or a store and if you're not using a store table without it's oscillatory saying earlier on it's essentially a key-value store directly inside a database and answer you can show that the dictionary right in there and just work on Sujin generally is when you have multiple columns multiple values inside a single column and so arranged to obviously makes some sense the gist of an here's where you have a essentially a couple again to the thing that can overlap in so a full text search shapes and rough rule of thumb working kind of fall under you multiple different ones the on in in uh is for similarity on STG I think I've had that explain to me 20 times and I still vaguely understand it and I here's that for phone number so when you guys kind of different densities within different subgroups and is the best I can do there and I'm sure that some people in the room and maybe can explain it better and I would love to hear from you afterwards but I here's differ from members so you're doing anything with the university track it but with a few more indexes right so there's the index types on there's also the ability to do conditional and functional on indexes which the fertile applications are hugely applicable of so conditional and so basically you know from selecting certain places maybe only 1 index of my my talk of cities are index places that have a population of over 10 thousand and so this is going to keep that index for only whereas conditions in kinds of creating a pretty straightforward and created on the table in the column and the condition there on you can actually have functions on this as well if so what functional indexes also includes awesome thing here using Jason interaction the the people I'm betting it's not performing well for you unless you're doing some functional indexes and moral issues shoving or all adjacent blobs of the JASigning press arise getting better and hopefully we have the same CMB and actually you know it's a lot more performance on the right now all J. Sommers from inside the advantage of a gene index just indexes because its text with the uh with functional index is always something like LDA so PLGA flows of is the VAD JavaScript engine embedded directly in the post breast on so you can execute whatever script you want of which is also introducing design so using it you can uh start to do things like this so that the basic ideas some blog down there on time the function called the numeric which is actually the reverse down into the design of the full else uh they see population from my column data which state is an awful awful awful Cullinane except I think it works for days on there so all sorts are inundated and there is that's so this in this case and actually combining in the fossil indexed the phenomenon the so here we go on the tree index on on on that function and it's going to do what you expect there from 1 probably important thing is declaring it is immutable from a mutable essentially says if I have the same thing in a minute as a result out of versus if you're who you're function has the ability to do random things to right there's a change data on which stocks and other cases for the you're going to uni suffered in essentially is designed to be able to be immutable and pose stresses that optimize that different and so if you got something that you have always returns the same results based on the same input and then he would be mutable you get a former states the the the the the and then of course you can combine and so we can do a conditional combined with functional indexes and the pretty straightforward of created index concurrently is anyone here not aware this I does anyone have friends use my sequel friends may be wrong but I would say this is the single the easiest way to convince them and so hopefully everything required here but it's going to be roughly 2 3 times slower of when I say there's a lot the table doesn't take out a box the entire time the table it's going to take a lot but it's essentially a noticeable on minister middle of your index in the background and you can keep writing DAC doing shit and you have to worry about the bringing down production for her per data to white tried in the performance which certain other databases like you can have high and was a big thing is pulling on the path to everyone years 1 distinctive and so on this is a new relative performance chart and right there at the bottom is the connection time from cycle 2 and and this is pretty normal so some people of common that seems really high on the bulk of the costs there is SSL negotiation on which does have some overhead but this was the devolver time and Django on that it got a connection every request the and I'd say there's a couple of things here to fix it up and there's the application framework under which often has connection pooling so I think rails as by DeVault and the seek well in Ruby library j going eyes persisting connections which is 1 states
