Bestand wählen

When PostgreSQL Can't, You Can

Zitierlink des Filmsegments
Embed Code

Automatisierte Medienanalyse

Erkannte Entitäten
manage fits and talking about some projects have been working on with the press lately and indeed for on the TI where tech-support for some i traffic with its applications based will provide support from hardware level with application development and mostly for companies that have very high volume of traffic and have very large amounts of data on the examples of companies that work with the may may have heard of our ideal at the work to be relied on to other companies have every year we also brought conference called surge but is most the thing that really the this is that supported how things go horribly wrong and how you recover and fix things from that and very large environments but anyway in September and you check on the test site the exact date right now and of course like many other companies here we are also hiring on here are tree is the CEO he did the previous talk about online for features and image itself is another people person here so looking for a job you speak any of us this is the 1st product I worked on called the extracted from EEG dump is pretty limited and when you can actually filter out as for the training at the objects of the database those are pretty much the voltage you have their table schema and the P is for functions in only works on the restorer actually does not work on the don't so I end it the it only dumps out the on access control lists of privileges for tables and comments upper table so this pre limiting in way if you just like when I get a functional areas you try to get a specific pieces of stuff that actually challenging so I originally wrote this world and recently in a version 2 . 0 and every written by an expression 3 and you can run the 3 to to scrape and it does work perfectly fine and by the time to set by the major thing is any considered all the things that are that there that can filter by she puts each object into his own individual file not in 1 giant SQL files and organize them into folders and and files like that you can also here a regular expression matching so if you need to be done about partition tables in you know the pattern in the tables you can it a pattern like that you can also give it up like a list of files and another like Alyssa tables in a file given that file as a filter the only ones filter out specific approach to and had some other features and actually you can actually extract out the all the default list and like the for the full privileges and stuff like that now I figure can is going
to a demo this is explained that shows a lot easier than me talking about it so you have this 1 is made of a thing you don't have to you in the database name if you don't do it database they know it don't everything out in the far end of the older you specify at the top level or the default folder you're running a from so I usually always give a database named so to put it into a database for over and when he gets everything and you want to keep the dump file in this so that he made the key for database so there's no aggregate functions roles schema tables use all on our borders China tables they eat every table in its own file this is all in all outputs and I'm plaintext the plaintext format if you want a binary formatting give the data the dash option just like you can it has most of the same options as on the roles of although overall that has actually you have an option to filter out whether you can construct pasrage from the role wild so you know your by checking the sender version troll you can filter out the password hashes so does include them that well truck
is a sickly uses the only thing that the extracted uses speedy dumping bigger restore on when you originally wrote 1 the original bot program this was 1 actually like would log a database newsstand like rebuilding system catalogs those 4 8 3 9 8 4 1 9 0 came out system catalogs change broke the whole program so to avoid having to rewrite this end and keep the future compatible it only uses the down in a combination Pt down and the fiji restored objectless it passes through that to get everything out and if you have overloaded functions they will put all the versions of that function in the same file it puts the act of access control list for the functions in the file as well so tries to keep everything altogether from the filtering of the light years in IBM again all but you can region 1 able scheme of plates roles there's a get the full privileges from being actually get sequences out if you give it that it did not by default an output data want it day that you to get dashed edge get they'd optional get the data as well and lecture give you the sequence values as well to get data out of the get rules accounting if things with similar options and they the dump as well like in a schema Kaplan exclude came in the tables well you can any connection with a file name which each each other goes on use functions you can filter by owner or the fact that a lot of people found they only 1 dump object on by specific user you can do that it does have terror into things right things in parallel with the help over after the has reduction people right so it
