Duration 59:40
16+
Play
Video

Time Series Databases: A Deeper Look - Peter Eisentraut

Peter Eisentraut
CTO at 2ndQuadrant
  • Video
  • Table of contents
  • Video
PGCon 2020
May 27, 2020, Online, Berkeley, USA
PGCon 2020
Request Q&A
Video
Time Series Databases: A Deeper Look - Peter Eisentraut
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
Add to favorites
356
I like 0
I dislike 0
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
  • Description
  • Transcript
  • Discussion

About speaker

Peter Eisentraut
CTO at 2ndQuadrant

Peter has been a contributor to PostgreSQL development since 1999. He is a committer, core team member, and is involved in many aspects of the community.He works at 2ndQuadrant, a global service and support provider for PostgreSQL.

View the profile

About the talk

The term "time series" is popular (again) in database circles. What is it and what's the point? Clearly, a traditional relational database like PostgreSQL can deal with time and with series. So why is time series a special use case?

In this presentation, I want to look beyond the marketing a bit and analyze what the true technical characteristics of a time series database are and what use cases drive it. Then we can analyze how PostgreSQL handles these requirements, which workarounds are handy, and where improvements would be necessary.

In PGCon tradition, this presentation is both guidance for users and a call to action for developers.

Share

Hello everyone. This is Peter eisenhardt recording. My presentation for pecan 2020 from home in Germany. So, let's get right to it. So today I want to talk about time series databases and I usually do it specially to PG Khan. I don't want to talk so much about a specific way of doing things or any kind of particular products, but I want to take a topic and actually asked myself, what is this even mean? and then, Brother asking what does it mean for me as a user or the usual is that I support? And what does it mean for me as a developer of the post office? Quit self.

And that's what I want to do today, okay? so, first premise here, is that time series is a use case in database to call Saturday. So it comes and goes over the years as a trend, Oftentimes it comes as old as the marketing message that it's a new way of doing in the databases and it's a new paradigm or what-have-you. I'd sell my soul to the time series to be a used case just like we have, for example or TP. That's a term that is used a lot but it doesn't mean anything very specific if you think about it. If you are let's say a

database administrator or a consultant and you come to a new database system and they have somebody tells me this isn't cool teepee system, that's Home Inspections. But it gives you an idea of what the system is for and what the characteristics. All right? It's so it's probably You know, how to set the fair amount of updates and her mother reads usually a smaller transactions. But a lot of them into a big, big mainly continuous load, probably pretty good up time

requirements and the date on there is valuable. So it has to be probably backed up regularly and so on and then handy the term to use and then you have Olaf. Would you also tells you things that you do the beds and probably has a lot of weeds, maybe not so many rides water rides. Maybe even a bulk loaded depending on the setup. So you know those kind of understand in general what is system is doing and then now we have to think about it now we have to work with it. But they don't tell you anything

specific about it. Do you know what products to use anymore? Exactly what this is his doing. So and I think time Sears is useful to think of it in that sense. Also that it is a yusuke's, it gives you some of it. A hinge of Detroit to the cities of the system are time to resist nod. Do you know a new database Paradigm or a new way of doing things or I think like that. So I think that's it. That's a good way to frame it and I want to analyze it in a specific way. So what time's use database is useful for?

Here are some of the typical things you could find. So, most obvious one and perhaps in a way that a lot of people don't think about it, is any anything to do a server logs or Latino festival, logs on logs, that have the time stem in some information on it. I think that's the basic idea of the time-series data. And then that stuff has been around obviously forever, but where it becomes more popular over. So, the more recent times has I need anything having to do with measuring Real world, the equipment,

or whether sensors water sensors that you cannot imagine more than whether, but that's a good source. Example, two to think about, And then, you know, you can be in a lot of measurements and terms of air quality. And that's how it's done for us to become more widely available and she parted this more interest in that. But also income to do with the financial World. Those are that's a data that has time attached and obviously there's a lot of interest in that and just because

an obvious case is in 9 science where people need to keep track of things over time and do analysis. Where this is coming from. If you want to do a little bit more of a buzzword approach to this, right? Sometimes when you're trying to going to maybe sell the idea, you need to have a couple passports, you can also deploy these Side of things this you know, popular. It's basically the same idea is having lots of sensors connected to the internet. Mainly because there's a lot of data in those that needs to be analyzed. Either in real time or just do the development process,