that interplay framework has persistent connections and I was looking here your language framework it doesn't have it of figure out how to get in there mission is the default under the other is a standalone I think this is what most people think of when they think of flexible and I'm in reality both can work to be useful and then larger applications will have on the general opinion I'd say you don't need the stand-alone demon in cylinder maybe a hundred connections are more at of very anything in there and it was with me come so under the pulse Christine inside all there's 2 options that pretty much at the bouncer people of had pretty pretty much say answer is we should be using and I'm sure people may disagree with that by it's it's kind of 0 is 1 tool for the job which is nice for people there's a lot of different things in in terms of your caveats means element our doing the the on so as I mentioned early on cash is important it's probably the most important piece and it was the when no building LTP applications on the easiest way to when performance is crappy is just add more cash for more hardware the problem on 1 kind of nice the option of doing that is a replication of several replication is that a little bit better pose rests on but the nice thing when I say you know replication cash you can send reader queries to a slave that you can handle 10 ms on seconds of latency from the time it's written to it's red which most applications can and that other the re slave is going to have its own Kashin optimize for that which can be very very nice and so it comes replication there's quite a few options there's a Sony 1 but seem to say that the type of of the cargo of PG full and Wally and bar man I have personal recommendation and if you're not already using replication and wanting to get started on home I would start with 1 of those 2 the the simpler Orion and again is that really small lightweight tool to you have to worry about too many things and you'll know when you need the others generally but if you're deciding if replication they generally just work on them and they work at the Post Festinger station on the it's nice tool around it make it will be years set up the on so an updating post settings on i encourage most people on if years if you're not control is looking at us as can file on my encourage you not to look because it's scary so we need in each of these settings there's there's simply just a few I would really generally start by people and I'm I'm not really going to see tells his anything poster this starts on actually explain them pretty well overall on but essentially you know optimizing your your cache size for your database might allocated on shared buffers work man and maintenance work Nando apparently now it's going to be on the other vacuum I forget what it is now the new 1 in 94 and that's essentially for acting on but workmen's unity in a few doing a little more or less of your research between optimizing their little bit more so on the back of that's another area and that actually has performance implications on most people just add them automatically to the database on which is a good thing to do not saying to reduce your notes and there's a couple types a back singer does have some implications so the 1st 1 I think is this way most people really do our is what's called a lot about that so this area of the governor database on it can be human readable doesn't have to be but can be human readable on its portable so you can take it restores hopefully on cross-platform just works this is what most people do the other option is is physical so this is the actual fights on this on the generally is says the VAD armed kind of architecture of the opinion that you're on the same 1 I've heard rumors that it can work across ones by don't think anyone in the community commits that works on so basically you're going to have to have the same infrastructure their resources and so it's not horrible of setting up you don't let with by identical of this is the actual bytes and this and in the base factor by itself is is often useless on you that the red longest well and actually advantages of so kind of comparing contrasting it's you and logical it's going across multiple architectures from portability and it has those on the database it has fairly significant load on anything cool thumb is merely a I say it were true under 50 gigabytes of if you're over 50 invites is going to start going on and not perfectly consistently fire and it's just 1 get worse over time on the job is hereby range of might be shocked to hear if you're doing the dump it actually works and so search is failing pardon but the really nice thing about like the back of the arm is because it is human readable on portable it's actually going to do some validation there so it can be a way to find corruption so I would encourage people to do it but do it on a once a week once a month the basis of sex some now 9 3 definitely help on checking for corruption and but if you're Arnold older database this is probably the best way to actually find in isolated and then no physical lines and more initial set up this utterly some more work on early on on then just the down on but is less portable at but it does have a move the system so I think it's you know the scale and this is really really want the and the other thing is visible that helps our typically would enable the the replication so on and it's a good thing to do on-site encourage you not necessarily 1 or the other you should be looking at but what combination of both you need so for us we actually do the base factor of Staley I'm an ideological back it's time to users but I encourage afterwards at the type the to actually blew through all that but 103 capital in the direct questions and the people