also is a actually a full Python class so you can actually import this and use the public functions to get the order like this structured list of the PD don't the pedia down object list out and use it however also you feel you may need to use it so that it may useful in other situations as well any questions about this book I have links at the end with a where I can get on on the TI has their own labs and get repository and from I get repository at once told what happened at the end in questions all of my other tools will be talking about from that from here on out are also mention some extensions system was introduced in 9 1 that basically describes what extensions are logically grouping things together and and and or in a way that's easier to control the country modules that are posters are actually extensions now as a 9 1 so the old version 1 . 0 as far as I can tell that hasn't been updated but you can have actually birth control of groups of other objects in various now allows more controlled updates of your of your code and a license to be more of a word if you have the same code multiple databases like these probably would you can make sure they're all running same code and lot a predictable updates and if if the author of extension right you can actually downgrade as well in a predictable way so lot single allows control of your decoding database mostly functions and you can put tables and extensions but the extension offers didn't really intend tables for anything other than configuration data such as when you do a PG don't you don't actually about any of the extension code it all out creates is a create extension whatever and you don't file so if you need that data you can set flag on your own extension table to include to be included in the dumps but then that also include a table in schema down with the data so if you have a very very large tables as part of your extension you'll be dumping all that data as part of your game so hoping they make extensions to move more flexible but so far it's it's work really you can write you can write extension that many languages like with cost function C Perl Python but you need to do this is the 1st extension i worked on it is that it's autonomous functions the feature that's missing from post breast with that basically means you can't do multiple begins a commits in a single function so you can't like loop over things and and keep things going if you if and that means everything that runs in a function that that function fails everything in the world back so if you try to do some like have monitoring on function you analogs steps of function and see where fails and if the function value at rollback all everything a lot so you have no idea where they'll it failed so this extension was written I was actually code we actually had written back in 8 2 and a 3 but 1 9 1 came out I have organizer the liberals code sync across the clients so this allows you to log in little steps of function by using the link trick that DB-Link is another extension all the dealing does is allow you connect to other posters database and do whatever you want so used to be linked to connect back to the same database and you basically make an individual sessions that our individual transactions that can be committed so it does have a little bit overhead to functions but I'm usually the functions or running your head is michael length negligible and the monitoring the function is much more important than the overhead it onto it and so cell on but also allows and the integrate monitoring monitoring tools they can keep track of when something felt little this is basically have work and you and a job that all the edge of function you name you step the functions what that step will do and then you update the step what have been out there of this section makes it easy to keep track if you have a really really long running functioning wanna see how they're progressing along units call updates inner loop and then go query that the job 1 table and whatever your writing to that job it'll keep track of things how many rows it's going 1 of these 1 and others that you just go in and others that I'll keep going and also accounts for the failure people can put a failed of call and the captions and then you then you get whatever the error logs were there was of this basic what what the upper looks like and its own there's there's 2 tables as a job log table a job detail table I mean these functions users convenient for that you can create a double tables and a directly and this this is like these individual jobs here and then this is like if you want to get into the individual steps of a job so that was done by the 9 year you the individual steps of the job and what it is that I use this 1 a lot because I can never remember exactly what the job they are so I made a convenience like function the end at the end by default it only it'll give you back the most recent and if you need to get more than that packet of these put a number in there given the most recent 2050 whatever and also provides model with this 1 this 1 1 function even with those old it's a return record so can create however you need to create new user scenarios and we also only the I also have a longer graphs there is culture currency that we use and how to do this I have all along that but these 9 years before you should look familiar is what basic not use requires you to look like to the built in all of the tools of people already made so this looks like and everything working OK this one's stuff on that I'm so state so you have a you have a function that runs on what what runs it must run every day so you can actually there's some configuration options and here you can tell it then um and you tell you give the job name and you tell this this this function must run every 24 hours it hasn't run the last 24 hours when you run checked job status that will come back and say like this what this says some of missing and when the last ran so not just like you to what critical means it's just like you know what that means when it gets back a response like that so you make the command service checks and ideas and then would tell you of this function run and by the by the fall I made this set of flavor ever of function never fails 3 times that by default sets off a critical long so much of Hill's idiot something's runs runs every 5 minutes and you want to you in a like occasionally fail once and recover the effects itself so this'll like a lot you with its failed 3 times in a row you can change that however you want also has an escalation so that you can see is that actually the warning so something goes in warning 3 times the last played critical like another failed so any other questions about this is that
