Merken

The PostgreSQL replication protocol, tools and opportunities

Zitierlink des Filmsegments
Embed Code

Automatisierte Medienanalyse

Beta
Erkannte Entitäten
Sprachtranskript
let's get started good afternoon everyone i'm of magnesite I'm here to talk to you some about the replication protocol enables this consultant primarily on from Stockholm in Sweden work for a company name rental improve those little red link or not and we do open source consultancy and writing focuses on various sites and obviously PostgreSQL instead of but here to talk to him about because Chris replication protocols of the tools and the opportunities around this and things that happen like that so I notice a number of people in your group actually being in previous versions of this talk about how many people know that they've been in previous versions of this talk really I want to give you the chance to you know sneak out silently beginning because if you see the previous version that hasn't changed much there a few minor things but in general the roadmap for was happen has been fairly clear over the past couple years so the only major difference is that a number of things have moved from the area will be opportunities to now actually being tools which is the direction of would like to see things so when we talk about the post replication
protocol we talk about the new stuff that we had imposed 9 right the streaming replication part uh hot standby is obviously important but streaming application is the particle level in the lower layer of how we transport and how we deal with the actual replication traffic be between our primaries lights that is based on streaming the transactional or as we often call it imposed was that the right to have low it's exactly the same thing 2 names sometimes but that sort of what we always start from what we call it base factor that's the terminology we use for whatever some databases will call the same thing a full backup for a sample and we're going to base back and it's basically small it's where you start from and then we use standard recovery can the same 1 you use if you restored from backed up the same 1 you use if you're just no crashes of these posters doesn't graph the this is the however you're running it on my crash particularly if you're running on virtual machines are applied during the strange things like that and the opposes replication protocol is layered on top of the standard but of product so basically all of this replication stuff is layered on top of something else we got the actual replication is layered on top of recovery and restore functionality the protocol itself sits on top the standard posters for which has given us a lot of flexibility and functionality completely for free and the post the the current version of the post protocol has been around for several radius I think by now it's not something we change often and then we're going to change it follows into new version but the core of the protocol is very stable thing to sit on top of and has a lot of functionality so looking at the
different parts of the puzzle that we're going to talk about 3 this talk uh the collection processing start of how that actually works in close stress for those of you who are not aware that the posters particle itself which is again the foundation of how we do this by the replication specific protocol that sits on top of this and then sort of moving into the tools space which is has been very much focused on the PDE-based back to that was added in post 9 1 1 which was actually when I started doing this talk the 1st time that's the part that was entirely in the opportunities part because wouldn't have any of these tools we just had the role particles in mind 1 0 that we get working so sovereign offline connection has posed work uh so what happens
is it's a TCP connection you can also have a local Unix domain connection posters and we're getting more that they are they work exactly the same as in replication of scenarios you probable that I have a TCP connection replicating locally over 2 to the same machine and it doesn't make much sense other than possibly test so we get it sees the connection and we get the kernel taking care of all the TCP packets going back and forth and have these be connection running on port 5 4 3 2 now at this point there were posters does is for what you'll see lot of other databases is still spent a thread aerial associate a worker thread somewhere with using existing connections and that the mean you will use professional user multiprocess models would create a completely new process that will take care of this specific connection and now we've done this we do a potential SSL negotiation this happens immediately after the 4 and this is 1st level where the part because with particle itself negotiates as a all by not checking does the client and server both support a where the settings all the client parameter of SSL mode and what those potentially HPA file say about cell and once that's done if we determine that SSL should be useful its connection then we handed over to the actual SSL protocol level negotiation to set up encryption verify certificates and all this this is the 1st thing that happens as soon as we get him what we've done this we get a number of parameters that are sent from the client to the server unconditionally at all times we get the database name I we get even these are we using to map to the row in PG HPA that called to figure out how we're going to do with allocation of this and a number of other options such as client encoding the sense that we need to know for every connection are sent at this point and then we reach this we have enough to find the role in PDA speed of called so we can refer them to speak perform authentication however that happens to be configured whether it's username password we might do a covers exchange to arguing with my checking all that long and we might use SSL certificates log in but happens during the early start anyone who performed authentication assuming it successful right if it's not successful were just permit the connections and kill the process complete but assuming it's successful and we select the database and that basically means posters will will run ch deer into the data directory all that specific database but will also prolong the cash with the number of entries from the system tables that are specific to a set of and once we've done this we are ready to start processing queries to start doing things as we enter the query processing where we just stay forever in this