Bestand wählen
Merken

Understanding PostgreSQL timelines

Zitierlink des Filmsegments
Embed Code

Automatisierte Medienanalyse

Beta
Erkannte Entitäten
Sprachtranskript
my name is taken in the areas of work for the emperor on his talk about timelines imposed so the highlights
of this or it's back to release 8 . 1 when and where he was quiz got point-in-time recovery on the on the table what point in time recovery 1st we understand where this is coming from and point-in-time recovery is a feature you use when you do something stupid in your database like don't drop a table you didn't can you or you forget where cost interval instead of so what you're going to do in this situation is that once you realize you've done something stupid you're going to kill the server and you restore from the back up and you try to restore the back up to this point just before the and then looked at usually happens is that you have interleave restored to far and then you restore from the back again and you try to find the right location where there are happen so I think it kind
lines is that they help to help you manage the situation where you restart recovery many times on every time you do when you restore from a back up and you find lines on a single
server timelines of the warring basically looks like this just a single line in time and and there isn't much to of the
however there is I'm using the same notation all lights and so kind flows from left to right and these things happen during times like dangerous and objects and so forth on interest will create a record of transactions of all its operations and so the little the features called point-in-time recovery what it's really all about it's about restoring to a certain point in the transaction log so time really defined by transactional on you say you want to restore the certain point in time we need to a certain point in transaction so we
funding time recovery the final and get more interesting so what you have here is the base back up so we should use this small overhead our and this is where you do the delete of the clause and then time and so on until you realize that he did something stupid and at that point you can look back in time so what you need to do is you go back to your back up here and you restored to the private law until you reach this point and then and then you start up the database and you realize that groups thing too far already went past the so that created timeline number 2 so now you go back to the back up again you restore and now you hit the right point just before that stupid delayed statement and resources such as the create do what you want to do and that becomes of trees it you the so all this
time lines have an ID number which you can use to differentiate and or the part on which kind of an important because otherwise if you restore from back of recovery store from the back again there's going to be multiple well I minus for you have the same positions in the transactional but they're not actually the same thing here is you recovered and knowledge you have different data that had that point in time so the time myself you to differentiate between transactional positions on 1 time and the same position on another another time 1 1 interesting
point is that whenever you those the timelines branch but if you are if you have a running server at a certain point in time like here if you fall backwards you can always trace their history back to the beginning of time at the beginning the beginning of the wild and is always linear from the point of view of a single server so single-server doesn't really ever see these different timelines this season its own timeline which goes back all the way in history of what's which might switch to different time the way that they it is clear a culture history goes back to the beginning of on but with the same
questions just showed this is kind of a collection of random thoughts related of runtime so if there's any questions just ask on so this is what the timelines look like in your while archive or in Europe directory the posters writes the transaction what's in the 16 megabyte segments which are named like this so their numbers increased but if you have if you have different timelines in archive of the bonds the 1st number here and I think at this level together so
what you can see here is that this is the 1st time in line with the ones in here and the transaction log numbers increased and this is the point where the forked all the 2nd time line by the 2 and you have the same number still if we didn't have online ideas we couldn't tell these apart which would lead into all kinds of problems like you couldn't use the same while archived archive mode of the time was it had to keep them in separate directories or something of the banks timeline ideas can tell them apart and you can just collect them all in your similar quiet and the system will keep them separate the so at the end of