of the Internet and working on the new it's a logical logical or pour table replication extension some of the mean for us this winter and the source dark I'm looking for copy synonyms and that definition came up from the thing the government on it from mimeograph was low cost in small quantity this is very very simplified logical replication solution compared to things like Picardo which are will 1 that much more powerful than this replication people the need something that's it that's you're going directly in a lot of tables need need to be more easily manage those would probably be better off going for something they just replicate a few tables or in December specialize replication and this can do this may work better for you the big thing also is a so both into those presses most you know is spinning about shipping but that's pretty much all or nothing you can't take up except if you go to the next talk about how the tuning logic replications were built-in tonight for some and hopefully this may make this extension obsolete but you know there's 3 basic types of replication of read about individually on the biggest thing is this this requires the out of insulation required to produce around his grant the permissions to the users running it and people to work this is the sole or from the destination database so this is a whole thing is they pull replication it's not push to set this up on your destination where you wanted to go and it pulls the data from however many sources you find invite as many sources as you want all points in face and specimen has some other features you can tell the only 1 of replicate certain columns and the countries are aware condition only at rows as well and it also uses the PG job 1 extension provide monitoring and logging of what the snapshot in his way copies of table every time so this let's I go over a small table or a table is not doesn't change very often in the the last option there if the source data hasn't changed just does nothing so what it does doesn't specialized way to minimize locking actually to view on top of the tables and replication runs it populates the other table and swaps the for a brief locked so while replications running it doesn't actually have your whole table locks it's just doing all that the background and swapped for every 2nd this is much faster if here if the majority of the table changes at all of the time this is much much faster than replaying the ML on replication system so you can this report will hold the whole table select star rather truncated repopulate table I'm actually checks in the system logs signature look and then he says the juice that stay all tables something like that I can be right now there's there's actually our columns and that that have an incremental count of the number of inserts updates and deletes done under the table so I just store that number and watch for teaching and sensors redoing the whole table actually made it so it'll of column changes so if you had a column struck columns working plates it'll it'll get lot of you from 1 of our clients had an issue with this slope is they're actually using this 1 of the development databases can view is not the same thing as a real table has sequences all that kind of stuff so I made this basic cable propagation option locks the whole table every time a do replication but it actually will let you know at the sequence numbers and foreign keys knowledge and stuff working and it'll mandatory resetting the sequences and foreign keys on the stuff so it's good for development database that each 1 copy data from production to to to the development database that's arbitrary to you but could be could be a gigabyte could be 100 k kind of up to you and this is the other specialized that replication calling him and all this is based on there being a column on the source that changes with every single answer update and that's the key part of it has to change with every single answer update and updated to a new value so this is actually very ideal for things like like a webpage tracking table this is constantly getting hammered with with new data 1 replicate that over putting most of the logic replicational report relies on triggers which popular another Q table so that makes your high traffic tables having erected 2 tables every time when you wanna replicate them this allows you to some avoid that so that both snapshot people and that's only require read access on source database requires nothing else so very very non-invasive 1 replication you run into issues it's it with serially on 1 and 2 communication with timestamp but if you you run into the daylight savings problem the easiest way is to not use the word is to use GMT or UTC time for your database find that will eliminate all the problems but not everybody can do that in the client we developed this for our runs the database and Eastern time so the easiest way to avoid that is to just not run replication during that 2 hour period and that that's the only way to really account for so it's different places have different DST time so that's configurable when you want to not run so but that's how all yes that is that that can that can be a problem and that can be the candidate in there that that is 1 of the other issues with the with stable replication of yeah found by the bullets are for time but if all the 10 minutes so it is your destination will essentially always be 10 minutes behind so he said that he said that boundary variable to when you know the sources than doing what it's doing I give it takes 15 minutes draw your answer use the boundary of 15 minutes will account for that not happening the same thing with serial