back just run a query we get a query that response in new query center response we keep doing that until fluid on the connection but this is a neuronal posters quite if we change this into what it looks like when we're talking about a replication clients it's very similar this actually just a couple things we changes we do the username option thing would change it because there is no database name 1 work great creating replication connection but in that it's actually a hard coded database and use the another point selecting a database if we don't have a database and obviously we're not enter a query processing loop because we're not going to process works where we are going to accept that is the collection of exactly the same place and we're going to 4 and the US association same way ever going to perform authentication the same way in exactly the same place in the code to the same line the same C 4 and the big advantage of this is any way you can set up and use authentication for your regular posters client you can also use authentication for your replication flights so if you set up you know Active Directory integration with covers and GSS API and what not for your clients if you use this for your replication of all and I found this little of myself I found this to be really really useful in many deployments and so I'm happy to say that no I was initially against layering this way on top of the authentication because I thought it would be too complex and it was definitely worth it I withdraw that objection or rather I did a while ago and is shown to be very useful so the things that we changed again uh which change the database username options to just get the database it's going to be replication we still get the username was still get the option and that was not something called around centers instead of entering the part there were still inside the same process we don't start yet another process is just we call a function called whilst and remain in the post back so what is is under it is basically
it's a special purpose both the back and like me a regular back is always connected with a single database this is why you can't change databases cross database queries post-cursor back and look to a single database loss and reason not just exist it only accepts very very very simple queries and comments that I think there are 3 commands in total that incentive last very very simple it will return the these are commenting and this is where running on top of standard protocols leads a regular commands it look the same way you can actually run these commands from PS if you're careful with exactly how he started and you enjoy watching binary data on history and it returns a mix of result sets which are formatted just like any other posters resulted in streams which are fermented exactly like the copy strict and if we're looking at basically because the line for knowing where this is the very simple case it also supports a single thing which is the client connects it tells the server give me all the transactional starting at point x the server goes into the mom would start sending the transactional and it never stops ontologies reconnection goes away but no further amounts ones the client has said give me all the data that's where it's just and keep something so taking half a step back we said this is a layered on top of the posters particle
so most of the posters for local it's very simple thing about this very quickly and it's nice when there's a simple protocol for minute repulsive way too complicated so it's only can be TCP well it's always going to be screaming stopped in we can have a Unix domain soccer but it works the same way is a message-based protocol it would generate its fixed messages pass along its bi-directional is no special streaming protocol that is only 1 way the particles always 2 ways we can optionally SSL encrypted and when we do these the entire TCP stream is wrapped in SSL we don't try to so do encryption at the layer of an individual message or anything we just use whatever SSL library we have currently that's open a cell it's the only 1 we support and the protocol those restrict that obviously and just wrapped the whole screen in SSL in that case and the message as simple
Israeli and and this is the simplest way you can write you have a message type that is 1 right there are some 30 sure so I think message type defines in total in the form of all the message length that's a 32 bit integer so a single message will never be greater than 2 to power of 32 and usually there be a lot smaller than that and then you have the message itself which is obviously the fabric of message is depending on exactly what the message types so a standard query exchange between
client and server just look something like this we have a message type already for query that sent from the server to the client that is the code Z. or zee depending on where you're from we do try and let Q and posters to understand both the differences between British and American English so the final that year truth understands canadian other like you might have some issues with the view of trying to connect relegating their messages French that point and you get a sizable we do these days as opposed to focal version 3 s along with this was some the transaction status information which is just are you currently in a transaction and it's this transaction new broken or not but the client will respond or will at the time time run a query will send a simple query message which is the Q message and it has a length which can be the length of the strawberries that we're running the coming up there and then we got the actual SQL for just straight and as a text it's sent to the Paris it's always that's optimized it's executed we get some sort of a response and the response to a standard where it is going to start with a row description message which tells the point the format of the specific resulted and the team message that's size it lists the information about each individual column like what data type is in this column how big is the data fits variable-length data types sort fixed and then we send 1 or or more of their own message if you get 100 rose to get a 100 data of message as the size is about and when the whole theory is gone we return all the rows were returned a command tag message that just tells you basically what did we just it sort of the unbound now message so in this case will send the command taxi and select telling us exactly that that that's what we did now if we apply this a change to the