recovery 1st of all if you have a master server that's running on the future you recover from a crash you do and free you replace all the lots on the you read stand and that's the of recovery on however but if you're recovering if you're doing point-in-time recovery things work differently than the trade is timelines and so the recovery gains when you stole reclaimed the wild and you start accepting connections that can write over to new transactions of right to say that a hot standby will never and recovery until you will take a lot of standby mode so at the end of the timeline system chooses a new timeline idea and in this case we people with was used in the online ID by scanning the wall archive and checking will timeline ideas are already in use so if you have time lines 1 and 2 and entry there already it'll see that OK 1 is used to to use use trees used social think the next 1 that's well and the next time along with the common for so after to decide it's what and the idea of the new time is going to be it's going to write the timeline history file and the timeline history file helps the system to trace back to go backwards
time on history file contains a line for each of these points where the fork to a new timeline so for example if we had the 4th time line of force of here the timeline history file and tell you that OK this is review you switch the timeline for this worries reached the time 1 2 and so forth but I think I have more of that OK so here's
an example of what it looks like
in the wall too many and recovered you can see here it's starts part of recovery starts reading the lord's until it hits the end of which in this case it's it's because it doesn't so find the next file more on so it's gonna tell you that OK on and on and this is the new timeline ideology of after that it's it's also known after
after an hour after the end recovery it's going to call it the last transaction log files under the new name because so what happens in the computer while while it was still writing timeline it's going to stop recovery somewhere in the middle of the files typically and you can't we don't want to have a causal files were you have like 1st solve this entity and continues in the 2nd half sold it it's got the partial contents of the previous work of from the people online under the same name in each of the changes that timeline IV and this part is identical in both files but stopping after the start of it's going to write a new transactions of the rest of of the file so I mentioned timeline
history for these are little files that will appear in the TXM directory and in the wall are and so I said earlier this is the final details what the history of this timeline is well all the points coding arrive in this point in time it looks like it's on 4
so every time the systems which Estonian timeline is also called the gold history file for the previous time and 1 so you can see that this is the 1st time aligned it forked from and he considers me his father's name of the trade off the streets means this is the timeline history file for timeline number trees and what means is that the previous line was to where this force from and you have here the location in the transaction log where that happens of by the way this is new is new in my country in 1 . 2 is used to be just a final name but in 1 country is actually contains the exact location of the timelines which on this was it's just nice to have more accuracy there but the the reason I needed that was in my country for that some changes the instrument replication of on and you may know you have a big suffocation transaction log where this which happened and there is also the reason why it happened In this case it shows 4 point called before pinch con which means that I have a specified I think the recovered look con file at specified that I want to recover the certain name restore point which was called before the chicken and the reason we system the timeline is that we found that point and system which acts as request yeah all
of this is the head and so they have their so the question is how does it know which histories with time is already in use on used to restore command the same this this what this history files are archived together with the actual transactional for us Our so they get archived with command just like the the conventional files or in the country using string replication there will actually be replicated just like the transactional falls on so would basically doesn't just by calling the restore command and polling checking which numbers already in use to its if it's continent timeline to it's going to check if timeline tree exists if there's a history file for commentary if it finds it then it checks the next 1 is the history history file for timeline for and so forth and you find 1 that's not very on so they could be a race condition here if you have multiple servers and you do this at the same time into service but that would be kind of weird using all of all this states are yeah that's correct in the year old yeah that's sure so this assumes that everyone is writing the same while
so 1 consistent point here is to remember that the 1st line here for the 1st column here is the timeline both the parents timeline so as you can see it is this 3 file for number 3 and there is no line for number 3 because we are still on the timeline we never switched off that online but we do have records for number 2 and number 1 and that's also the reason why there is no timeline history file for the 1st time like online 1 because that is what you automatically get when you just you need you never really switched in that they can be cast parents so other reasons you might see here is if you set of if you can recover from of conflict he said the recovery target to a certain point in time if you have time there then this will say what the constant or if you have an exciting area to tell you on what was reason we do this which so 1