usually with serial it's saying it's a counseling committee members of the video but if you're instance a number or a long transaction variant and he had the same issue but you can set the same kind of boundaries boundaries is actually another very big problem is you think this would be an easy replication type is not there is a lot of edge cases with this boundaries being 1 of the big issues with that with handling things when when replication runs you grab a value but that value is still being inserted on the source and you're using that boundary is your next as your next group of data you witness a completely so there are some configuration options to light and all that but time I also added some and some recent functions in this recently for for this but this 1 and this despite that will moderate the source columns change it will let you know they can't replicate them automatically but really sensor monitoring to let you know the source changes you can go and this is the type of biological evolution of almost everybody does I call it the amount of base measures replace updates with the trigger an acute able to keep track of everything from like is it doesn't actually replace everything they replace all incident elites but it's a you have a hundred thousand updates in between that time period it will only require the last 1 because just going by the primary key values and this this option does require primary areas in the that you don't have that the or snapshot that replication methods can work for you on additional so little automatic that the trigger and the Q table they put on the source automatically manages giving a direct grants and stuff that it needs come and you can have multiple destinations this does mean putting multiple triggers in multiple Q tables on your source so for every destination you're ready to that many tables on your source so I have a hard limit of 100 but during that many tables here I think you are hearing and other issues so also did this other option we have data warehousing uses that and you don't know what you don't want to do an audit every and every update or anything but you want audit when Rosalita you wanna keep that role round this will light the the deleted rows on your destination this adds an extra column on the destination of 1 row is deleted and keep rest on some users with us where we really do it so you have a on a table in your source database and you want to keep track of the art at all but the entire order log on another off-site data warehouse you don't keep the entire audit log on your production database on that work really well with the incremental replication so you can you partitioning on your source table and drop all the old keep although are all cables on brother of played in the National so users to do upgrades across major variance I wouldn't upgrade from 8 1 2 9 2 with this and basically took the way you do is you get you take the bubble the largest tables which is usually not the majority of the tables is usually in the top 20 or 50 so this does required is that each individual to individual table of replication but setting up 20 30 tables is that they can deal sees at stables up was replication from the old to the new 1 and that could see the majority of the data across the entire database and then used to a dump on the smaller data on smaller tables and scenic significantly reduces the downtime the upgrade look at they gained 5 6 hours is now down to maybe 5 minutes because he got the majority of the data already replicating and action somebody else that uses for they had a had a sharded was system that uses they use a UUID so they had 512 shots shot of outputs database cross 512 databases but they needed to reporting on it so they set of 512 replication jobs both 512 tables to a single database and then that reporting so use cases I would never have thought that somebody else that uses it with itself In this you know just just think of yes that this this this snapshot 1 can handle getting the getting the the new column changes over the other 1 the other replication methods cannot so it will automatically do that I have some monitoring functions to monitor for assessing strange on with the game actually tricky you have to do that you have to the counting a specific water yet Avenue new column to pursue other company founded destination and you add a new calendar trigger and you have a new column to the actual source table because you add out of water then you start getting errors on replication so it is a tricky issue I might I document I have in the documentation how to do how to handle column counting with the ML so I that that that's the way it is I just don't hand I don't I can handle it and this and acceptance snapshot sense him on any other questions about this extension living in my biggest project in the about the most on it was found was press has no automated partitioning built into it and has a very very extensive documentation on how to right partitioning and he's stable inheritance in the core documentation that's pretty much what I base this extension on so if you look at if you look at the source code of it you'll see this looks exactly like the way the core developers and 10 partitioning to work so hard but this takes care of but if you wanna write partitioning yourself you have to write handle keeping all the table that the thinking Europe given function up-to-date using I like basic alphabetical partitioning that's cattle once and then set up and you're done but things retirement and serial IT partitioning you need have ongoing maintenance so this is what this 1 is designed to do it only handles
all the time period the partitioning they started out with including some basic pre made from time time series of yearly quarterly monthly our only recently actually got because sometimes the connections that any time interval you want and add a little bit more overhead to it so that all given the what static and then figures are all later for partitioning but on it'll it that you can basically partition by 3 and a quarter hours if you want and what you do that then the thing with partitioning is some if you wait to create the partition when the partition is needed like like as glycogen serial and right when you need a new table is the way to do it till then you run under race conditions and this avoids that the race conditions by pre creating the partitions on by default face forehead you can that that whatever you want to date and I also have it managing all most of the properties of the the of the child tables