streaming replication probable it looks so surprisingly similar are made of surprising but it does look similar we got the same ready for query message is being sent by by these while center protocol this is why we can talk to it with something like to PS 2 all we can filter with anything that he's asleep picking you can't open the JDBC driver if you want to and look for it's a good idea of what a good and we send again a simple query message this is the same message that we used to run a regular part so we some simple prior message if you want to start their application we send this topic engine and the 0 slash 0 that's the post way of formatting the transaction and location 0 slash 0 means you just set this system of that's the very 1st transaction would actually don't think we have 0 slash their more I think those change it actually starts at 1 so that we can use of 0 to identify in our the those change in 1 1 million 108 try to look around at people could confirm this formula there looking confused thank you but I have 1 reason I wasn't looking at you but that's for confirming anyway you know they're not there were there were some issues around with it you recovery actually crashed exactly at the segment boundary it would come up thinking was not that was bad so the difference now is that instead of sending the data had the message was sent a copy of response this is the same kind responses if your query was actually and that the W. message with a size number 2 flags centered in streaming replication they will be 0 and 0 1 of them tells you how big is true where is the how many columns on each row if regular copy from a table with 3 columns 1 of those will be afraid but we're not stopping regular tables that 0 the other 1 identifies this is a binary copy is not a takes the binary not because obviously are transactional is a binary loss and then undercut the day message which is just the the size and then some binary transactional data and after that I will send another copy data message and I will send another copy data message and will just keep going and we will never solve the actual sizes so these excellent data messages will depend on how your data is being generated but as soon as the data is written to the transaction log it will be sent across the replication connection and as we were an impulse because my point 0 this is really all there was to the replication fork that's so we have and then in 9 1 we started building a little bit further on this added some more functionality well
for 1 thing we have a Synchronous replication right but I'm going to cheat mold actually talk about how that integrates the problem but it's really it's it's a secondary set messages to flowing from the uh replication slave back to the master telling it how far along it we have also added something called hot standby feedback at which use whereby the slave ships information about open transactions back to master to avoid standby query cost and that's also really more of hot standby feature on the lower layer when we're talking about the the replication protocol itself and the tools that is we actually had something we can sort of a micro SQL micro language used for the well so that it can now accept a number of new commands it can do more things than we could ever before and
basically we now have a full grammar running allows sentiment previously the whilst the command was no if string compare input would stop replication do this else out I don't know what to do but we do have a full grammar it has very few command and it has a very few options but it does exist there and it's being more is being added to requirements for what this is going to be extremely picky about how you if you felt standards conforming strings made it hard to make get things exactly right This is a lot more picky that it's not intended for manual consumption it's intended for tools it's intended for tools like EEG-based back its intended for the replication slates and nothing other than that but it's also forming a foundation for future improvements we can build more tools using this low-level binary access to the people persist in the future we might be able to use it to talk about things that aren't necessarily transaction for those sitting here work in science clustering and briefing as with all the other nite that kind of stuff from may well be layered on top of the things we can do here because we have a generic particle so these
are the commands that we actually have available imposed is 9 . 1 so it's still not all that much we have an identify system in a post was 9 1 0 3 simple command the connect around identify system using rollback that tells you the system which is used to make sure you replicating from the master that you're supposed to be replicating from not just some random other posters database and and that's pretty much it there is a 2nd column and I can fill my life remember what the 2nd call Mr. theory can be important and where the stock replication command works exactly as before some stock replication you get a position then if you want right it's going to be a stroll connect to the database named replication typing start replication 0 slash 0 and you know what you're transactional come back and be patches great but it's interesting when you do that by mistake but the new command that we had in mind 1 the 1st of the new utility commands added to the this so micro language is based back which it's a number of parameters and all that kind of tells you what it does for it will form of fact that's that's the only new commands fine on 1 I we haven't actually added any new commands at all in nineteen with added a new flag uh to base backup ugly but not on actually met so there is that it
does is it gives you a single command that starting to look like I was in in a talk earlier today about the wonders of sequel server and how it so much better impose 1st and 1 of the things that I was you can connect run back up over SQL well guess what we can do that if you don't necessarily want to do it from your mom might find uh that you run the base back so you no longer need to deal with this manual in connected via stroll run PG start back take your backup connect again from the you step back and and don't forget to make sure that everything worked fine and absolutely don't forget to actually run the stop back up or your backups are useless so we are all completely integrated because still control things like backup label if you actually care about lack of labels you can control checkpointing control on the sense as no it's not and silver bullet big this is not what everybody wants to use for backup but it's very simple it's very easy to use and terms of work pretty well below method is still there if you want to know but that most people don't need to know what the new method is more robust and so along with this