has a lot of data and the glass to be also be processed very quickly as can be act in the split. Second that that's an interesting, interesting area of research and development Guerra. golden Freddy not going to put any postgres database in the self-driving cars of anytime soon, but the same ideas, apply now, So another thing to clarify is Dad's time-series database is not the same thing at all as the temporal database. And is absolutely confusing us. Both of those terms have time or form of time in their name. And they

also do, you know, popular terms from time to time in the database field. But they're not the same at all. Today Amazon, but I'm serious. Obviously, if you interested in temporal, I could recommend last year's PG, Khan had a good talk about temporal databases. I'm so in just pulled up since you have a in a simple way to tell them apart, Perhaps it is that in Time series, you have records with thyme stems. Which is the time of Monday, piece of information was recorded or observed.

Where is in a temporal database? You have time stem ranges. Which employs could be in an actual range time or in a general sense, a start and end time stamp and that is the time to arrange when that piece of information was valid, or invalid, a little be balanced depending on what you do. I might be a little bit complicated, but Let's keep those apart. To know what makes a x your state of it. That's the most important thing I want talk about it. I did. So it's a use case a Fuzzy

term for a soul of a collection of characteristics and I have collected here and based obviously on other people's resource and so on phone correct, the mistakes I want to go through. What makes the time-series database? And then how is that relevant to post office? And how can post both satisfy these? Nigeria. I'll go through them in detail. So me a timestamp is part of the key that seems kind of obvious time Stones. Usually in increasing order. Did I usually more inserted? Not so much

updated that obviously becomes interesting for certain departments characteristics. Usually times him the time-series database is have a lot of data. Usually individual records and all the data matter less. It's more tedious aggregate is more interesting. And I'm suited by the usually try to do analytics based on time. So they need to be some support for that so that's look into that in detail. The number one time. Step is part of the key. So that the obvious thing that makes a Time series database, you have data and

every bro has a timestamp associated with it which is when the value was recorded, Or measured or something like that and let me know after I think it would look like this. You have a with thyme stem. And then usually you have some kind of indication of the sensor in this case, or what's your measuring. Now, if your measuring only is a single thing, Where you at when you have one point of measurements, you don't need that. But usually have a lot of those obviously height. So you have some kind of a reference to what device was.

And then, usually the combination of the X Stand and some kind of an indication of what the measurement point was, would be the primary key whether you actually put a primary key on a blast at a time. Stem is a key, right there could also be some reference data. So in this case, The Descendants or table, not having * 10. But these nuts Tables with the data, we have that. How old is sportscaster for Dad? It was really good support for anything having to do with time and date that

it is as Wily as widely regarded for that. You know, I've done t z time that they are tired that you can use for that. What if you somehow don't like that, you can find your own way of recording time, maybe not necessarily recommend, you not here right now, but it's something you could also do. It may be suspended by other the space on a Time. Stone time. If you can measure time somehow yourself and use it into drama, begin to Jehovah if that's something you want to do.

All the types of not special time, stems on a bed special, you can put in whatever you want. This is how I'm supposed to go see as a more General assistant is different from specialized * C. What is databases that with the notion of of time stamps is really eat usually quite heavily based in so you can't really use those systems very well for something that's not a Time series database for specific way of doing things, but if you somehow what makes soil different approaches in a database and they're not very good at that. Because everything is marginalized.

Second Point, timestamps are usually in increasing order. So, that makes sense as you record data as you get data, from the input points, time increases, the value said you record also increases It's not an absolute cuz in a different to a client's, couldn't report out of order, depending on dates and see and then other things like that. then you can go to this tomorrow and requiring that Optimized for Huntington, the will come to that later in terms of what Curry's and planets could do. But this is a An easy thing

to optimize fold it. That you think that? The values are. Channel a pre-sorted on disc. So, how does that work in postgres db3? Implementation in. Post-Gazette is has a special special authorization for this app and use case at Cassius Dio, right? Mostly page, depending values. It is has already has one free cash though. It's her point for that. This is only the optimization. It doesn't have any results of my father used in cases. It's already optimized for that was put in a couple years ago. Also, if you have

mostly sore today. That's great for range partitioning, obviously it will come to that little later, but obviously, in a lot of these, I'm serious use cases. You want to do partitioning Probably by times tonight that would make sense to that way into a few. Has ordered data than you usually only writing to one partition. Xitone partition can be damn hard and get the memory. All the index is memory and all the old part 8, don't have to be kept the memory. So that's how you want to use partitioning. So that works. Well,