with their indexes constraints all religions ownership somebody that had already set on the parent table and had a good patch recently later actually account Freudian being inherited wealth and you set all that on the parent table and it takes care of a phenomenon of labels on just a few days ago actually got foreign key inheritance working so you have on on the partition set to 0 another table they will inherit from the parent to the child and foreign keys in the other direction and PostgreSQL like to a partitions that do not work and actually there is a way to do it and the reason it doesn't work is because there's a lot of it doesn't work because when you do a foreign key on a petition it only looks at the parent table does look at any of the child tables and the hair and inherent 3 so you can actually make a trigger to do that for you but if you have a really really large provisions that every single answer to that worth it will take a very very long time so this takes care of some of the figures as needed that was a very that object in mind is the very 1 of those things you don't usually think about when you're sitting a partitioning especially in serial partitioning that can be a big problem with that the partition you should like you like underscore p and then like a date or number for the partitioning Priscilla continues to get bigger and bigger and bigger and eventually run into that 63 character limit and you you're you're partition suffixes Heiga identify with the data is in there so expressed truncated it doesn't give you doesn't like stop you from doing it allows you to create your table adjust truncates at all so if you had a very long cable name that was partition by day in a kind of operating over here is lost in the whole semantic naming of higher partitioning works so this handles that by just truncating the actually move table and then adding the suffix on it private around them and also in recent versions some constraining exclusion is 1 of the reasons people do partitioning so if if you're doing a select and you only want the most you you have 10 years of data but you only when the recent months and you do a query with the where condition on the partitioning column will only look at the tables with the most recent month 1 instead of the previous 1 to the dual recognition 1 of the other columns not include when you don't include the control column is a sequential scan or index scans across the entire partitions which is that is the downfall so what I did was I use this pre create number the falls for that actually that actually controls the trigger of where where it's on what tables handling the data so afford handles for tables and and for the future and for tables in the past and will continue moving that so so what I did was on if you give it other column names any tables over the nose for it will look at the current values that are in the state that are in the table and add a constraint on the table based on those values so you can actually have exclusive trained on the other columns the partitions that only works if the dad at static if you will that is changing obviously this will work but most the time where people need that the real data static so does that make sense and where any questions about that I don't I actually some users talking to me yesterday about that at the Royal Oak at at this month's worth of partitioning it automatically I mean if you if you can somehow I I don't it would be kind of hard to manage what you could manually can manage doing so partitioning with this but it doesn't have it automatically on much pregnant that Appleton have in my notes to to see if I can have I it's basically functions that are create by 1st create the parent table is dependable already exists and then there's a function call create create parent and you give the parent table the the column and and whatever other things you and in each so you can even do that with the new brand new table on existing cable human existence enabling partition data on this will do the part of the partitioning for you so that the Python script do it automatically it a command after each partition created by default you can actually make it do in smaller batches like for example I have now and this is actually going a full up the head stable which have had state this is the hit like a web hits table that we partition alive while traffic was still coming in from the only lock this however requires the brief to 3 cycle locum apparent that that's actually wall this this is a very very busy busy table see only reason it's a true 3 seconds to get lock on the table usually it's pretty and then and then of these yet we partition by day but there's options to the partition that the partition function to give the user commits and smaller blocks so there's less contention so this our blocks of a 5 2nd to not overwhelm the wall stream and think about but about 10 hours but there was no interruption to production and partitioning was done on my actually have that's actually opposite direction I was thinking to do but I was actually I'm actually working on a consolidation script so like say you have your like partitioning hourly or daily like after a year you wanna just change to get yearly partitioning working on a script to change to do that I will be back in the the opposite direction as well and see why that might actually a little tricky but others actually a get to that a little bit the connection to partitioning and this it is still under a completely but it doesn't do various things on this is getting into what the different strains static partitioning and then partitioning was 1 thing the duration of some people they actually have like the functions writing functions or functions running SQL they just have a right 1 long line and that is unreadable is this is actually alive function Figure function this right so I tried to make them readable and but which basically what style means is you actually explicitly main each individual partition for each condition and the if condition on this allows it to the cash query plan makes makes a lot faster so in in this instance and I have to actually also have like alternating this's with the pre make life at the 2 so you see like this is the current table this is 1 in the past when the future to in the past to the future so can alternates to try to keep the if conditions of the the most recent values so it doesn't have to go down through the entire list but this is the most efficient version of partitioning is what people mostly due for for uh the provisions of the issue with this is that if you want to keep being able to answer to a lot of tables like hundreds and hundreds of them then you have the list of if conditions as hundreds and hundreds of lines long and impacts the performance of the year of the partitioning so if you run into that that's 1