that the protocol is still you're not intended users from annual consumption by utilizing the SQL and you will get your vase back act as it could be resolved right on your screen uh which you probably know what so this is why we have told the due dates back but related this is the tool that we added post 9 1 1 so it's not available and I know it's a 9 1 will be in future versions of course and it's also designed to be used as an integration point for third-party backups if you using 1 of these super expensive enterprise backup systems they can integrate with I don't actually know if anyone does that yet so if someone you know 1 of the big backup systems that actually speak the same protocol that would be interesting to know so please let me know if you want but it's there for them to use so it's to fully documented procollagen official protocol it's not considered an internal particle it's just look for end-user consumption so this training the action backed up so that you end up receiving are simple far
towers very nice standardized format that hasn't changed since I since they started building computers so but is very straight many other archive months like if you use prior or something like that actually have a header in the beginning part and have a head it's the sign to be streamed because it is designed to run on tape and you really can't do seeking on you have to stream urinate Ontotext so that we can have a all files in in a car archive are aligned at 512 by byte 100 again this is also a legacy from talking to tape right but it turns out the imposed 1st data files are already aligned on an 8 kilobytes by punk so the only cases where you losing sort of wasting space by using this is for maybe a configuration file and control much money in PD control but style European for the majority of the database is already aligned at these 8 kilobytes which is obviously also aligned fire 12 so you don't actually what you going to receive is is 1 clustering per table space the content tablespaces you will receive 10 half also 1 for each if you were in the different scenario where you have only 1 table space well you will receive only 1 tough and will be transmitted sequential we don't currently have support for backing up multiple tables basics and how would just them 1 by 1 so if
we look at this in the the particle so it's very much like the other stuff we started with with our ready for query message was sent back to the base factor label food saying this is the thing I wanted to Monday's backup and then we end up getting his 1st we get a regular the PQ results so regular results with 3 calls SPC alright STC location and size that's the tablespace altered by the the tablespace location the full power of where this status was developed and how big is tablespaces by default this size is going to be minus 1 meaning it's not the lecture and and then for each sentence around with information about it and then we get that mI'kmaq act that says we're done with this letter now without waiting for any further
data were gonna send a copy of book this is something that you will look like if you're standing multiple queries to post before the semicolon between was just taxi and results on on the end of each other g a copy of response they and it looks exactly like the streaming replication which is we send the wealth of member of Poland 0 and yes it's going to be binary and then we just stream particle number times when we're done we send a copy down message which is which is the controversial the command taxing saying OK were gone this far and and another copy of response for your next tables and we just repeat this until would send all the data we need I want her down there we go back into the loop and you can run further comments if you want but some of the things that we will to be actually well absolutely energy use actually use acquire Wireshark they know how to disassemble particles and they will 1 December the Conference of the your Cognitive but they will know that it's a complicated message so you can definitely someone see that this is actually what happened and it's it's fairly simple so that that's the simple way of how this is done by an argument would normally used for this
phase we use PDE-based you're deviate actually use this feedback that use the tool feature-based factor that we should pass among 1 it's a simple command-line tool like was the ones we have it takes a couple of options and takes Dutch capital the should tell you which directory we're going to write false and it takes a Dutch capital f option which works like if you only if you work with Peter Dunn correctness number these the best capitalists the formant it in the plane or it can be talk but this does not to change the format of how were sending the data across the wire will always sending it in format but if you choose formant equals plane Pd-based backed up will unpack the tar and regenerated data directory whereas if you use format to plot the T is just broke HR file and then you can do whatever you want but it's far from the that's will control how we do the checkpoints whether we do a fast checkpoint spread checkpoint this is the same as the 2nd parameter if you were using the PDE start back up functions the 1st parameter is the backup label the 2nd parameters of Boolean saying when he wanted to a fast word spreadsheet that's this process that we can set the labeled with a Bachelor we can enable compression with best at the link if we do compression it happens in PDE-based back at the data stream that's coming out of PostgreSQL never actually compressed but we can write compressed files so messy for using compression pg . where the data flowing from posters PG down with the uncompressed and welcome precedent like this now on top of the cells were running this thing over the standard PQ protocol we accept all the standardly PQ final but the same once you'll seen PDE-based back in the dumps like that capital you use hosting the name but for number all of this if you're using Kerberos it'll go talk to cover if you need a passport will you have the same kind of parameters you have for the other tools were always ask for a password never ask for password and they are they look the same they work exactly the same but we have another parameter that the sensor problem that a lot of people have been complaining about that which is you cannot
