The Most Complicated Algorithm I've Ever Written: SQLite B-Tree Balancing
Vložit
- čas přidán 29. 07. 2024
- In this video we visualize how the SQLite 2.X.X B-Tree balancing algorithm works. Technically it's not exactly the algorithm from SQLite but my own version which I implemented for MKDB. My version is almost a 1:1 Rust copy of the original algorithm implemented in C which you can find linked below.
🌐 LINKS
Algorithm Implementation:
github.com/antoniosarosi/mkdb...
Original SQLite Implementation:
github.com/antoniosarosi/sqli...
✉️ CONTACT INFO
Business Email: business@antoniosarosi.io
Contact Email: sarosiantonio@gmail.com
Twitter: / antoniosarosi
Instagram: / antoniosarosi
LinkedIn: / antoniosarosi
🎵 MUSIC
• 【Chillstep】Wayr - Agai...
• Swell
• Could We Say Goodbye
📖 CHAPTERS
00:00 Introduction
00:41 Databases & B-Trees
03:30 The Balancing Algorithm
09:24 Deleting & Balancing
12:03 Final Thoughts
🏷️ HASHTAGS
#programming
#computerscience
#algorithm - Věda a technologie
Me after writing my first linked list
Lol
Me after forgetting how to write a linked list and then re-learning it and feeling like a God again. 😅
I remember putting that on my resume when I first learned linked lists 😂😂
@@minimumt3n204 Google must've been fighting Amazon and Microsoft to hire you 💀
Me after writting my first binary search.
There are thousands of beginner videos out there, and very few which really digs into the details. This is fantastic.
I love this channel!
Solid engineering that isn't just the basics, nor so high level that anything useful is abstracted away. Instead real algorithms and explanations on their utilization in a real world manner.
Keep up the good work
I try to keep it in the middle, too high level would be "B-Trees can self balance, that's it" and too many details would be getting into how the code is actually implementing the ideas that I explain, in this case for example the distribution computation is kinda weird with the way the loops work 😂, but whoever wants to understand the details can read through the source code.
Freaking crazy. All this is super interesting! I like how you explain and guide us through the process. Starting from scracth knowing nothing, but researching and building up until you figure out what to do next
excellent explanation of b-tree insertion, deletion and balancing, I love your visualizations too, thanks a lot!
I have implemented it in my third year engineering almost 24 years ago. It felt like a great achievement.
It is amazing that there are still people interested in algothims since nowdays people are mostly interested in chat gpt and the other AI models. I really appreciate it 🙏 Thank you!
how are people still interested in chat gpt?
most developers i know have turned off the ai assist stuff again as it is need but also kinda distacting
i feel like by now its 90% businessy people trying to sell each other AI hype
LLMs are cool and all but they aren't world changing yet, we gotta wait a few more months for that
Thank you for this series of videos, I am starting to build my database too, and by research found your videos. Was really useful!
Please keep these vids up there are 0 DBMS dev vids
This is gonna be the last database video for now, I'll move to a different project for the next video. All the source code I wrote is available on GitHub and I put a lot of emphasis on documentation for people interested in writing databases, so feel free to read through it if you want to get into the details.
@@tony_saro thank you , you are such an amazing developer wish you all luck
There is a CZcams channel called "CMU Database Group"
@@tony_saro there are lots of components in your dbms implementation though, i see no reason why you shouldn't make a video on them as well, for example sql parsing
@@energy-tunes I'm reserving parsing for the "Writing my own Compiler" video 😂, it's gonna be much more important there.
Phenomenal video… very few people make content about the algorithms and technology we take for granted… the db implementation video was insanely great 😊
You've explained the principles of a B-Tree really well that I would be comfortable trying to make it myself.
I think a basic implementation would not be that hard, but a performant one would be substantially harder.
A basic implementation is simpler, you don't need the distribution algorithm.
Keep going, you are the best doing the content I saw in CZcams for a long time!
Great content Tony. You know you've mastered something when you can ELI5 a concept like this. Keep it up!