and from random thought here is that this time on history files we have a bit of an unfortunate naming scheme here so it but of all the previous slide was Timeline history files but these are not to be confused with back history files which are completely different thing but you create backup history files and you do backups and timeline history files when you restore OK so the system just to show you what the backup history file looks like a here's 1 to a new duty to start back up we create the file that looks like this it's name like that is the current look segment and followed by a location about story thing and or maybe it's worth start and back on this file this file just for information purposes post has never actually reads this spot on its there just so that you know you can look at it yourself manually and see when you need a back and hold for you need to keep the to still be able to restore the file and so this is called the backup history file it's different from a time when certified this is kind of
confusing continues there's also something called a bag of label file which is again different from back history files and different from time history files and what makes this confusing is that it looks kind of similar to the backup history file on the previous slide but this file is important if you do a back up it's important that you include these files in your back up and this is Audio commended but I've seen people in several instances of people actually want purpose either leaving this fall out from the back ups thinking that it's just you know since the label you don't really need that or how is going to refer to actually exclude them from the back which is wrong we have concluded but we improve the don't this lecture will make this more explicit and the naming naming is a in inconvenience here just to
summarize the tree files with the bag label file which you shouldn't touch you can read it it's human readable but it's also by the system and it's very important for giving a consistent system and this is back in history file which is only 4 of your information you can delete them if you want and then there's the time on history files which are created when you create a new timeline and these are also if you ever want to recover you know true those timelines which is you can old history file you really don't need them anymore but the going like a 100 bytes of something so you might as well just keep them all straight questions at this point you know for sure that there are also has also been so that the only thing people yeah this this actually change so you might want 1 or at some point when there with his 9 change this so that we no longer read about history of in order to reduce to read it in earlier versions of but we don't do that anymore so nowadays it's just for human consumption this 1 of the things that the so yeah if you have if if you've tried to restore from formant-based back up onto to service at the same time that both pointing to the same while about what could happen is that they both there's a race condition where they might both see that OK timeline number 2 is used timeline mandatory is used the next 1 that's free is number tree and then they both at the same time alignment street and then you get an error when you try to put it tries to actually write the archive of 4 of our we don't have any fiction against them well yeah usually wouldn't have that but you might if you for some reason it's perfectly OK to point 2 services the symbolic value of the borderline of on a different time line if they are actually both copies of the same system just running on different timelines I'm not sure exactly what you would do that but maybe maybe you have a database production database and you want to fork that temporarily for some reason we try out something and so on were all they were all yeah it's not really stated if you it's a very small 1 you want run into that in practice very often I think on but is it's possible so the race condition only happens when you start up the database and its switches to a new timeline so as long as you make sure you don't want to recover both of them at the same time the dimension is is also what you arrived that's through the new timeline is determined when it's actually going to switch that happened after recovery yeah I guess we could try to do better and actually try to write and archived the history file and then back out of our own choosing new 1 if it fails so some of the we haven't bothered the so so 1 thing I
here sometimes is that well so this doesn't really apply to me because I'm not planning to do point-in-time recovery right point by recovers on you know what's the big deal with these timelines just confusing but as just like to point out that you are going to get new timelines whenever you do have a lower on a hot standby system and there's a good reason for that is some people have tried to actually pack around that way you know shutting down the database and manually removing Ricardo from so that it doesn't create a new timeline but I wouldn't recommend that because typical if you do failover on you might have some of you might have already done some transactions on the master server before you shut down you probably want avoid that but it happens so it's actually very useful to have