that capital P. that he would get a progress report and it will tell you how far ahead in your back up your from you know 0 % to 100 % it and it's fairly expensive because what it does is when you start the back at it enumerates all the files in your data directory and some of and some of the size of the and sense that belong to the diverse back and then start all over again by sending the fire so the enumerating all phosphites now if you have a large number of databases or a very large number of objects in your database this can be quite expensive was for traversing along files if you have don't have large number database is a huge number of objects and it's gonna be the cash anyway and that this is not to be that expensive to just figure out how the the files are it's also going to be an exact right because it gets the size of the files when you start to back up the files your databases still online it's still making changes while taking the back up the size the file might well be different to the point that you know somebody else might drop huge table while you're doing sizes instead be wrong or sentences slowly table and the SI system it's going to be wrong so actually in the universe is that it didn't always end hundreds % what they know what we always ends at 100 % because when it ends we just adjust the value to 100 % this we know we're done at but sometimes you can see it actually running up to 100 % and then staying at 140 few minutes because the database increased in size as the back of mind and we don't try to adapt to that in any way during the wrong we just add up and when we get to 100 per cent we stop adding up I wonder if we might actually sold at 9 unintelligible but the point being it's in exactly gives you good hint it gives you a hint whether you know you're halfway there are no time to build a couple more profits come back again based
package combined with transaction logs to give us some other things when you when you restore from the base factor you use of but require while packed transactional work like many of you here have set up an archive commands and you know remember that the files over there make sure you check the exit codes all this stuff it's kind complex to set up it can be complex the monitoring of based backup brings us 1 more feature that's very easy for the simple use case which is if you add the bashed its parameter to Europe EDA-based back up all your base backup will include all transactional all that's required to restore this back you don't actually need a lot of archive anymore for a physical back and if you do that you don't get the ability to the point in time recover that will still require a lot of but you'll be able to restore justice 1 back because of based backup without any transaction log built in a post this will start up and say sorry I can't do anything but this is what happens here is that the about center will just include all the required while false at the end of the traffic the very end of the back of just send the transaction log that's required and this is basically all the transaction log that was generated between when you started the back up and went back for those of you who work with the replication fork and we have a parameter called barkeep segments that you will need to use in this case that tells the server to keep enough while around because if you run this parameter and have activity on your server all your alone we rotate the transaction logs at checkpoints you get rid of them and posters that leads the transactional before PDE-based backup has received it well Pt-based backup will say you know sorry I spent 5 hours they back about naggin really sorry I don't you know it's not going to silently back the doesn't work but it needs all those transactional files that keeps segments lets you keep expressing affection around to make this work and that's the stuff that we have so far that's the stuff we're having imposed 9 1 and this is where those of you have seen this before it on my was when I switch to the opportunities side but we got 9 2 heating paid enough so that these things are no longer on the opportunities side there now on this help with testing site but
1st of all we have a tool now that lets us use training of our and so on just show mental blend set up logo archiving using of the time for a number of you how many of you have note to set but that time not by the state so that words and POS previously a when you said archive command business will run this command per once per 16 megabytes of transactional it will then copy this transactional to the archive this is perfectly fine if you have a high-velocity database if you have a low-velocity database that my you know take a day or 2 to generate 60 megawatts of resurrection more commonly made another day but maybe an hour that means that this is your recovery granularity you want the point of recovery you generate 16 megabytes every 2 days well you know your granularities to date so we have around for that which is a parameter Pollock at time of which says ship these files maybe every 10 minutes even if we didn't generate 16 megabytes transaction which would be perfect if it wasn't for the fact that we still ship 16 megabytes even if we only filled with a hundred bytes so you can really I mean you like this that is all that time of 5 seconds and get a greater granularity but you feel you archive with terabytes of data report this would send 16 megabytes every 5 seconds which might be too much but never really this is that the post of replication particle if you have a applications like already dust replication probably already sense the exact same data would go you about how so for this we have a mutual interest is mine PG receive x that you run somewhere that's not on the database server it will connect to your server as replication clients and reconstruct the log archive for it just as if you were running archive command except it will receive by by by not 16 megabyte by 16 megabytes and it's basically it's supposed replication client without this sort of took out a big bloated thing and just ran a simple thing it takes replications the rights directed files that gets you out of the trouble or rather having to choose between recovery granularity and filling up your eyes with with empty set no if you actually have a replication client already this doesn't really help you because you use the replication client to get it is the last of 15 . 4 megabytes that never hit the box but if you don't have a replication clients and this is used so you can get rid of his chronological and based on the same thing that we have a new parameter for PET-based that