We need more of video like this!! Good stuff.
aquí apoyando esta nueva iniciativa
i did this shit last year, i had found this book on data structures at work so i had the bright idea to implement them in my favorite language, c++. (i do nodejs at work and i would rather do low level c or c++ so i better get cracked at it before i try switching jobs.) i can safely say that deleting a node from a btree and rebalancing the tree was some difficult shit. it was hella fun though, and I built a lot of cool data structures that i might have a direct use for in 50 years
Great video Tony. !!
Thank you for sharing your knowledge; thank you for your excellent job. !!
My best wishes for your channel.
Fernando, Buenos Aires, Argentina
im taking the CS50 SQL and watching this videos is like, me watching what i would be able to understand in a few months, but is 5am and i cant sleep so lets watch this now
Love this series, thank you ❤
Amazing 🤩, Really appreciate the hard work.
Nunca comento nada, pero vaya. Todos tus videos son top!. Te sigo desde el otro canal y siempre que puedo le doy un vistazo cuando tengo burnout. Me recuerda que todo es con paciencia y muchos prints jajaja. Felicitaciones, Sarosi. De verdad, increíble!
Una barbaridad de prints 😂
Beautiful narration for the algorithm.
This is a really good exercise. I wrote one a few years ago when I was learning about databases myself. I never thought about looking at SQLite source code though, I was implementing it based on the wikipedia article - definitely a grind. Nice to see more resources being created for it.
SQLite is pretty simple compared to MySQL or Postgres so it's great for learning. I'm talking about SQLite 2 though (written in the early 2000s), SQLite 3 is more complex but still simpler than MySQL and Postgres.
Amazing work, please keep it up! :)
Fantastic content!
Also please keep sharing your experience about developing mkdb. your first video was very informative🙂🙂
Beautiful piece of art!
this is insane, really impressive
Aww this takes me back to my data structures course 20 years ago
hermano te felicito, que buen vídeo es este, me encanta la informática a bajo nivel y tu contenido apunta a eso mismo, algoritmia, arquitectura de computadores, SO, etc, en lenguajes de programacion de hombres rust c/c++, al área y conocimientos mas puro y duro de las ciencia de la informática, esto si que es contenido de Ingeniero, seguí así.
Awesome video Antonio, Goat 🐐
it's a monton. Excelente video, lo tengo que ver como 2-3 veces para terminarlo de entender.
"bored"?!?! This was amazing!! New sub, thank you.
I like you trasnparency. Keep doing 👌🏽
Buen video Antonio, ahora puedo aprender backend e inglés a la vez. Éxitos.
Great job dude! Please give us more of this senior juice lol
Hermano, buenísima explicación!!!
Tus videos son buenísimos como explicas y las animaciones que utilizas hace que sea fácil de entender, yo estoy haciendo una app web con java para visualizar las operaciones en arboles B que indique todas las lecturas/escrituras. Complete la inserción y hacer la realocación de punteros cuando un nodo se divide y promociona a medida que el árbol crece fue bastante mas complicado de lo que pensaba, pero ni de cerca es tan recursivo y elegante como el que mostras acá 😅
As a front end dev, this video glued me to the screen
this channel is a hidden gem
I remember failing the assignment of doing a b-tree algorithm.
We must protect this man at all costs
🫡
Thanks for the great content
Thank you for the super
I learnt about this and red black tree but have never implemented them. Will implement them soon.
Good luck 👍
What a great video!
The time honored tradition of reimplementing perfectly fine software in rust, but worse. Thanks dude.
Gotta rewrite everything in Rust and make it blazingly slow 🤙
Hi, you and your channel are amazing! Thanks for this content really.
One question, if I may. Could/would you consider doing some more 'in depth' coding walk throughs we could even follow along with you? I'd personally love to see more code, even if it means longer videos and/or topics less complex than b-tree balancing 😁 But even this code I'd really like to see more of. It's a wild one man. Amazing work!
I am not sure how to do that, I wouldn't do 1 hour tutorials of basic "hello world" projects, there are a lot of channels that already do that. Streaming is another option, maybe I can stream smaller projects that would take max 10 hours / streams and then reupload to CZcams. But for now I'm focused on explaining ideas/concepts, trust me, as I said at 12:11, if all you have is the code then you'll have a very hard time figuring out how it works. I don't even know how many hours I spent understanding this particular algorithm but in total I spent months with it. If I had had a simple 12 minute video like this one I would've probably spent days instead of months.