a go at the time of OK so here's a plan that so if you if you have a master server and domestic crashes you from all the you have a standby server and you prove that it will create a new timeline so you might think it looks something like this but in
reality it's actually more likely to impact recover it happens because in a typical scenario you fail over the mass atrocities and you have a hot standby system but it might not be totally up to date the mastermind already have applied some transactions or you might not even be what user trans actions that might be a vacuum that run in the background that created some some of the records in the transaction log so when you do when from although standby server it creates a new timeline and the right thing to do here and this is what the system will do is timeline because it it might not continue from where the master level and you might have lost updates here or maybe not if you're looking or if you you it's this process well and it's a control shut you can avoid that body but in the general case on the new timeline is the right thing to do so
1 thing I hear people saying that is that I really have a problem because I Synchronous replication so I won't have any lost updates because there will be in single well it's not true unfortunately and in saying this replication on what synchronous replacing DOS is delay sending back the acknowledgment to accommodate so if you do a comic if you want to return to the user application on until it has been onto the Committee record has been replicated the but that's all it just on it still retains the disk on the master 1st so it's entirely possible that when you do an update and then try and the client doesn't commit it never returned and it was which almost all day I think to that OK on to speed up a lotta against have on which 1 is it that I don't think so as to be doing something good so the the a the 3rd and so forth on the have avoided doing any back to the future references in if anyone has any this would be a good time to promote that doesn't work now no now hello all I can just try to speak up can you hear me in the back OK I'll just continue with of this 1 the 1 that's actually recordings working so where was so if you even if you use synchronous replication you still have the same problems that there might be some transactions that are in the master but are not yet replicated the standby what synchronous replication would do it will let it will reduce that window but it's still always there in general you can and and it doesn't solve the problem at all and even if you somehow managed to stop any trend transactions running on the master of this still post this still doing all kinds of stuff in the background of this checkpoints all the vacuum analyze all kinds of stuff running and they will write a new transaction log even if you don't do anything the database itself OK so you do use
point-in-time recovery even if you might not realize it again or you might not want to that you do on so it's important to know what these timelines are and just embrace them and learn to use them and it'll keep it'll keep you sane when you have to deal with these situations and in my country this a
few enhancements this and so this is something I spend a lot of time in the mother country release cycle much more than the half so I spend a lot of time on this and I thought it was just that would be a small few days maybe a week to do this but it turned out to be in months and months so far to hold open and fixing bugs and but it's there so before my country in previous versions if you wanted to do if you wanted to have a standby server and a master server set up in way actually if you have a master around to standby service and you want to them to be set up so that when you do a failover the 1 standby server other standby will automatically connected the new stand by and it will follow all the time on speech and keep following standby and the new timeline of OK the should use it so so so the former country you have to have a wall quote for that you couldn't do it with just of extreme replication and in my country you can have a look at the 3 replication will argue is that when it hits when it's connected to a standby cascading standby server and the upstream standby servers switches a new timeline it'll tell their downstream standby OK I mean you know it's the point where I switch to a new timeline of stop streaming and it'll stop streaming and this was a real asked them to master the new master from what is your new timeline it will fix all the time on history files from the upstream server this is all new that I have to write because there wasn't anything previously in the recognition for the call to actually get the history for our social now stream that history files and intellectually fits all the history files but not only the 1 you're recovering and this is for the reason that was asking how does the server know what what's the next time line that's not still not being used on so extreme replication it will fit all the time on history files even deflect future ones that are not interesting for the server itself because if you then later fail from that server needs the history for us just to tell you which ideas are still on use of and so this is kind of complicated but it now works we can do you can have to standby servers and