basically take the same process but could you receive excellent us and runs in parallel during based back so when you run a loop EEG-based back in this mode where you take the base back up and you want include all all the required transactional they FIL just fire the 2nd connection that streams the transaction while it's running the back so you no longer need to set while keep segments master you know that longer need to keep the while around master because receiving while almost only backup node while we're receiver-based back in power that that is very simple you just add the best aspects of equal screen instead of just that X which is also believed that test pixel equals copy is the same 1 as the old and it'll just for the big difference being used you now need to replication connection to with running the service separate connection with to makes stream some same connection another thing other we were working on that found further use of the base factor in mind you is
relocatable tablespace when you were thinking based back and stress between version 8 . 0 and 9 1 1 and there is a very annoying thing which is you can move the data directory school is supposed to move your data directory so you restore it into different locations but if you have any non-standard tablespaces they must be restored into the same place it's not supported to move their tables this is the new location is not how it works but it's not support and it kind of works out the thing is that the implementation prior to posters non 1 surprised was his 99 1 another is there is a system table called PD tablespace space which has a column called SBC location it has an overriding not dictate like every person subject and SBC location that tells you this table lives over here and elsewhere restoring we can modify this because it's part 1 moments system table so we have to restore 1st and then change its meaning that we need to restore data into the right location 1st now when you create a table space we also put a symlink in the PG underscore TBL as director in the data director that points to wherever this thing and the fact is even if you look into earlier versions proestrus the prior to 19 the SBC location field in PD tablespace is never used but once it's SEKT when you create the table space never used it's only the semblance it is only used by the tools we need to find out where tablespace let's result useful we actually access tables they only use the similar to what we did in the 9 . 2 is we just that the lead that the FTC location feel it's no longer in the details instead we have a function call pt get tablespace location that but turned it by looking at the same that means that acid mine to its fullest supported to move the table space to restore it into a different location than it was before by just changing the similar uh which is quite useful now has over in 92 to uh it's still going to be manual right you restored tablespaces you manually modified similar that will leave that part of the all here's are still the opportunity space and 4 9 3 is to just add parameters like Pt-based or hybrid sports will do this for you but it's not only stored in the database so we can actually access and modify it before we start the restored there but I have 1 other thing that's sitting on my list on the opportunities part
of that used to be a very long list we worked on this brouhaha which is some sort of ability to to do incremental backups I don't think anyone is actually figured out the best way of doing this 1 way of doing it allows people people to do incremental backups when you use the step back and if you step back up is use are so we can build something or single light into the poster from the question is whether it's going actually win depends on what problem really trying to solve but I found myself when running this on production system such that actually doing this last can be pretty darn destructive for performance on the server wonder taken aback resource inquiry generates a lot of random I Pt-based back generates nice sequential which is a lot faster with low so those things to be done that is also discussion what can we do using the Ellis and we have a another Ellison on every page of every day of uh every data for telling us which is the latest transaction the touch this page it we can use that to generate our own way of doing incremental backups there some issues around that as well it's not quite as straightforward as it might sound with this things could be done around that find a nice way to be able to decrease size of archive without taking more full that's like being in right now you can decrease I mean you have to keep every log file from your base backup and forward so the only way to keep your archive smallest tape-based backup very often but the base backups aerobic fitted find a way to consolidate these things together to optimize that but I think we're done a pretty good job because again about a 3rd of this talk used to be opportunities about a year ago so long as happened in both is 9 1 and some were numbers mine to I'm sure there's a lot more that we can build on top of this both from the concept-based back up and within replication obviously but what we have now is a very solid foundation that we can build on the and that's really mean that's the way we usually end up doing these things in both before I put in a solid foundation but somebody else Figure of apologies that's a furthermore tools in the space and the more enhancements of these tools obviously if you have any good ideas on enhancements that you'd like to do yourself another 2 black to just for people that this would be good to use were very interested in hearing about that and I think I actually have that I do still have about 1 minute for questions so any