And also bring in Texas or useful for that. if you have to rely on, Data being sold on this basically so that it can index Rangers, that's what the name. So if you have it entirely mix stayed up Romantics, can't help Renee Nexus. Summarize ranges of data is very well for you, can you bring in Texas as an alternative to be serving cases and then you have a much smaller index and that's good. So this is all quite well. Some room for improvement here and especially this is something. Maybe you might appoint here but they're

very useful, often ignored by users And that could be more work being done in the planners that to use. Bring indexes in more situations. For example, if you want to order values and do you have bring indexes that already gives you hints of the order of things and the plan or I could make more use of that. So that's something to think about. Other than that, I think twice, cover by post press So then we said the new data is inserted not updated. This is key to this whole web design approach.

If you all setting up a time series database into let's say in a normal oil teepee business database, you usually only keep the current date in. A what is your current set of customers different set of inventory, occurrence of orders, and things like that. And I want things change. You run updates to change another. The address of the customer, the price of a product and things like that died. In the state of bass, you basically don't do that and in the really extreme case, you never update anything. You

keep all historical data. and when there's a new, A new information, you just add a new record for the new times them. Now if you do, this would have been already between way for him. Obviously that leads to problems that space usage and application, Foreman's. And obviously have to be usually a middle ground has to be found. But that is the idea, right? You do and update data. Happen, if you keep all paid up to some degree. No. Anyway, this also overlaps with temple

databases is kind of a little bit of Time series way of looking at it. This is a recording of historical state. up to the current time was the temporal database is Intentionally, describing the range of when information is valid, or will be valid also into the future. but the general ideas, once you have reported something, And I'm seriously doing change it because that is the fact of the past. Now it could sometimes be that maybe Corrections have to be applied, if it turns out, maybe a sense of his faulty all the time, clock was awful, things

like that. So you can build a system around never allowing any updates. But you can certainly optimized for not allowing any updates. Will not having you usually not having updates. So I can post go to work, but I thought she was as graceful out of my to fault. I know he died. Relatively bad, if you need to update a lot. But if you just depend on So this is a really surprising you. Not a problem. Then as we mentioned and alluded to time-series database has often, have we usually have a lot of data.

Because you never did never change anything as we just said. We just keep recording everything new. Business more data, we have the more value we have actually factually true to be debated, but this is how business is to collect a lot of data and then do some kind of analysis on it. Also, do you know because it's possible? Basically do you storage can be had relatively cheaply, so it is more feasible than it was in the past to keep lot more old data. A lot of Thieves. Temp, service in measurement points are relatively cheap.

Maybe you can set up your own weather stations for really cheap money nowadays, and that also applies in it. All the other use cases. And actually, having more measurement points is relatively inexpensive. And, you know, you can also say, if you want more data more value, you can also measure more often, maybe, instead of measuring the weather, once a day on Saint, are you make every minute for all the time? There's no. Hardly any reasonable limit there? The main interesting point here is that in the

time-series database, there is no natural limit of how much data you could have. If you think about this in the old teepee database in the business database, there is some predictable limit of how much data you will have since. So that's a hotel and you record your own now without reservations and you can only have so many people in a room and one party and, you know, there are you going to have Thousand records per day, * 378. how how much data to expect, even if you have a chain of hotels, and you record all kinds of other things for breakfast and things like that, they're still

An offer limited to the multiplier of how much data you will have to even issue our big retailer and online and there is a limited number of customers you have and a limited number of items in your inventory so you can plan it a bit. Christian actors database because of all these factors. You knew there is almost no limit. You can collect as much data as you. You ones, we can afford. And whether that is sensible and valuable one, just passing out a question that, you know,

how much how useful is it to measure the weather, every you 5 minutes versus every hour and, you know, that's something that the main exports will have to answer. But certainly the depression is there are two I was getting more and more into these kinds of database is often very difficult to plan for cuz you don't really know how much data you might want to go. Okay. So how can post goes help with lots of data? As we know, he is not Rates for space usage. Mainly because of two bladder Lauren.