failover between them without having while archive before 9 countries you could still do this but you have to have the wall archive and restore comments about yes the all that you know the the yes exactly the same way you yes that's exactly right so that a not to be a bit difficult but it's minds to do it and then also also I did the same thing for kids you receive actual which is actually a very nice tool if you haven't used it is rejected as a tool that ships with posters that basically also used to store the actual files on a disk like a warlock higher but if you it doesn't require it doesn't use to restore command mechanism to do that instead it will connect to the server using extreme replication and it'll just streamed into a file until the end result is the same as a wall archive you have excellent us there but it's created by streaming so the PGA receive can also do the same thing that's the standby series will fall all the time I switch and if if it's connected to a server that gets promoted or or follows another timeline it also affects all the history files and right and the disk and then decide what to do next and fall for this so cruel so there's 1 more thing I'm going to cover and
this is something I just wrote a while back up so 1 problem we still have if you have a master and standby is that if you if you fail over from master to a standby server and then you wanna failed back on you have just have to be restart the old masters server from that and this is this is something that new users who are not familiar with this will be a horrified when they hear is like before I have a can survive database there's no way I can just recover from a base back up time on so
the weight becomes the standard way to do this is that you restore from the base back up and he said recovery target timeline to latest firmly and then you roll forward all the lost from the base back up on so this is just the same as the graph
of a faster way to do this is to use art of which is nice so if you have a master and a standby server failover and then each other fairly back what you need to do is to do the gist of back up on the new master server and then you are a sink instead of producing a new base by but use are assumed to sink optimize the directory and the other day the directory and the old masters and the coastal back up and then you can recover it like you would recover this and this is much faster than recovering full based by that because our single only have to send the difference but the problem is the field that if you have a 10 terrabytes database it still has to actually read through that can terrabytes on both systems so that it can see it's changed so yeah it's the
same with the graph so you want to see
my post reasonable hackers I wrote a little tool on cold PG rewind which does the same thing as are think it determines what has changed but it actually does is wearing transaction and I think this was pretty cool but I originally started working on this when Andreas point up did some work on the 9 when 3 to make it easier to write tools that we the transaction actually lots of those now the way to create small right tools and see that can read the transaction log and do stuff with them so input of my country there is a tool called all PTX-loaded dont which is just a debugging tool and that's the white reference implementation of how you use this but I wrote this little tool that can read the warts and to most of you are started doing this just basically to check that the interfaces OK with the expert reader stuff that Congress world but the Minister of Robert Frost put all the list but this doesn't actually work it's kind of unfortunate but I think it will work if you also enable checksums which kind of solves the problem with that indicates that so there's a little caveat is that it requires you to also enable checksums which is another new feature not to intrude on but I think of the work of that so would would how
it actually works it 1st of all it will copy all the files that are not normal relation follows like all configuration files and out our control file and all kinds of map visibility and free space mapping also currently just copy at is that's that's always say it's always to just copy what's there and the master over but so but the optimization here is that it uses transaction locks check so wouldn't have to check that copy over the actual data files which of the majority of your database on the user's transaction logs to determine which blocks have changed and then people only confuse over those that
is on you can use it pretty much way the way I think it's so the typical scenario would be that you have a master and the slave it's called a and B and a goes down and you from all the to be the new master and then you bring in a back up and you wanted to become a standard wire so you at that point you would use this tool be rewind or 1 of the best mechanism to send it back up with that of the new master of the you the way to the mapping it yet yes we you can do on if you if you can control this thing is good you don't like have a coveralls or anything you don't actually need to like this you can just shut down cleanly and you have to somehow ensure that you get all the transactions what's copied over before you do the time which then you wanted to like is it will be safe to just make the old master a slave I'm not sure if we have documented that or if we what kind of unnatural kind of checks cities have signed these checks for that long but it should be possible to do that just created recovered from filing a mass 0 it'll become a slave all the indicates that we should go on and on what didn't understand which is the source of the other you know what you would do you want to do that and the only thing the Montserrat can't understand perhaps we can figure out of line