questions this is the 1st of all your 1st this you actually better at so I didn't want to live with that and that argues that companies and that see of in thing addition in fact a lot of balls never happened was that yeah exactly and that the dashed exist for is if you that X and in plain text not going not text what they know it will recreate the data directory will put in the PDX so you can actually just PDC tailstock right in that director interrupt that's the use these fanatics he don't need to settle of archive you don't need a recovery of call for anything to get back up and running it's 1 of back up that includes everything you wouldn't you wouldn't necessarily use this that creates and uses the image that's of so you others so that certainly will get you logical replication systems that often work across the use Pd-based backup to set up a replication streaming replication slight but then you don't use the best expert that x is from 1 of back but what if the Union of these which but this should all be actually it might not be the tablespace moving part by helping the everything else should be for document as well as all of these other than the the fault about incremental backups such a upon actually exist but the thing is to get so what mean what or you can connect the to I think there is specific mention that you can do with the PSQ all that but it's fully documented it's standardly PQ connection so that no so there it should all the information should be there some of it is under is actually in the protocol documentation and not in the you end-user user documentation but it's it's all fully documented except I actually think you may have a point in the we never actually updated localization of interest the there you can use that the you will be able to run the best backed up against standby and in 9 . 2 yes injustice looking at the bacteriostatic this yesterday so that that's a beta version right but uh there there a couple of ago to open box on the specifics of running the base factor honestly right now but it will be they will be fixed before 19 itself and then 1 you can't run conflict can only run mastering on 1 of the was the waiting for the check point having a master thing is enough of the offended somebody standing before you that I believe this a protocol PGR man and there have been doing that and I would never use the incremental backup there myself I've heard someone who used it at an early point in time and it looked like it work and everything was correct but I think that fix that problem about that united looking to have the diamond dust that is obviously a good step forward figure out I don't know of anyone other than the german stuff on at but I have would be the main issue yes so that only those with the use of and then get from from the courts of perspective we have this particle or we have the core part of it were hoping to see maybe get some money to court and that all men use these features instead of some of the things that the German has been doing and sort of happy because that's the only thing what we can do it it's what everybody's numerous of hacker on the limitations were trying to take the way the limitations of the model with axis well so that's so I think we are premised on my maybe running over a don't have schedule right now but but so this will be the last round of talks so then after this we have the closing session which is right back where we all started with the opening session the keynote and I don't think there's much break so thing down would appreciate it if once you're done here just sort of had straight over there and get ready for the session so thank you as you
Subtraktion
Web Site
Protokoll <Datenverarbeitungssystem>
Open Source
Güte der Anpassung
Gruppenkeim
Versionsverwaltung
Netzwerkbetriebssystem
EDV-Beratung
Zahlenbereich
Datenreplikation
Binder <Informatik>
Computeranimation
Richtung
Videokonferenz
Gewicht <Mathematik>
Flächeninhalt
Datenreplikation
Protokoll <Datenverarbeitungssystem>
Mereologie
Prozess <Physik>
Versionsverwaltung
Wiederkehrender Zustand
Regulärer Graph
Systemzusammenbruch
Kartesische Koordinaten
Datensicherung
Datensicherung
Raum-Zeit
Computeranimation
Wiederherstellung <Informatik>
Übergang
Virtuelle Maschine
Umwandlungsenthalpie
Standardabweichung
Code
Datenreplikation
Protokoll <Datenverarbeitungssystem>
Transaktionsverwaltung
Umwandlungsenthalpie
Einfach zusammenhängender Raum
Lineares Funktional
Radius
Protokollierung
Prozess <Informatik>
Protokoll <Datenverarbeitungssystem>
Graph
Datenhaltung
Datenreplikation
Biprodukt
Quick-Sort
Teilbarkeit
Rechter Winkel
Mereologie
Wiederherstellung <Informatik>
Speicherabzug
Partikelsystem
Streaming <Kommunikationstechnik>
Normalspannung
Resultante
Betriebsmittelverwaltung
Retrievalsprache
Einfügungsdämpfung
Vektorpotenzial
Konfiguration <Informatik>
Prozess <Physik>
Punkt
Binärcode
Computeranimation
Eins
Übergang
Kernel <Informatik>
Streaming <Kommunikationstechnik>
Client
Regulärer Graph
Datenreplikation
Mixed Reality
Notepad-Computer
Gerade
Softwaretest
Lineares Funktional
Parametersystem
ATM
Vervollständigung <Mathematik>
Datenhaltung
Stellenring
Abfrage
Mehrprozessorsystem
Konfiguration <Informatik>
Arithmetisches Mittel
Chiffrierung
Menge
Rechter Winkel
Client
Server
Verzeichnisdienst
Standardabweichung
Tabelle <Informatik>
Total <Mathematik>
Fluid
Mathematisierung
Zahlenbereich
Zellularer Automat
Code
Datenhaltung
Überlagerung <Mathematik>
Loop
Virtuelle Maschine
Informationsmodellierung
Domain-Name
Datensatz
Authentifikation
Endogene Variable
Thread
Passwort
Abfrageverarbeitung
Ortsoperator
Assoziativgesetz
Einfach zusammenhängender Raum
Digitales Zertifikat
Ontologie <Wissensverarbeitung>
Protokoll <Datenverarbeitungssystem>
Programmverifikation
Einfache Genauigkeit
Datenreplikation
Physikalisches System
Elektronische Publikation
Menge
Integral
Objekt <Kategorie>
Loop
Mereologie
Authentifikation
Partikelsystem
Streaming <Kommunikationstechnik>
Bit
Total <Mathematik>
Zellularer Automat
Computeranimation
Streaming <Kommunikationstechnik>
Message-Passing
Domain-Name
Bildschirmmaske
Datentyp
Programmbibliothek
Protokoll <Datenverarbeitungssystem>
Ganze Funktion
Touchscreen
Leistung <Physik>
Dicke
Protokoll <Datenverarbeitungssystem>
Stellenring
Abfrage
Chiffrierung
Ganze Zahl
Ganze Funktion
Richtung
Partikelsystem
Streaming <Kommunikationstechnik>
Message-Passing
Standardabweichung
Einfügungsdämpfung
Punkt
Formale Sprache
Gewichtete Summe
Versionsverwaltung
Wiederkehrender Zustand
Kartesische Koordinaten
Oval
Computeranimation
Deskriptive Statistik
Client
Regulärer Graph
Standardabweichung
Fahne <Mathematik>
Datenreplikation
Lineares Funktional
Dicke
Sichtenkonzept
Abfrage
Marketinginformationssystem
Sinusfunktion
Randwert
Transaktionsverwaltung
Menge
Rechter Winkel
Server
Dateiformat
Information
URL
Simulation
Message-Passing
Standardabweichung
Tabelle <Informatik>
Rückkopplung
Subtraktion
Mathematisierung
Zahlenbereich
Physikalische Theorie
Code
Ausdruck <Logik>
Datensatz
Endogene Variable
Datentyp
Störungstheorie
Transaktionsverwaltung
Einfach zusammenhängender Raum