were all out the below the topped pulls there and I'm sure they can lead to a pretty important to you but the order and this is helpful on it but it's an interesting for the natural the the riori comes with a lot of caveats freaking essentially room in your database if you're wrong but it is really awesome I think were moving toward a little bit of that functionality to appear in court over time and there's an interesting things there but I don't think the full factionalized any activities and the other thing is you start to get more more hardcore in the the solutions and on top of a stress on the eastern the primary products and or it can be you know resolutions with things like the proxy for a little CP on the the basics of and data is in cash as much as possible so 99 % and I think for us any time the season Ruffalo that we encourage them to upgrade have to optimize so you correlative P-Stat statements on I was solely started the steps and I would look at 1 of queries and look and see how much over although they have 1 on the system and less for some reason there's 1 report you really really care about and efficient use of indexes and and and and and scaling of the database the easiest way to scale is it for more money the problem that actually does work and databases need memory and work pretty well with the the and gets all the slides to be online speaker that focusing and then all of them questions we the here are the 3 I don't recommend using the setting sun but it's got some overhead but it's pretty low overall and I think the the overhead that introduces you easily get fact as you optimize and we we have this we started turning it on for every single customer might be part of it is so valuable and inevitably you're going to say hi optimize the database and having that they already there it's it's hugely valuable the yeah and something this with the the the level of you know I mean it will die out of and so the question in case anyone didn't hear it was on fair lateral more PD down there try not being reliable over 50 gigabytes and at a certain point it'll just fail and it won't give you a dog and the crew won't really clearly say why you and PD don't doesn't have the best error reporting and so sometimes it works sometimes it won't on and typically it's when you get to that size is usually the 1 large-scale from problems funny that 1 table it's really really huge and for some reason it just doesn't work well I think there's some inferences being made to the and it also fails when you have thousands and thousands of people and so some of the management they're getting that information the catalogs and work and so essentially you nothing and you're in no it's not working but for the solution isn't eternity inching along more and more and more its its sister to go the other way the that's you can do about it and I don't think it will work because was the question it in In case in here was a is very important 1 which file system should be used for a purpose that I'm not alone yeah so file system layout and I think it really depends on that depends on a lot of things of of what you're doing I say look at crystals talk he gives us pretty basic version of if you're on cloud drives to this here on these sheet here Cloud Drive see this if you're on real argue this but essentially what you're doing here you just setting the sum the page cost settings of how long does it typically takes its rampage on and what you read that for later on and PostgreSQL pretty much you're writing with the finger on and terms of you know is the preferred file run on and I think it's probably a huge fight shared really know that there's a ton of different opinions on trying to think of war run on its escaping and the the the yeah I think you're going to find people in all camps on from from I don't have a strong opinion personally there but most of check with me after knighting ineligible and analogous more information like the the if you you know you yeah so for automated backups we use the g will depend so for all of our customers and we do use Bayes factors every 24 hours and and our cluster results see that we just do it we needed we worry about it they'll have to think about it and because people like to feel and titleholder back up on we have something called the PG backup for her OK and that is a pd don't want and Minnesota's people do wonderfully chemically on I'm not discouraging the use of PG don't just know that it has learned a system under you should be doing other things and know that it will start scaling for you at some point and having a lot of you out on a daily back up and instead of replication as well were replication should lend itself to replace some of that and by running the backups of our the dumped on a daily basis the introducing unnecessarily unnecessary was and then we have the requests every so often come in Peyton I have run this back up every hour it's like why you you don't Miyanaga recover that way and there's a better approach the so yeah it's their users use it it's helpful for marine data around on but it's not hard default back also but yes we thank you and me the kind of you know it's on top of the search for the 1st of the of the of the of the of the the of the and that the and and you know and and and and the the thank you for the the the the the people and the release of the the of the of the of the of the original read and it was all about the world and I think it was in this see in the in the starting of the some of the some of the some of the things so so as yet so there really and that's where the the and the thickness of the new guys very near and that was the yeah I don't know if this is the time series of