When you write code you also have to get into the details of the code, not only do you have to explain the idea behind the code but also how each line works. For this particular algorithm, explaining the code would probably take 12 hours instead of 12 minutes, because you also have to understand other parts of the code base such as the pager and cache system. The code base is 25K lines, code only is probably 12-15K, the rest is documentation. So projects like this one would be impossible to stream / teach, it would take hundreds of hours to write and explain all the code.
Simpler projects might work, as long as I'm not getting into "hello world" territory. In the meantime, if you want to read the code, it's on GitHub, it's thoroughly documented, some parts of the code are explained in my videos, so it should be easier to follow than most other codebases.
amazing video 👍👍👍
Amazing content
That is brilliant
Hi, I have several questions about databases in general: Do the b-trees store only the references to the data (and it's stored somewhere else in the file in a heap-like fashion) or are they also used to store the actual data? How are the variable-sized data types like TEXT/BLOB stored? What happens when they're inserted/modified/deleted?
My full database video answers some of your questions. I'll link it below and I'll answer anyway.
czcams.com/video/5Pc18ge9ohI/video.htmlsi=uTQ94tOYGIk4cQuo
B-Trees store the entire rows, all the data. Text and blobs are stored using overflow pages, you basically break the data into multiple chunks, store each chunk in a different page outside the B-Tree and link all the pages.
Inserting texts or blobs is no different than inserting any other field, you deallocate or reuse the pages previously occupied by the blobs if you're removing or modifying, and when you're inserting you just use the algorithm from the video.
Quality❤
I once spent an hour solid going thru the SQLite source code, also not understanding what I was looking at. Still don't. But it gives you an appreciation for the genius of D. Richard Hipp (creator of SQLite). Anyway, this is a cool project and I'm motivated to make my own simple database. Thanks!
This indices corresponde to offsets in a file ? I mean indice 5 for instance corresponde to x bytes from the bigining of the file ?
No, 5 is the primary key of the row. There are no offsets in this video.
Buen vídeo, bro.
this is really impressive. it's almost like software engineers are still needed and ai isn't going to take over any time soon. lol
AI won't correctly implement an algorithm like this one from scratch any time soon 😂
agreed. i still have work to do. the only thing i need to watch out in the ai world is which ai is giving me less errors.
This isnt really even software engineering either. Very few software engineers write these type of stuff, as fascinating as it is.
@@Alexander-zt9kz I'd say it's actually more "engineering" than application development, at least in the traditional meaning of the word "engineering". You have the theory, or the "science" which tells you this data structure has such and such time complexity / space complexity, the data structures must comply to such and such rules, there's some random paper that proves it, and now good luck translating that to code that actually runs on the CPU.
Mechanical or electrical engineers are supposed to apply math and physics to real world problems, so shouldn't "software" engineers apply math and computer science to real world problems? The reality though is that most of us build APIs and we're sending JSONs back and forth 😂😂😂.
Ah …not so sure about that. AI has started to develop algorithms…Ai has already created one to improve matrix multiplication. Something we thought was already optimized. Don’t deny our soon to be lord and savior.
During Uni I found classic Database algorithms always simple compared to the nightmarish stuff needed in Distributed systems.
Any particular algorithm from distributed systems you find nightmarish?
Mine was the Guassian Blurr for image processing. We had to do it in C and cuda. I still have nightmares about it.
Bravo!
Te felicito por haberte expandido al habla inglésa, muy buen contenido .
Great video! Great animations and visuals (is it manim?)! Great explanation! Great pace and rhythm. But the code font is too close to Comic Sans for me... 😅
Not Manim, it's just standard Adobe Premiere. Code Font has many haters 😂
Hey this is amazing! Do you mind sharing the source where you learn from? Be it book, website, course or whatever you used to actually learn to implement this algorithms and the whole database
I explained it in the full database video, check it out. The resources are also linked in the README that's on GitHub. But basically I used the "CMU Intro to Database Systems" course and the code of SQLite.