Protokoll <Datenverarbeitungssystem>
Antwortfunktion
Überlagerung <Mathematik>
Datenreplikation
Physikalisches System
Quick-Sort
Druckertreiber
Rückkopplung
Loop
Offene Menge
Mereologie
Wiederherstellung <Informatik>
Streaming <Kommunikationstechnik>
Systemidentifikation
Konfiguration <Informatik>
Ortsoperator
Zurücksetzung <Transaktion>
Formale Sprache
Formale Grammatik
Zahlenbereich
Datensicherung
Physikalische Theorie
Computeranimation
Physikalisches System
Fahne <Mathematik>
Datenreplikation
Randomisierung
Cluster <Rechnernetz>
Ortsoperator
Formale Grammatik
Videospiel
Parametersystem
Datenhaltung
Softwarewerkzeug
Systemaufruf
Datenreplikation
Physikalisches System
Ein-Ausgabe
Dateiformat
Neunzehn
Konfiguration <Informatik>
Transaktionsverwaltung
ATM
Partikelsystem
Zeichenkette
Standardabweichung
Binärdaten
Einfach zusammenhängender Raum
Wellenpaket
Protokoll <Datenverarbeitungssystem>
Desintegration <Mathematik>
Gruppenoperation
Ablöseblase
Versionsverwaltung
Fortsetzung <Mathematik>
Physikalisches System
Datensicherung
Term
Datensicherung
Computeranimation
Modul <Datentyp>
COM
Server
Gamecontroller
Vererbungshierarchie
Numerische Integration
Partikelsystem
Bitrate
Unternehmensarchitektur
Touchscreen
Resultante
Einmaleins
Stromlinie <Strömungsmechanik>
Computerunterstütztes Verfahren
Extrempunkt
E-Mail
Datensicherung
Datensicherung
Raum-Zeit
Computeranimation
Reduktionsverfahren
Streaming <Kommunikationstechnik>
Vorzeichen <Mathematik>
Turm <Mathematik>
Konfigurationsraum
E-Mail
Default
Transaktionsverwaltung
Schreib-Lese-Kopf
Leistung <Physik>
Tabelle <Informatik>
Folge <Mathematik>
Raum-Zeit
Datenhaltung
Magnetbandlaufwerk
Systemaufruf
Abfrage
Marketinginformationssystem
Elektronische Publikation
Dateiformat
Teilbarkeit
Quick-Sort
Packprogramm
Datensatz
Arithmetisches Mittel
Gamecontroller
Dateiformat
URL
Information
Partikelsystem
Streaming <Kommunikationstechnik>
Message-Passing
Tabelle <Informatik>
Ebene
Resultante
Rückkopplung
Konfiguration <Informatik>
Hyperbolischer Differentialoperator
Prozess <Physik>
Zahlenbereich
Zellularer Automat
Datensicherung
Datensicherung
Computeranimation
Eins
Streaming <Kommunikationstechnik>
Loop
Multiplikation
Verzeichnisdienst
Standardabweichung
Datenstrom
Datenreplikation
Endogene Variable
Kerberos <Kryptologie>
Passwort
Quellencodierung
Phasenumwandlung
Lineares Funktional
Parametersystem
Protokoll <Datenverarbeitungssystem>
Abfrage
Elektronische Publikation
Binder <Informatik>
Teilbarkeit
Konfiguration <Informatik>
Energiedichte
Tabellenkalkulation
Client
Lesen <Datenverarbeitung>
Dateiformat
Wort <Informatik>
Speicherabzug
Partikelsystem
Streaming <Kommunikationstechnik>
Verzeichnisdienst
Message-Passing
Tabelle <Informatik>
Standardabweichung
Web Site
Punkt
Mathematisierung
Zahlenbereich
Protokollierung
Datensicherung
SI-Einheiten
Datensicherung
Komplex <Algebra>
Computeranimation
Arithmetische Folge
Grundraum
Hilfesystem
Tabelle <Informatik>
Softwaretest
Parametersystem
Elektronische Publikation
Datenhaltung
Elektronische Publikation
Ultraviolett-Photoelektronenspektroskopie
Menge
Gerade
Teilbarkeit
Packprogramm
Objekt <Kategorie>
Transaktionsverwaltung
System F
Komplex <Algebra>
Rechter Winkel
Server
Codierung
Verzeichnisdienst
Verkehrsinformation
Tabelle <Informatik>
Subtraktion
Prozess <Physik>
Punkt
Wellenpaket
Zahlenbereich
Kartesische Koordinaten
Datensicherung
Datensicherung
Computeranimation
Loop
Streaming <Kommunikationstechnik>
Knotenmenge
Client
Datenreplikation
Protokoll <Datenverarbeitungssystem>
Parallele Schnittstelle
Parallele Schnittstelle
Touchscreen
Leistung <Physik>
Softwaretest
Einfach zusammenhängender Raum
Trennungsaxiom
ATM
Parametersystem
Pixel
Datenhaltung
Zwei
Applet
Datenreplikation
Elektronische Publikation
Teilbarkeit
Quick-Sort
Packprogramm
Transaktionsverwaltung
Dienst <Informatik>
Differenzkern
Rechter Winkel
Server
Wiederherstellung <Informatik>
Wort <Informatik>
Partikelsystem
p-Block
Streaming <Kommunikationstechnik>
Verkehrsinformation
Aggregatzustand
Resultante
Maschinenschreiben
Stereometrie
Folge <Mathematik>
Momentenproblem
Atomarität <Informatik>
Versionsverwaltung
Zahlenbereich
Implementierung
Datensicherung
Datensicherung
Raum-Zeit
Computeranimation
Homepage
Data Mining
Prozess <Informatik>
Datenreplikation
Hybridrechner
Figurierte Zahl
Tabelle <Informatik>
Parametersystem
Lineares Funktional
Synchronisierung
Datenlogger
Datenhaltung
Gebäude <Mathematik>
Güte der Anpassung
Einfache Genauigkeit
Systemaufruf
Mailing-Liste
Physikalisches System
Biprodukt
Quick-Sort
Packprogramm
Arithmetisches Mittel
Transaktionsverwaltung
Datenfeld
Rechter Winkel
Physikalische Theorie
Mereologie
Server
URL
Verzeichnisdienst
Normalspannung
Tabelle <Informatik>
Punkt
Quader
Versionsverwaltung
Abgeschlossene Menge
Unrundheit
Kardinalzahl
Mathematische Logik
Datensicherung
Interrupt <Informatik>
Computeranimation
Informationsmodellierung
Gewicht <Mathematik>
Perspektive
Datenreplikation
Kontrollstruktur
Inverser Limes
Hacker
Bildgebendes Verfahren
Metropolitan area network
Einfach zusammenhängender Raum
Umwandlungsenthalpie
Addition
Expertensystem
Protokoll <Datenverarbeitungssystem>
Betafunktion
Stellenring
Systemaufruf
Physikalisches System
Packprogramm
Quick-Sort
Teilbarkeit
Scheduling
Rhombus <Mathematik>
Offene Menge
Rechter Winkel
Mereologie
Speicherabzug
Wiederherstellung <Informatik>
Information
Partikelsystem
Verzeichnisdienst

Metadaten

Formale Metadaten

Titel The PostgreSQL replication protocol, tools and opportunities
Serientitel PGCon 2012
Anzahl der Teile 21
Autor Hagander, Magnus
Mitwirkende Heroku (Provider)
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 10.5446/19038
Herausgeber PGCon - PostgreSQL Conference for Users and Developers, Andrea Ross
Erscheinungsjahr 2012
Sprache Englisch
Produzent FOSSLC

Technische Metadaten

Dauer 49:45

Inhaltliche Metadaten

Fachgebiet Informatik
Abstract The new binary replication protocols and tools in PostgreSQL 9.0 and 9.1 are a popular new feature - but they can also be used for other things than just replication! The new binary replication protocols and tools in PostgreSQL 9.0 and 9.1 are a popular new feature - but they can also be used for other things than just replication! PostgreSQL 9.1 includes server changes to allow standalone tools to request and respond according to the replication protocol. From these, tools like pg basebackup allow a number of new possibilities. And the infrastructure put in place in 9.1 opens opportunities for further enhancements - some already on the drawing board and some just wild ideas so far.

Ähnliche Filme

Loading...