There could be some more Altima stations. Are there are other storage engines being evil considered or have already been put out for that reason. And related reasons to have just more compact storage. Certain partitioning is they are to help with space usage and then based on partitioning charging Virginia partitioning over multiple hosts something that there's some important person for that. But they're certainly lots of ways to improve that. So here's a, a long list of

how do we make? Do I have to leave? Watch the NFL in pussy us so we could work on making the table storage itself, more and more contact different to the rebels representation, or the new batteries details like that. That work is underway. After could be lot of work in the area of compression in all kinds of ways. Compression is Jesus. Is toast applicable to time, Syracuse case. That depends. If you use the scheme, I suppose, I should really aware. You

have mostly numbers being recorded. Then it probably not so much. But in practice people also store in a Json data. Maybe that's the Jason day, does it take to get from whatever the measurement pointers? And then you just recorded, it's obviously not optimal at all. The optimization. That's certainly be useful. And then there's always maybe to do compression to think about a block level, you can use my system if I can compress. Oh, there's also perhaps ways to compressed in a way that specific to the data. So let me explain that.

If you. In a record. A Time series records are consisting of a timestamp and then some items. That is DD items. I'll be measured between the data is probably not going to be very different if you measure every minute. You have That the next time stems only one minute different from the previous one. But also did a data be measured is probably not that different. If you measure, let's just keep the ball in case the measure of the temperature. Again, the temperature is probably either, not going to change it. All was going to have some more. So, you could use.

Some kind of a run length, encoding, or, or something similar to that and there's something more specific there to optimize this Voyage of that. So, instead of recording everything explicit, leave time stamp value value. I'm stuck value value, you could Optimize that somehow and save, you know, the first row you store times in value value and then the next to adjust for the difference. And you don't need to use the pool and 12 x 4. You know, the times table that will advise or whatever you have for the values, you can

just store like a small difference. And that. Could certainly have message massive storage exactly how to represent that you that, you know, what compression method is that a storage method? How would you fit that into the system? I don't know yet. That solar system is also. You know, it would be very difficult to make updatable. So that's more of a 40D. And you would have to see the great and it's been too great for this with all these lower. That's with you. Very

good way to compress things. Lots of options there for compression. Certainly, the partition management could be improved, I think partitioning itself is really. Humans are pretty good now, but the management of partitions is has not been considered all that all that much to do. For example, a very straightforward. Use case would be I would like one partition per month or one partition per day. This is not easy to set up right now. You have all the commands and there is some extensions that can help you manage that it should pass Keys in Florida.

That was supposed to be. I think I would be humble. And then under the bigger scale anything to do with charging is obviously you currently Olney building blocks. And proven, they're all Darrell. number of, Things that could be improved if so, that's a long list of the time. Sew-in individual records medalists. I usually keep my interest in the aggregated data on also older than I was last two points here. And so the fact that the temperature reading at a specific time, a week ago was that

is not that interesting. If you lose that one record in, you have to record of you know 5 minutes before or after that, that's not a disaster. What matters is DD data? So that gives you certain options for. Optimizing apps. Now, just depends on. How do you say sometimes? If this is system of record, obviously, you can't do that. If you're recording, financial data, somehow Propst and copy of all the requirements. But if you or just, you know, measuring stuff out there in the world trying to make

little bit of analysis and modeling and one individual record is not The most important thing. Also, as older data is usually not that important again, that depends on Tuesdays. If you are doing a story on the houses of weather patterns or air pollution patterns and you need deal, they are just as much as new data but if you are observing. Traffic was so much traffic in, you kind of fine, maybe your phone is deviation, then the new data is more relevant than the old attitudes

access lesson. the only that for reference, so, How can post office help with that? Again, partitioning as I mentioned already before is good for that cuz then you can keep a hot date on the cold at a separate. You sense, the kind of use case, you can a synchronous Commit This Very applicable to this. So that you don't have to know. She have a lot of data being inserted all the time. So single robot single, bro. You can turn Asian kind of scum it on and off but we did And then you get a little bit of performance boost and the

risk of losing valuable data since the war in Terminator because you are useful for this. Yeah you can also buy perhaps. This is for the pens on, really do use kaise materialize to use, can you use full data sheet, 1 through old data away and compacted and we compute to materialize fuse up to you before. And you can offer them, you may be using table bases. If you Put all the data. Who's Improvement again petition management. And in this case, specifically the lifecycle management of the position themselves as, you