student under going to dynamic can attend that part inventory now but basically what it's doing is it's doing next each statement and partitioning there's some variable so at the time of insertion it determines what the partitioning with traditional goes in interest that the problem with this is excuse statements are the plans are possible so this has to re-evaluate the part that the crew plan every single time the insert done so depending on your partitioning method that's why included both of these options because of depending on your data model if you need 1 or the other so I can write with both of these the the custom time partitioning on options actually uses this and also there's a look up table so that actually is performance of the partitioning methods but a look up tables basically the only way to do that sometime partitioning that I've been able to figure out so it's it's still pretty fast but obviously that you start running into really high and insertion writing 1 of those kind of issues using with some for time based partitioning the new the creation of new partitions is based on running across job with the with the of the function of this includes by default serial publication will go by when the when the current table reaches 50 per cent of its but whatever its merits constraint is it will go on to making the next partition the future on some people have reported for a for a very very very high traffic cables because a lot of contention because 1 1 of the book 1 insert will come along and say I need to make a new table and the next answer comes from long right away before the new table you created a hostile way so that that method is is 1 of the issues so actually made it so you can I did it that way because I Base partitions using a predictable of how often you have to run the the cron job to keep it date but I made optional now so you can actually if you know how often you need to run it you write that often and it doesn't do the 50 % creation anymore so it avoids the contention there any questions about static versus dynamic partitioning automated
creation and also on destruction of is the other big reason people do some partitioning is only a very very large tables and you have to delete 100 million to undermine rose as a very expensive delete operation and massively bloater table and only the clean up is either back in full which will like a table or using something like repacked but on this allows you you just drop the table which is very fast option and this allows you to automate that so you just set the time incentive to buy anything older than 3 months automatically drop the table by default just an inherited the pressure drop it so it and make it safe to that you want to drop the tables you have to tell you what up there and you can actually have only dropped the table or just drop the index if you want to keep the table and say the space so they you only keep it around the 1st and possibly the 1st serial the more complex products when the cycle and you give it a boundary value and although there was a current max minus that value is all ruffled tables of so many actually at another conference this other reason I got about the circumferences because I get ideas for stuff like this is actually needed to dump out the tables they don't even with the data in the database they wanted compressed archived outside of the database but the way he didn't use that he LSH language on which I started to try to do but realize it was very very limiting you would essentially require a set the shell of shells being able to be available to whatever platform knowledge makes it pretty much useless so window of so I I actually it that will actually moves the table of the schema and then there's a script that I will dump about the tables from that schema and lonely drop the table at the PDM command comes back successfully so safely handling it said you have on on partitioning and do so what they want to like use your saying to and automatically change of career with a monthly or something like that you can undo your current purchasing and redo the partitioning again so the out of the and portion will actually partitioning works with the the the the uh the table herons feature of our stress so that the end you will actually work for any inheritance they not just once the partition that PG Partner managers so you have another table inheritance see table you have set up a new 1 that this lecture this to watch you work with any of this tool on since they
Wheeler's in this talk but this tool has saved my sanity developing these extensions of the tab is a is unit testing proposed breast and basically you can tell at I want I run this query I want this result back there a stable exist as primary key exist as this triggers this all of those conditions human check if you run something this allows you to the unit testing for and this has been the only way have been able to release things even someone bug free on manual has about 300 and PD tap tests the partition manager has close to 4 thousand the try to manage all of the different i partition types and for time all like can I'd never be able to make sure all that stuff works without something like this so use this Co-funded delivery can talk to you about it but I love full of I would say I would make this a this is expected essential if you're doing any extension development at all have been made it may be useful you're your own functions but these are looking into extensions at you released to the public I would highly recommend using this but use of went through