so so this is how you use it 1
cool thing with this tool is that it doesn't actually require additional acts as it does all of these truly PQ on happens to be this nice helper functions in process which requires superiors access and the necessary but there is there happens to be this nice little functions you can do the job building functions the least directories like Achilles today the directory and there is functions to fetch or read files and send them over so I just use those and select where is the result of this was a mastery of so it's Fernando
itself there but it's not country because of the stuff that underestimated there so so we
are seeing gender history rewind
as well you can you can actually use this mechanism to seeing any as long as they have a common ancestor right of so if you have a situation where the if you like for core from here and you have like for timelines that all branch off from some ancient history timeline you can use ROC teachers still sing out on any of these timelines is the was work and the same with the C rewind it just means that it has to read all although all from where they branched off to different branches in time that's really all I have constant questions instead of big questions more questions this you know what you think about it and it would be of the you know the all of the time of the year and it was not part of the unit of work that the new OK and it's also called who the been adding more sanity checks the recovery my coat that over years and in general this has because of all this messing around with timelines and restoring from back ups on stream recreational the stuff that we've been adding more sanity checks in this code that's over the years so if you take 9 1 0 there's a lot more but the opportunity to screw things up and have a database that starts up silently and you will notice that it's actually could completely corrupt it's getting harder to do that but there are still ways to screw up yes what are you going to do is you have to be on the the all the question is now when when you when is it safe to do a little time on history for us and what do you get if you never intend to based back up and that's the oldest based back up you ever interested in than anything older than that and on thank you feed your car architectually not all that's a common yeah I know I have no idea how it feels to all this pixel for while the low-noise I guess you could extract the timeline from that I think it's designed to do but as I said they're small so just keep them around the house because do you really think you really really in the life of the people in the presence you know what you're going to have to store the nest in the light is associated with the data is stored in a database through the constant and you can also yeah so does no more questions on on unfortunately done thank you
move yeah this is 1 of the things that we need to work in the area of the of the of the of the of the union of the of the people of the of the of the of the of the the and here is the way in which we used use in the in the in the in the in the 2nd half of the universe changes and if you really want to say that the only way you can be in the way in which the history of the region and the world the at the end the of the the the the view of the data and we know that we have to do it in the in the in detection the data the then in the in the in in the in the in the in the likelihood of the data and this is the only way you do this in a that the in to the in the in the in the the in the in the in the in the in the in the in in the in the middle of the room and this in you know there we and this have is a lot of the media and the beginning of the year in which you're going to use the the the the the
Server
Punkt
Flächeninhalt
Rechter Winkel
Datenhaltung
Server
Punkt
Wiederherstellung <Informatik>
URL
Tropfen
Datensicherung
Wiederherstellung <Informatik>
Tabelle <Informatik>
Server
Punkt
Einfache Genauigkeit
Datenfluss
Computeranimation
Wiederherstellung <Informatik>
RFID
Objekt <Kategorie>
Zahlensystem
Transaktionsverwaltung
Datensatz
Server
Punkt
Wiederherstellung <Informatik>
Versionsverwaltung
Gerade
Befehl <Informatik>
Punkt
Ortsoperator
Datenhaltung
Gruppenkeim
Zahlenbereich
Datensicherung
Gesetz <Physik>
Computeranimation
RFID
Wiederherstellung <Informatik>
Netzwerktopologie
Mereologie
Punkt
Wiederherstellung <Informatik>
Overhead <Kommunikationstechnik>
Speicher <Informatik>
Gerade
Lineare Abbildung
Subtraktion
Punkt
Sichtenkonzept
Verzweigendes Programm
Zahlenbereich
Rechenzeit
Packprogramm
Computeranimation
Übergang
Transaktionsverwaltung
Server
Randomisierung
Verzeichnisdienst
Einfach zusammenhängender Raum
Trennungsaxiom
ATM
Punkt
Systemzusammenbruch
Zahlenbereich
Physikalisches System
Elektronische Publikation
Login
Packprogramm
Computeranimation
Datenhaltung
Wiederherstellung <Informatik>
Eins
Netzwerktopologie
Transaktionsverwaltung
Rechter Winkel