know, making new partitions, but you're getting rid of the old position facing Lee. we have an open item that in puskas currently, you can attach new positions without a heavy lock, but you cannot detach But that's when they were being worked on. That's definitely an open item. And it just in general, maybe you might think how about? I just want to have automatic, you throw away old partitions at older than 6 months or something like that. Seems an obvious you call me

after you that manually. All the tools are there. Someone is scripts automated but maybe the cookies. And also, the management of tablespaces, is it just basic? and if you want to use materialized views d, Often discussed materialized view incremental, materialized, do you refresh could perhaps be useful? So that you can You still have. New data arriving, but instead of storing it all individually, you could just update to make it Wednesday at 2 depending on you. Use phone that something is off of already being worked on so that might just come in handy.

White Hunter, finally free for all like, how is the database being set up and how is the date of being put in there? And then the questions would you actually do with it? They are. I'm doing that talk to you really depends on the application but generally people want to do I bathe the holidays. If usually also depending again but usually sold quick analytics. So these are not long and Elite exteriors of run might run 4 minutes or hours. You really want a quick summary

is up What, what was my in? What was my traffic in the last 5 minutes? 15 minutes. What was my last traffic per hour over the last day or two? If that's certainly the processing, large files and website traffic, that's interesting. But even if you were monitoring that stay Industrial Equipment, wanted to know, do you know quickly if anything is wrong? And what's happening of his or eating or whatever the case may be so, You want to eat. But you still want these four is basically.

But then, they could also be a boy's long. You no longer running for as if you're doing it again. Weather analysis, over a longer time does good, be happy for Aries and Taurus pistol price. But you usually don't you just a single don't look up. Unless it's maybe you need any men. Are you looking at financial data? And you need to have to look up something that you don't use this system to look up like one record in really quickly. Mostly aggravation. And then on top of that

generally, you want some Cooling. You look through to you, no answer those parades or Explorer data, or we have any kind of visualization or front end zone, in more advanced math, on top of the person. Requirements in high-level tooling as well. So how am I supposed to get support? This know, obviously analytics to port in a skill and general is pretty good and push girls implements. Most of that most of all would be relevant grouping is pretty basic

window functions are in a well-supported prosperous. Now that's useful for all the times you was used as compared to any more specialized databases that have a very limited query language that might be really quickly. A specific use Guzman. If you don't want to do no break out of what. They supported really well then you might not have any support at all. Why do you have to program them to yourself and climb plan? So, if y'all database is naughty, There is some

support in a postgres phone during daytime processing, but actually, surprisingly not a lot and it's an area to move in. So, for example, The youth games that I alluded to the moment ago. If you want to have an app show me my traffic which would be an account. Show me my head in your comments for our over the last. that's a pretty straightforward straightforward Group by Prairie and you could use a trunk to truncate the time stems to the hour and I would park what

doesn't work in the same way as if you want to have traffic by every 15 minutes. How do you truncate a timestamp to O'Reilly and a truncated? If you might want to 15-minute involved for that? You have to somehow make you write that yourself, but it's complicated. but it is acceptable in specially in this area functions and operators to Whatever you need, you can add yourself. but yeah, roof Improvement especially what I just mentioned that taking day trunk and expanding that to support auditory in tolls

and I could also be more more advanced functionality if you want to do maybe histograms based on times and things like that. But somewhat more be, there's some more basic apps that are missing. Which would be relatively easy to, to flock. So there's some work already going on and then, Necessary Improvement in that. Area for that. Same used to give me. Give me up. Information by some Isabel like an hour or 15 minutes. He used a trunk or something similar of the plant that this kind of

losers information in the planter this year, if you an accomplice Brewery & you have sort of input, what the plan of things is for the input, almost, and then you run in as the, in the case of the time, Then you run date trunk and over them. The planet doesn't know anything about that. So it doesn't think that the operative that is sort of anymore and then it would have to use other plans. So this could be some of the difference between the group aggravated and it has racked up yet. Even though we won't leave. We know as soon as we know that

something like date trunk with preserve the order of the input for the planet doesn't have any information on that. So that would be something that's also already been vaguely discussed some time ago to add another function, attribute of some sort to tell the planner that this is an order preserving function of you want to call it, or preserving is probably a good way to describe it. To those two first things off would be really useful. Or supporting basic time blocking, packaging careers.