Wow, That is a great explanation. B - trees are advanced version of 2-3 trees. Which holds most of the properties of b-tree. I have written AVL trees on my own. So, To my understanding up and till this point, a b-tree or even a binary can be represented using a linear data structure like an array using an expression of (2*i)- left child & (2*i)+1- right child. In contrast, (i) is an index of the current element in the array.
So, can leverage these to store information on disk as we know in storage we can get access to a 4kb page for each read operation and 4kb is size assured by hard disk manufacturers that if the power goes off. It can write 4kb data to disk without losing it. so, it helps us with crash recovery.
I am trying to build a very simple database without complex data structures to make clear how this disk stores data & is retrieved at a low level. I am from a PHP background. I know some basics of c programming & javascript. Why do you choose Rust to build this database? why not c or C++.
C doesn't have a standard library so you have to write a lot of code or use third party libraries. The standard library of Rust allowed me to write the entire database with 0 dependencies.
On the other hand it has modern features that C lacks (genetics, pattern matching, closures, traits). C++ I guess you can use it, but I don't know any C++ I've never used it. I chose a language I already knew.
Because C is outdated feature languge
Are btrees shallow generally shallow? The algorithm probably has some configs, like the node capacity, nr of children.... how to choose them? In theory, the more capacity the nodes have, the less depth tree will have, is this correct?
It's correct. If by shallow you mean not deep, then yes B-Trees should be shallow. It's impossible to calculate the exact depth since nodes store variable entries so you don't know how many entries each node will hold. But there's a configuration for that, each node must store a minimum of entries, in my code it defaults to 4 entries. So root has a minimum of 5 children, then each child has at least 5 children, so third level is already 25 pages. The depth will grow very slowly. The node capacity is determined by the page size, which is a power of two between 4 KiB and 64 KiB. The greater the page size, the more entries each node will store so less depth.
I may be weird but I like this algorithm and even if it seems complex I have some urge to implement it myself :D I remember from university when we went through C and we implemented a lot of basic trees and algoritms and it was fun (until some memory problem appeared :D) Nowadays I just use some library from someone a lot smarter who is able to use math to prove it. Thank you for video - again cool animations and very easy to follow. Cheers 👍
You'll learn a lot if you implement it but it requires a lot of effort. You need to implement the entire B-Tree if you want to write this algorithm 😂, and you need a system that loads pages from disk into memory and writes them back, SQLite calls that system the "pager".
@@tony_saro Yea, sounds like I'll pass pager and implement just B-tree part with numbers/strings for fun :D I don't feel that good about low-level IO operations :D I would like to check Rust or Zig when I have time so that could be good combo
Thank you very much for sharing, it is great you continue this topic. Unrelated question, only now I noticed your chair. What brand/type it is? Are you satisfied with it? I am asking because my back is killing me (too many hours in front of monitor) so I am looking something sane 🙂
It's Razer Iskur. It's not a bad chair but it's pretty expensive, if you want a good chair and you have enough money you should probably get an ergonomic chair like the ones made by Herman Miller, that's what I'd buy if I had to get a new chair right now. But don't take my word for it, I haven't personally tried those chairs, I just heard good things / reviews about them. Gaming chairs are mostly for aesthetics.
@@tony_saro Many thanks for your honest reply, it is very helpful. Yes, I also heard heard about HM, one thing that stops me more than price is weight -- this is wrong direction, steel and heavy bicycle is around 13kg, digital piano 14.5kg, a chair something like 20kg 😀
@@DevelTimeWell the Razer Iskur is 30Kg 😂, so I wouldn't have a problem with weight, especially if the chair is comfortable.
20 years ago I studied data structures and algorithms. I remember red black trees and decided I would try and do a b-tree for laughs… I lasted a day before giving up and going down the pub. We’ll done on your success. I’m pretty convinced I made the right decision. The enlightening part was the special case of the toot node. I’m pretty sure that is where I was getting confused. I might even try my own implementation… though I may end up on anti -depressants this time round…
20 years ago you couldn't use the internet to find implementations or explanations about how the algorithm works, at least not as many as today, all you had was probably books. I don't think I'd be able to write a database in 2004 😂.