all my tools on user I knew 0 see not all these funct always our extension there in the open you already so there will be easier to fit into the code of how they work I knew 0 see before so working on the Ernest over much yeah well it had the energy function so both of the database so I don't know what I'm doing in that sense but I still we will write extremely extremely useful tools for our proposed rests on and think the community of Oakland Post person the announcement was conferences our what kind of the all of the stuff to exist in the popular for people know about they give few new questions about love was resident have partitioning of the now now you can't there's no point having built in this table inheritance but that's not that's not partitioning that's it I mean you can have as like that no 1 in the other side the
Gewichtete Summe
Schreiben <Datenverarbeitung>
Ungerichteter Graph
Service provider
Fahne <Mathematik>
Skript <Programm>
Automatische Indexierung
Lemma <Logik>
Einheit <Mathematik>
Ordnung <Mathematik>
Tabelle <Informatik>
Lesen <Datenverarbeitung>
Objekt <Kategorie>
Folge <Mathematik>
Selbst organisierendes System
Fächer <Mathematik>
Mathematische Logik
Interrupt <Informatik>
Weg <Topologie>
Endogene Variable
Äußere Algebra eines Moduls
Spezifisches Volumen
Tabelle <Informatik>
Konvexe Hülle
Binder <Informatik>
Wort <Informatik>
Familie <Mathematik>
Serielle Schnittstelle
Disjunktion <Logik>
Vorzeichen <Mathematik>
Elektronische Unterschrift
Konfiguration <Informatik>
Arithmetisches Mittel
Automatische Indexierung
Autonomic Computing
Projektive Ebene
Web Site
Inverser Limes
Strom <Mathematik>
Bildgebendes Verfahren
Einfach zusammenhängender Raum
Physikalisches System
Objekt <Kategorie>
Regulärer Ausdruck
Dreiecksfreier Graph
Innerer Punkt
Demo <Programm>
Nabel <Mathematik>
Streaming <Kommunikationstechnik>
TUNIS <Programm>
Trennschärfe <Statistik>
Funktionale Programmierung
Befehl <Informatik>
Kategorie <Mathematik>
Web log
Dienst <Informatik>
Rechter Winkel
Cloud Computing
SCI <Informatik>
Stabilitätstheorie <Logik>
Zurücksetzung <Transaktion>
Klasse <Mathematik>
Automatische Handlungsplanung
Open Source
Globale Optimierung
Skript <Programm>
Inhalt <Mathematik>
Open Source
Elektronische Publikation
Patch <Software>
Konfiguration <Informatik>
Formale Sprache
Kartesische Koordinaten
Arithmetischer Ausdruck
Einheit <Mathematik>
Prozess <Informatik>
Figurierte Zahl
Funktion <Mathematik>
Inklusion <Mathematik>
Nichtlinearer Operator
Ideal <Mathematik>
Eingebettetes System
Funktion <Mathematik>
Technische Optik
Overhead <Kommunikationstechnik>
Diskretes System
Formale Sprache
Serielle Schnittstelle
Funktionale Programmiersprache


Formale Metadaten

Titel When PostgreSQL Can't, You Can
Serientitel PGCon 2014
Anzahl der Teile 31
Autor Fiske, Keith
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 10.5446/19085
Herausgeber PGCon - PostgreSQL Conference for Users and Developers, Andrea Ross
Erscheinungsjahr 2014
Sprache Englisch
Produktionsort Ottawa, Canada

Inhaltliche Metadaten

Fachgebiet Informatik
Abstract After using PostgreSQL for a while, you realize that there are missing features that would make it significantly easier to use in large production environments. Thankfully, it's extremely easy to make add-ons to enable some of those features right now without even knowing a single line of C! After using PostgreSQL for a while, you realize that there are missing features that would make it significantly easier to use in large production environments. Thankfully, it's extremely easy to make add-ons to enable some of those features right now. And you don't even have to know a single line of C code! Over the past view years I've been developing tools that have made PostgreSQL administration for our clients much easier. Table partitioning is one of the best methods for providing query performance improvements on large tables in PostgreSQL. While the documentation gives instructions on how this can be set up, it's still a very manual process. For time/serial based partitioning, the maintenance of creating new partitions is huge part of that process. PG Partition Manager is an a extension that aims to provide easy setup & maintenance for the most common partitioning types. It also provides advanced features for taking advantage of constraint exclusion and retention to drop unneeded tables. Another area where we ran into limitations was when one of our clients needed an easy method to keep the database schema checked into version control. The pg extractor tool grew from this and has become a popular method of finely tuning the extraction of database objects from PostgreSQL into individual files. Autonomous transactions is another one of the more popular features in other RDMS systems that is missing from PostgreSQL. Begin able to reliably log a function's run status requires that function failure not roll back the entire transaction and erase entries made to the log table. PG Job Monitor takes advantage the dblink contrib module to give the behavior of an autonomous transaction, providing logging & monitoring for mission critical functions. PostgreSQL's built in replication is all or nothing, but sometimes you just need to replicate specific pieces of your database. There are several 3rd-party systems capable of accomplishing this, but their setup can been daunting to most new users and overkill if you just need to grab a few tables. Mimeo is a logical replication tool that requires minimal permissions and setup and provides several specialized means of replicating from one or many databases to a single destination. This talk will discuss these project and hopefully bring insight on how easy it is to contribute to the PostgreSQL community.

Ähnliche Filme