and they're well within reach, not too hard so and then nurse on the tooling side, there's so many opportunities outside of the score for Visualization framework, Silver Lining Panic reconstruction framework, silver edition, his own extended math extensions for that would be. Cancel. Those is my discussion of Dos. X Series characteristics. here's a summary of the development projects that I mentioned and proposed and ordered in a way that maybe I would do them. This is not necessarily priority by the combination of most useful and easiest to do perhaps

A lot of those two are already in progress in one way or another. Not all of them under the time-series banner makers goes better for this. 70s. I mentioned some of these already in progress over the not so much, so Busy. Most of these are on my radar somehow that's only if someone else wants to look into that or collaborator be welcome. All right, so to summarize. Do I think postgres is, is great for Time series? And it's a universal database system and it has shown itself to be adaptable to many different use cases at the theater showing. Give all your scores

Json and things like that, and get that attempt has sound fundamentals and can be extended in different ways for different use cases. Not that there is. There are always going to be even more specialized systems out there that don't have a specific used case in mind and then we'll beat any dinner up assistant. You know that very well. There's value stores that are In memory, perhaps into optimized and generalized database systems are the anus to be a general Universal database system, but

also have your facilities. That's ghetto. Specific use cases. Lots of improvement as possible but they're all within reach and they're all reasonable. They're almost there already conceived. And in progress. Do it is all that comes together on many of those come together then be even greater time-series database. Okay, so that is my presentation at the actual conference. There will be no question. So look forward to answering those If you're not there, then also feel free to reach out to me. Why are my contact information? If you want to chat about

some of these development projects, while you have some comments on my classification of things? Play slingshot by postgres. Feel free to reach out. That's at I hope to see all of you soon again at a press conference until then take care. Bye. And cut. Is here live for the key. When I go ahead. Feeder. Oh hello. So I'll just go through the Irish sea. Hear the question of how the Bee Tree append. Optimization BoRics was already answered in the chat, that's automatic out of the couple released to go so

you don't have to do anything about that. Ulta beauty point that was made in the chat. About the new feature on puskas 13th at all, the vacuum is triggered appropriately for into. The only tables was actually, it's very good. And that's great and thank you for pointing it out. Question about a migraine in Texas. So if you are doing updates on records, that will potentially break a Britain exit, what would you have to do in that case. So there's nothing really automatic about that. Chance, that question, what exactly should do in that case? Really kind of depends on

the exact pattern of what update you would do? If, you know, if you if it's affecting maybe your entire Plantation, then maybe you can just meet us at that table. Possibly is also special functions for Brendan dax's that they call Brynn, summarize range and bring summarize new values to to make some specific changes. Their props to look into documentation of Brendan Texas to find out details about that. But yeah, I'm principal in general, if you do updates on Bring it back since. You could do about it. It's like problems.

You might have to rebuild. Question from Vic traits about anyone working on automatic partition tration. So I don't know exactly what you would you and Visionaries of the, what I think will we will not do, is create, the partitions was at the time of Insertion. I think that people have to agree that that song Cradle handle is all locking and concurrency problems and real cash on that, that kind of stuff. So you looking for that, I don't think anyone's

really working on that. I would I would Envision as the next steps would be indicating at the time, you cried the partition table to say, and I want this to be a petition by month and then you just have a command that says all the table or in my dad said, make me the next full moon and then he kind of knows what pattern you want. We could also imagine that flash quotations, which is not applicable here are but you can say, am I 160 more taken to scrape them all? So that is something that I think we've looked

into it internally at second quadrant, and we might work on that supposed is 14. Other than that, I don't know any other work on that question. From dawn, dropping old position is the same as a basic layout. So that also requires the to have you locked, that's something Alvaro. Is looking fixing a team? Initial implementation of the concurrent attached turned out to be different from his initial implementation. That's why you had the rewards that something,

but yeah, right now it's, it's required to have, you lock. Okay, what else? Question from just bored. Do you think that with the new beat rain next in 13, that sets to become should become an option for Time series? Well, that's a lie. I didn't think about that at all. I think it's certainly an option to to think about depending on how to use an idea. She she has to specially for a very wide Rose as it's an option more. I'm really thought about it too much but

certainly I would just mess up now and do what they're supposed to do. I did watch those that still talk to her yesterday and then try to make sense of it, but yeah. Chez. More questions. Coming in. Great timescaledb. Extension already exist? Does it already is one of those things you mentioned, you can use it at such as a different license in postgres. So, you know, just a value that for yourself. I think that's great and they are certainly doing a lot of good work there. What I would