GOAT🔥🔥🐐🐐
Hey , great work man ! Btw which tool you use for animation?
Adobe Premiere
I have two questions:
1. In case of underflow root, ehy the page numbers are not modified when the depth gets decreased by 1? In that case the child pages have a wrong page number
2. Are the indexes loaded into a btree upon initialization?
Thanks
Only the page pointers are modified when reorganizing entries.
Indexes are not "loaded" upon initialization, they are created with the "CREATE INDEX" statement and then they exist on disk.
so many steps and cases
Great viz. This is quite different from the B-tree as I understand in book, maybe there are different variants and implementations like B+ tree B* tree. I need to review. Is file system work in similar way? I know I can look it up though
Yes, file system works in similar way and explains why there is limited number of C:, D: drive for us to choose in Windows file system formatting. There is also B# tree. I have no idea about it at all. B in Btrees, is possibly for Bayer-McCreight, 1972
I think implementing a self balancing tree is medium size project, meanwhile I may choose to solve 75 Leetcode problems
Technically this is a B* Tree, but it's just a name that Knuth came up with in his book "Sorting and Searching".
@@tony_saro Good Classic Reference, What I talked about is from Sedgewick's, Sedgewicks is student of Knuth (Ka nooth)
gracias por los subtitulos crack
Do you have links of learning ressource you used ?
Go to the GitHub repository, they are linked there. The respiratory is in the description of the video.
@@tony_saro cool thank you !
If a page overflows while inserting, how does it fit into the page? Like if my page is just 4KiB how is it possible to allow overflow even temporarily.
Ps: I’m trying to make my own dbms from scratch. I had the idea like months ago. Your videos have been a great help, thanks. Keep uploading!
Good question, the exact implementation is a little intricate, but a basic straight forward idea would be, just use a growable array, Vec in Rust. For performance reasons, I used something more elaborate, here's the direct link to the documentation:
github.com/antoniosarosi/mkdb/blob/bf1341bc4da70971fc6c340f3a5e9c6bbc55da37/src/storage/page.rs#L1023
Hace poco lo implemente en java y fue bastante complicado,sin la recursión no podría hacerlo
La insercion y eliminacion de un Árbol B fue realmente un reto
Hahaha... i remember doing this B-Tree rotations on the uni. But didn't seemed so dificult back then. Have to try again and see how rust I've become.
The rotation itself is not complicated. The complicated part is generalizing the rotation for variable length data and page allocation/deallocation.
Dude that's dope! Could you share with us how you build such good animations? Just After Effects / Davinci Fusion? Or any additional tools? How much time do you spend per minute of final video? I guess at least 1-2 hours per minute?
Yeah it's about 1-2 hours per minute. I don't even use After Effects it's just Adobe Premiere with the "Essential Graphics".
@@tony_saro Hey man, thanx for responding that quick! I have same Wallpaper as you @ 12:09 on the right Display! :D
Did you learn this kind of video production with basic Premiere Pro tutorials or are there special kind courses? Can you suggest any to learn from? Or any payed / premium ressources?
I want to create vids like yours / let's get rusty's /Dreams of Code's etc on Linux, FOSS, Vue, Golang and Rust as well, but can't find the right resource to get started.
Anyhow, subscribed!
I didn't use any courses at all, you can start with a basic Premiere tutorial on CZcams and after that search for specific tutorials, "how to do X in Premiere". That's all I did, but keep in mind these are not my first videos, I have another CZcams channel in Spanish which I started in 2019, so I have a lot of experience with editing.
@@tony_saro thanx man! yeah, probably just need to get started and then it comes from alone. Going to do a Davinci Resolve course next days on Udemy (canceled my Adobe CC long before I switched to Linux from Mac - even more evil to me than Apple, Google & Co, on the same level as Oracle). Looking forward to more vids on your projects here on CZcams!
"I decided to write my own database" ... lol how much red bull were you drinking that day? XD
Yeah you don't wanna know
So, it is used for indexing in SQLite - is it used only for "primary key" indexing or any other index ?
Primary Key and other indexes like "UNIQUE email" or any column you want.