the when happening culminating in the work in and the thing about yeah I think in the in the in the world and what you can do this kind of thing to think about how they don't work and in the case of the in in the direction of the wind which and it's not the only because it was still maintaining the induction was and that's what I mean by that is that the the only thing I know all about what I'm actually a lot of people around the rest of my career in the police force on that and so on later on most of the time you see a lot of the problems because there's been a key part of the variable operating on and some were better than others but I think they're all improving the answer trying to actually all of them and say we have time and number 2 on induction I'm thinking on the lot of things this is the notion of good is that the here we have a higher you know model and it's something I don't want you to be changed and this rate now there's the tune there's something in that was used in the name of the point here x like you so you look at your and maybe with private branch within push something like that yeah there was there was no others went on it that is making that and it never there is there is there is 1 of the things that kind people of thinking to most of the people they thought they had to do with just the moment when the rules governing the point which would have you have the right to the rest of the world so a more remote monitoring agent because I just want to be on the we all a motor environment for the existence or anything like that just send it on the number of that you think you know and and and I'm only going to be the thing in the morning and it has an init during the time because it it seems that the real is in the whole in memory no reported that their member had run it in the solution of the parameters and so you know the the way started actually like all and from what I read what I'm going to the door in the face of the amount of distance to the moon to Sunday's go there and you can reuse the memory of the event and is the author of the of these simple and the it before this 1 and now you know the the the the resulting in and the and the number of although the adenine and in this mind because of a lot of people and that's the part of the data that is what we call the speed and the memory management in the model that the line of in management and and that usually where all of the I don't know if anything like that and and there you have the option value if you're just of the running of it I will remind you that 99 which was right about the time that it was a big at all on the same or more elegant on the on the right the right side of the and here is the idea of looking into the union and now and here I am supposed you just going with some of the ideas for a general election on you know it's an oblique 1 or 2 people that have done that and that's basically the of and but I don't remember the effectiveness of health and so on and so it has to work on and this right the this and so it the the of the and the end of the world and it is you for the the the and good for you and and and will be the the so you and the this of the the goal is to the the and the reason and look at the end of the week the back and book the the the the the the right and so what you need to know the history of the history and you know that sort of thing you and the 2nd to the original if you want me and if the work with all the and you we if you really think the and in we we the and there a in a the this the the in the so the group we were in and out
and the group and the whole world that used to be the and you can this is 1 of the things that you would like to go the this in the with the and and this is the kind of and the the the the the the the the scale of the so that we can work in and the question of is what the reward and the people and screwed time the good the the the foreground and trust you we should be in the the the and the in the in the the the so this the the In the but but but this is the very the but the good but but but but but but but but but I the the the the the the and I if that the the the what we used to have the and some of my parents were students would be and should be the thing about and the the the and the the the the the the it the the the the the of the the end the and the and in the the and in the bottom of the page and and the you can the and the the you and so on and and so the the kind FIL the the the book the the people is the of we these this theory the the and and and of the the so we we was the the the the the and all the in the the the and and and the the would the In the end the the and the the the and and the the and the the
the if you to the the and
the government and we we we we could to
use the water that that and the and