Look for his little bit more, some general Solutions. Like for example, the the Planner, we had to scrap. The plan are the function attributes to hit the plan are the way I described that. My understanding is a timescale w, kind of, doesn't it? No more. So it's hard to attacked way and say like this function treated that way you know that's a little thing. So yeah you can use it seems to people like these people seem to like it but I would look in the long run for more generalized, push Chris like

acceptable Solutions in all of these areas. But yeah, that's a that's a step in the right direction that timescaledb. Recently implemented in needed width with buckets. What time is Sun Tzu? I was going to submit those Apache. I submit that, I don't know anyone else that work by at John, Naylor to implement the enhanced version of a trunk that I mentioned. There's not a work in progress right now. Yeah, I do. Look at the time scale to be just, you know, I'm not going to look into the source code cuz there's licensing differences. So I can't

really comment mobile that I'm just reading their documentation and, you know, a lot of the right questions, just may be difficult. Stockpot. Automatic Nobody's really working on that. I missed a bunch of your hands are. First of the video is cut off. I'm sorry. I arrest. Your question was about the Petri Ben. Brennan know. It doesn't do that automatically. Read the documentation. Sorry. I'm going through the Irish sea. Are you read the questions? I read out loud.

Yeah. Okay so you re-ask the question about Prince. Yeah, the question about during the dust. Yeah those out, they do break, there's nothing automatic that fixes it take for Britain to do at a certain adjustments depending on what you were doing. Question about custom type creation documentation, says, types of implement, using mostly sees a possible to type. Using SQL language the answer to that is I believe no. not right now cuz you you do need to Basically, take input string input and converted to a bite for storage to the

bad girl. Can't do that. You could maybe you know, one example is that it's not possible that you can imagine, you doing it in something like Ross, way to lay out and write them on that level. But if not just plain as possible right now, Quotes from David feather. Any winds to be had from simd? I'm sorry. I don't know what that is. Anyone know we still have a couple minutes. Has anyone that got that question card off, please ask again, I'm just watching the RC

I also have a question that I just answered some Mary s46 about wanting to convert units. Like beetles gallons is actually an extension for that postgresql. I should just check it out if you interested in that. Got no dick. If you got an answer to your question, but with Bucket, please, just send me that patch. Yep. There you go. Thank you. Anyone know what simd? Single instruction multiple data. So, is that swim back to processing? I am, I'm sorry, I still didn't know really what that is. I tried that before but I have not thought about that. Probably.

Even if I knew what that was, I don't have an answer. Improvements in execution, could optimization possible that go beyond what I have really. That's all the questions now. Now. That's a processing. Yeah, yeah. So I think it would I think it's could definitely benefit from that. That's Good little bit beyond that. Would I have thought about? And obviously that would also affect, you know, a bunch of other mostly definitely also now, maybe. Those are not there

supposed to be shoes, but yeah, you're the bottom of the list. You could Point thank you. All right, so the in a lot of the items I mentioned are in progress as we have also just discuss. Now some of this already in the next, very next to us some of those being worked on and there is anyone wants to know specifically, like who is working on what may be? No, just drop me a line and otherwise I'll have to see you all soon again, or I'll see you at the next to me first.

Thank you, Peter. Thanks. Thanks everyone.

Cackle comments for the website

Buy this talk

Access to the talk “Time Series Databases: A Deeper Look - Peter Eisentraut”
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free

Standard

Get access to all videos “PGCon 2020”
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
Ticket

Interested in topic “IT & Technology”?

You might be interested in videos from this event

September 28, 2018
Moscow
16
177
app store, apps, development, google play, mobile, soft

Similar talks

Alexander Korotkov
Major Contributor, Committer at PostgreSQL Global Development Group
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
Chris Travers
Head of Databases at adjust.com
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
Jesper Pedersen
Principal Software Engineer at Red Hat
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free

Buy this video

Video

Access to the talk “Time Series Databases: A Deeper Look - Peter Eisentraut”
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free

Conference Cast

With ConferenceCast.tv, you get access to our library of the world's best conference talks.

Conference Cast
735 conferences
30224 speakers
11293 hours of content