@@tony_saro What about "CREATE INDEX idx_lastname_firstname ON employees(last_name, first_name);" ?
It’s used for any index where you’re indexing based on a data type that can be sorted/compared. If an index can present data sorted by some criterion, that index will be done via b-trees (primary key is just one of them).
"i don't think anything will come close to that in the near future"
*B+ Trees sipping a beer in the corner...*
I mean from my perspective, I won't code anything as complicated as databases anytime soon. B+ Tree shouldn't be that complicated with all the code I have, the algorithm from this video is reusable, the overall structure would require many changes but going from a B-Tree to a B+ Tree is not as complicated as going from nothing to a B-Tree.
@@tony_saro fair enough, i didn't code neither of the two, i just know how they work, so shout out to you for the effort.
Just wait until you realise there are special optimized algorithms for when you are looking up more than one key at the same time. Those will make B-trees look like cute little things in comparison.
I implemented some ranged queries, for statements like "primary_key BETWEEN X AND Y" or "primary_key > X". Other than that you can't do much else with B-Trees.
I think I need to correct that I mean the kind of datastructures you use to optimize those searching. You typically end up with some variation of a k-d tree, which you then need to turn into a page optimized version. The optimized ones also have "fun" optimizations such as fractional cascading, which is somewhat of a nightmare.
Oh I'm pretty sure there are data structures more complicated than this one. I'm just speaking from experience, I did not implement any of them. I wrote some complicated data structures for a memory allocator project, but not as complicated as disk B-Trees.
Great 👏👏👏👌👌👌💪💪💪
so how did you do the animations? :D
Adobe Premiere
por fin señales de vida
🤙
Removal in BTree is a nightmare
Just a bit hahaha. This is the worst part, took me a lot of time to realize:
github.com/antoniosarosi/mkdb/blob/bf1341bc4da70971fc6c340f3a5e9c6bbc55da37/src/storage/btree.rs#L821
Me, when I've passed a variable with a reference
Por que ya no subes en el otro perfil tus videos en español ?
Lo expliqué en Twitter y en Instagram
I like rectangles.
Ah yes, rectangles 🟦
What do you work on in your actual job? Also do you have a spanish chanel?
Web dev.
czcams.com/video/HLMPUrm376E/video.htmlsi=5CGkJ-axdHpVx3wB
Ya pero qué fondo usas?🤔
😂😂😂😂😂 nooo, otra vez no jajaja
Who knows prof Saraj Sahni behind this Btree algo?
commenting for the algorithm
Thanks 📈
I would like to paid a course that how you make the database with Rust from scratch
I don't have any courses yet
You convinced me to never write my own DB - thank you! 🙏 🙏 🙏
You're welcome 😂
Que paso con el español?
instagram.com/s/aGlnaGxpZ2h0OjE3ODUxMDI2NDY1MjQ5NzA1?story_media_id=3399154947095379898_1492796422&igsh=bzl3MTJ4ZTNvNHd4
on my dsa course at university we learned about btrees, and usually to learn about programming i prefer learning by doing instead of just listening to theory. bad idea, could only manage to implement an insertion on btree after like two days of trying to get it to work, and even then i couldnt manage deletion. then also came the b+ trees and i couldnt do it lmao
I also learn by doing but if you literally have no idea what you need to do it's kinda difficult 😂. This algorithm in particular I just read through SQLite, added my own comments, print statements, modifications, etc, until I understood how it works. Then I made my own version.
Deletion on Symbol Table is hard. Refer to Sedgwick’s lecture, LLRBT deletion took years to complete! I tried hard to go through why it works
We studied balanced trees in second year in cs , and it was brain cracking 💀💊, that was because we learned in heavy subject (file structures and data structures), wich i saw useless untill 4 months later
No seguiste haciendo videos en español?
No
Isnt this B* tree? B tree doesn't borrow space from siblings when insertion overflows
Yeah, but that's just how Knuth named this variant in his book. Technically it is a B*Tree, but nobody even knows what the B stands for 😂
Llevo un año sin saber que te habías pasado a la comunidad inglesa XD
Me he pasado hace 1 mes