if you and we found that is we note that
the and this and that and you and
and and the and and and this that's the and and and on the right the kind of the the
Resultante
Bit
Gewichtete Summe
Web log
Web-Applikation
Gradient
Arithmetischer Ausdruck
Einheit <Mathematik>
Prozess <Informatik>
Mustersprache
Minimum
Zentrische Streckung
Chromatisches Polynom
Hardware
Sichtenkonzept
Datenhaltung
Güte der Anpassung
Abfrage
Bitrate
Elektronische Unterschrift
Rechenschieber
Automatische Indexierung
Rechter Winkel
Festspeicher
Normalspannung
Programmierumgebung
Portscanner
Tabelle <Informatik>
Subtraktion
Web Site
Zahlenbereich
Abgeschlossene Menge
Virtuelle Maschine
Datensatz
Reelle Zahl
Perspektive
Mini-Disc
Vererbungshierarchie
Inhalt <Mathematik>
Maßerweiterung
Speicher <Informatik>
Konfigurationsraum
Hilfesystem
Soundverarbeitung
Videospiel
Zwei
Systemverwaltung
Schlussregel
Physikalisches System
Binder <Informatik>
Fokalpunkt
Integral
Fensterfunktion
Caching
Mereologie
Hypermedia
Faktor <Algebra>
Wort <Informatik>
Resultante
CAM
Bit
Minimierung
Web-Applikation
Kartesische Koordinaten
Fortsetzung <Mathematik>
Eins
Untergruppe
Homepage
Netzwerktopologie
Fehlertoleranz
Perfekte Gruppe
Reverse Engineering
Minimum
Visualisierung
Meter
Skript <Programm>
Einflussgröße
Metropolitan area network
Umwandlungsenthalpie
Lineares Funktional
Befehl <Informatik>
Siedepunkt
Datenhaltung
Abfrage
Ähnlichkeitsgeometrie
Quellcode
p-Block
Ein-Ausgabe
Biprodukt
Dichte <Physik>
Motion Capturing
Rechenschieber
Menge
Automatische Indexierung
Rechter Winkel
Konditionszahl
Information
Overhead <Kommunikationstechnik>
Ordnung <Mathematik>
Normalspannung
Aggregatzustand
Tabelle <Informatik>
Lesen <Datenverarbeitung>
Subtraktion
Web Site
Total <Mathematik>
Thumbnail
Quader
Data-Warehouse-Konzept
Mathematisierung
Automatische Handlungsplanung
Interaktives Fernsehen
Zellularer Automat
Zahlenbereich
Sprachsynthese
Framework <Informatik>
Kontingenztafel
Multiplikation
Weg <Topologie>
Datensatz
Unterring
Mittelwert
Datentyp
Endogene Variable
Programmbibliothek
Spezifisches Volumen
Speicher <Informatik>
Grundraum
Analysis
Einfach zusammenhängender Raum
Schätzwert
Transinformation
Relativitätstheorie
Schlussregel
Physikalisches System
Vektorraum
Quick-Sort
Energiedichte
Dreiecksfreier Graph
Basisvektor
Speicherabzug
Größenordnung
Resultante
Bit
Gewichtete Summe
Punkt
Inferenz <Künstliche Intelligenz>
Minimierung
Formale Sprache
Versionsverwaltung
Kartesische Koordinaten
Element <Mathematik>
Datensicherung
Homepage
Übergang
Eins
Datenmanagement
Prozess <Informatik>
Datenreplikation
Dateiverwaltung
Default
Figurierte Zahl
Gerade
Bildauflösung
Metropolitan area network
Lineares Funktional
Zentrische Streckung
Befehl <Informatik>
Hardware
Datenhaltung
Gebäude <Mathematik>
Abfrage
Ausnahmebehandlung
Teilbarkeit
Konfiguration <Informatik>
Rechenschieber
Softwarewartung
Menge
Automatische Indexierung
Hochvakuum
Festspeicher
Ablöseblase
Information
Overhead <Kommunikationstechnik>
Ordnung <Mathematik>
Normalspannung
Lesen <Datenverarbeitung>
Tabelle <Informatik>
Kreiszylinder
Sichtbarkeitsverfahren
Proxy Server
Subtraktion
Thumbnail
Hyperbelverfahren
Schaltnetz
Online-Katalog
Term
Framework <Informatik>
Puffer <Netzplantechnik>
Multiplikation
Spannweite <Stochastik>
Zeitreihenanalyse
Datentyp
Arbeitsplatzcomputer
Mobiles Endgerät
Hilfesystem
Einfach zusammenhängender Raum
Zwei
Validität
Physikalisches System
Elektronische Publikation
Portabilität
Flächeninhalt
Last
Caching
Mereologie
Basisvektor
Gamecontroller
Speicherabzug
Computerarchitektur
Dämon <Informatik>
Streuungsdiagramm
Verkehrsinformation
Punkt
Momentenproblem
Statistische Schlussweise
t-Test
Gruppenkeim
Zahlenbereich
Physikalische Theorie
Richtung
Informationsmodellierung
TUNIS <Programm>
Datenmanagement
RPC
Existenzsatz
Minimum
Vererbungshierarchie
Abstand
Gerade
Autorisierung
Soundverarbeitung
Zentrische Streckung
Parametersystem
Verzweigendes Programm
Schlussregel
Bitrate
Ereignishorizont
Quick-Sort
Konfiguration <Informatik>
Forcing
Rechter Winkel
Festspeicher
Mereologie
Speicherverwaltung
Programmierumgebung
Offene Menge
Open Source
Rahmenproblem
Wasserdampftafel
Videospiel
HIP <Kommunikationsprotokoll>
Hacker
Computeranimation
Offene Menge
Open Source
Softwareentwickler
Rechter Winkel
Raum-Zeit
Rechenschieber
Sichtenkonzept
Bildschirmfenster
Hacker
Computeranimation
Zeitrichtung

Metadaten

Formale Metadaten

Titel Postgres Performance for Humans
Serientitel FOSDEM 2014
Autor Kerstiens, Craig
Lizenz CC-Namensnennung 2.0 Belgien:
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 10.5446/32609
Herausgeber FOSDEM VZW
Erscheinungsjahr 2014
Sprache Englisch

Inhaltliche Metadaten

Fachgebiet Informatik
Abstract To many developers the database is a black box. You expect to be able to put data into your database, have it to stay there, and get it out when you query it... hopefully in a performant manner. When its not performant enough the two options are usually add some indices or throw some hardware at it. We'll walk through a bit of a clearer guide of how you can understand how database is doing from a 30,000 foot perspective as well as analyze specific problematic queries and how to tune them. In particular we'll cover: * Postgres Caching * Postgres Indexing * Explain Plans * Extensions * More

Ähnliche Filme

Loading...
Feedback