Server
Wiederherstellung <Informatik>
Verzeichnisdienst
Gerade
Lineare Abbildung
Punkt
Logarithmus
Elektronische Publikation
Computeranimation
Wiederherstellung <Informatik>
Datenhaltung
Physikalisches System
Vier
Verzeichnisdienst
Forcing
Mereologie
Wiederherstellung <Informatik>
Transaktionsverwaltung
Gerade
Transaktionsverwaltung
Punkt
Mereologie
Mathematisierung
Wiederherstellung <Informatik>
Inhalt <Mathematik>
Elektronische Publikation
Verzeichnisdienst
Computeranimation
Punkt
Mathematisierung
Zahlenbereich
Physikalisches System
Elektronische Publikation
Computeranimation
Wiederherstellung <Informatik>
Netzwerktopologie
Dienst <Informatik>
Transaktionsverwaltung
Konditionszahl
Datenreplikation
Pinching
Server
Punkt
URL
Gerade
Zeichenkette
Aggregatzustand
Bit
Elektronische Publikation
Punkt
Versionsverwaltung
Zahlenbereich
Nummerung
Strömungsrichtung
Physikalisches System
Elektronische Publikation
Datensicherung
Datensicherung
Computeranimation
Wiederherstellung <Informatik>
Rechenschieber
Datensatz
Flächeninhalt
Vererbungshierarchie
Punkt
Wiederherstellung <Informatik>
Information
URL
Gerade
Maschinenschreiben
Subtraktion
Punkt
Hausdorff-Dimension
Mathematisierung
Zahlenbereich
Datensicherung
Datensicherung
Netzwerktopologie
Gerade
Elektronische Publikation
Datenhaltung
Physikalisches System
Biprodukt
Elektronische Publikation
Packprogramm
Rechenschieber
Dienst <Informatik>
Konditionszahl
Ablöseblase
Wiederherstellung <Informatik>
Information
Ordnung <Mathematik>
Instantiierung
Fehlermeldung
Transaktionsverwaltung
Punkt
Rechter Winkel
Datenhaltung
Automatische Handlungsplanung
Server
Systemzusammenbruch
Wiederherstellung <Informatik>
Störungstheorie
Physikalisches System
Systemzusammenbruch
RFID
Prozess <Physik>
Datenhaltung
Güte der Anpassung
Gruppenoperation
Ruhmasse
Kartesische Koordinaten
Datenreplikation
Störungstheorie
Physikalisches System
Systemzusammenbruch
Computeranimation
RFID
Übergang
Client
Datensatz
Transaktionsverwaltung
Twitter <Softwareplattform>
Hochvakuum
Mini-Disc
Bildschirmfenster
Datenreplikation
Client
Gamecontroller
Server
Resultante
Kraftfahrzeugmechatroniker
Punkt
Reihe
Versionsverwaltung
Systemaufruf
Sprachsynthese
Datenreplikation
Elektronische Publikation
Mustererkennung
Packprogramm
Computeranimation
RFID
Eins
Programmfehler
Streaming <Kommunikationstechnik>
Dienst <Informatik>
Rechter Winkel
Mini-Disc
Dreiecksfreier Graph
Datenreplikation
Server
Wiederherstellung <Informatik>
Gerade
Ereignisdatenanalyse
Server
Gewicht <Mathematik>
Euler-Winkel
Graph
Datenhaltung
Server
Wiederherstellung <Informatik>
Datensicherung
Physikalisches System
Subtraktion
Verzeichnisdienst
Synchronisierung
Graph
Datenhaltung
Server
Physikalisches System
Verzeichnisdienst
Systemaufruf
Gerichteter Graph
Computeranimation
Punkt
Minimierung
Implementierung
Regulärer Graph
Login
Computeranimation
Homepage
Wärmeübergang
Hacker
Konfigurationsraum
Schnittstelle
Expertensystem
Elektronische Publikation
Synchronisierung
Datenhaltung
Relativitätstheorie
Mailing-Liste
p-Block
Elektronische Publikation
Ein-Ausgabe
Mapping <Computergraphik>
Transaktionsverwaltung
Rechter Winkel
Gamecontroller
Binäre Relation
p-Block
Mapping <Computergraphik>
Open Source
Kraftfahrzeugmechatroniker
Fehlermeldung
Transaktionsverwaltung
Punkt
Ruhmasse
Quellcode
Gerade
Resultante
Open Source
Lineares Funktional
Fehlermeldung
Prozess <Physik>
Prozess <Informatik>
Gebäude <Mathematik>
Verzeichnisdienst
Computeranimation
Kraftfahrzeugmechatroniker
Videospiel
Subtraktion
Synchronisierung
Datenhaltung
Verzweigendes Programm
Regulärer Graph
ROC-Kurve
Code
Computeranimation
Streaming <Kommunikationstechnik>
Einheit <Mathematik>
Geschlecht <Mathematik>
Rechter Winkel
Wärmeübergang
Mereologie
Wiederherstellung <Informatik>
Speicherabzug
Ultraviolett-Photoelektronenspektroskopie
p-Block
Sichtenkonzept
Synchronisierung
Flächeninhalt
Finite-Elemente-Methode
Likelihood-Funktion
Mehragentensystem
Mathematisierung
Hypermedia
Simulation
Grundraum
Computeranimation

Metadaten

Formale Metadaten

Titel Understanding PostgreSQL timelines
Serientitel PGCon 2013
Anzahl der Teile 25
Autor Linnakangas, Heikki
Mitwirkende Heroku (Sponsor)
Lizenz CC-Namensnennung - keine kommerzielle Nutzung - Weitergabe unter gleichen Bedingungen 3.0 Unported:
Sie dürfen das Werk bzw. den Inhalt zu jedem legalen und nicht-kommerziellen Zweck nutzen, verändern und in unveränderter oder veränderter Form vervielfältigen, verbreiten und öffentlich zugänglich machen, sofern Sie den Namen des Autors/Rechteinhabers in der von ihm festgelegten Weise nennen und das Werk bzw. diesen Inhalt auch in veränderter Form nur unter den Bedingungen dieser Lizenz weitergeben
DOI 10.5446/19063
Herausgeber PGCon - PostgreSQL Conference for Users and Developers, Andrea Ross
Erscheinungsjahr 2013
Sprache Englisch
Produktionsort Ottawa, Canada

Inhaltliche Metadaten

Fachgebiet Informatik
Abstract Whenever you perform point-in-time-recovery or failover in PostgreSQL, a new timeline is created. In the best case, timelines help you stay sane in complicated recovery situations and keep your database consistent. In the worst case, they will thoroughly confuse you. This talk explains timelines, how they are used by the system, and how you can stay sane with them and even take advantage of them. Version 9.3 brings some improvements to the handling of timelines, these are discussed as well.

Ähnliche Filme

Loading...
Feedback