Day 3 Lightning Talks

This video is from DjangoCon Europe 2026 in Athens, Greece.

Day 3 Lightning Talks
0:32:19
Published October 26, 2026
55 views

Video collection of all the lightning talks from Day Three of DjangoCon Europe 2026.

Follow DjangoCon Europe on social media:
*X - https://x.com/DjangoConEurope/
*LinkedIn - https://www.linkedin.com/company/djangocon-europe/
*Mastodon - https://fosstodon.org/@djangoconeurope

#django #djangocon #djangoconeurope #djangoconeurope2026

Transcript

4,401 words · auto-generated · en Show

Automatically transcribed, so expect mistakes in names and technical terms.

0:02

Speaker 1: So as we're going towards the end of the day, we have lightning talks. So if you know that your talk was accepted. Please go on that side of their own place. If you do not know if it was accepted you have to go to Zulip and verify uh do you remember the list or shall I oh

0:19

Speaker 2: Yes, so I've had slides off most people. Alex Gomez, I haven't had any slides, and one other person I haven't had any slides. So please if you submitted one, please check now. And if you have slides, send them now, please. That would be great Uh but luckily the first person is here and he has some slides, I believe, and uh we thought we'd give an opportunity to someone at the conference who hasn't had a chance to do much talking. So here we go. It is Carlton Gibson with the first lightning talk.

0:44

Speaker 3: Thank you very much, Andrew It's not a tyranny if the rules apply to everybody, right? I'm jumping in here just on behalf of the Steering Council. Um One job we have is about the technical governance of Django. And um when we were elected uh the existing governance stock was felt to be too difficult to understand and Too difficult to work with on a practical day-to-day basis. So one thing we've been working with with in the background is updating the technical governance. So if I would the request I have for you now is to please check the Django blog today It's a new post being put up by Tim Chilling who's on the Steering Council and it's about the progress that we've been doing and there's a new

1:30

Speaker 3: it explains the background and there's a new DEP which is DEP nineteen which you can go to. DEP is a Django enhancement proposal, and so there's a new There's a new depth that we would ask you to look at, we would ask you to comment on, we would like to get feedback because we would like on the timeline that Tim explains in the blog post there, we would like to get the governance for Django on a modern footing. So that's the that's the main thing there. But just while I'm here, you'll see there's a banner at the top. As of today, there's a new PyTime promotion where it's an annual thing and It it amounts to about ten percent of the DFF's budget each year. And over all the years that PyCharm has been supporting um Django, um JetBrains, PyCharm are probably Django's all or the DFC's

2:15

Speaker 3: DSF's all-time number one financial contributor. So please do look at it. From your perspective, you get a 30% discount on a PyChan Pro license. From the Django Software Foundation's perspective, we get 100% of the revenue that goes into that goes into that. So JetBrains, they don't take a penny from this, they give it all to the DSF. So please do consider it. PyCharm is a wonderful editor or a wonderful IDE. It's very powerful. I think if you scroll down here it tells you all the new features. They've got and whatnot. But do at least give it a consideration because it is very important to the DSF. That's all I've got to say. I've got one more tablet. Oh no, that was something else. Don't worry about that. Come in September to Jankwantov. That's it. Thanks. Andrew.

2:58

Speaker 2: Thank you very much. Thank you. Next up we have Mr. Tim Bell again giving chance to uh let people speak. And do we have the slides? A V Tim. There we go. Okay, Mr. Tim Bell.

3:34

Speaker 4: Thank you very much. Um yes, I'm Tim. You might know me from Wednesday's talk on uh uh table partitioning with Postgres and Django. Today I'm adding new UIDs into the mix. There was this quote from my talk, partitioning often works best if the application is designed from the ground up with partitioning in mind. I didn't have an opportunity to demonstrate a case of that in my talk, so I'm going to do that now. So this is a case study of something that uh we're we're looking at it cracking. Um I discovered literally a couple of hours ago that this actually got merged ten days ago and I didn't know about it So we've got this new bookkeeping event record model. Yeah, there's some of it there. It uses event sourcing, which it means it's an append-only table. Why do we want to partition this?

4:19

Speaker 4: Well, we know it's going to get quite large over time. We expect it to get to 20 billion rows. So that's big enough that it's worth partitioning. The question then is what partition key to use? Almost all the access is by the ledger number, and sometimes we need to retrieve all the records for a given ledger number. So those are the key things we know we need to support. Should we partition by the created at field? That would put all the recently created objects in a partition together. But objects for a given ledger number would be in multiple partitions. Which means that second use case where we were retrieving all records for a given ledger number would access lots of partitions. We want to avoid that. What about partitioning by ledger number? All partitions would get written to, which is perfectly fine.

5:04

Speaker 4: But all the objects for a given ledger number would be in the same partition together Which means that if we did need to retrieve all the records, which we expect to, um it would be good. It would be just accessing that one partition. So I said at in the talk on Wednesday that partition key must be part of the primary key. If we partition by ledger number, that would be mean using a composite primary key. And they're not well supported in Django. So can we partition by ID? Just a reminder that the ID here is a UUID field. Let's do a detour into UUIDs. They're 128-bit long, usually displayed as hex strings like this. They come in different flavors, UUID version 1 through 8. I want to talk about two versions.

5:51

Speaker 4: Version 7 uses a Unix epoch timestamp as the first part of the most significant bits of the UID, then some random stuff. UUID V8 is entirely user-defined apart from a version field and variant field. These two versions were standardized in May 2024, so they're quite new. This is what a UUID V7 looks like. The top row uh is is bit numbers. Uh UNIX TSMS stands for Unix Timestamp Stamp in Miseconds, and that takes the first 48 bits of the of the um of the the the the whole thing. Uh then you've got a version number and then some random fields and there's a variant number as well. UUIDV8 , if I flick between the two, you can see that there it's the the the by the boundaries are similar.

6:36

Speaker 4: It's just got custom fields all over the place. So the first 48 bits are custom A, and then there's some custom B and custom C. If we partition by UID, uh V7 ID field, it would be like partitioning by created at. That would be by time. We don't really want that. If we partition by a UI V8 ID field, it depends on what data we go put into the custom A value. And we get to choose what that is. If we put a hash of the ledger number into the first 12 bits of the custom A field and then put timestamp in the next bit and the rest of the timestamp a bit further on That will let us effectively partition by our ledger number or hash of the ledger number.

7:25

Speaker 4: So what's that going to look like? We're going to do a range range partition over the ID. Our partition bounds for all our partitions will cover all possible values. There'll be a fixed number of partitions. We've chosen 256. It could be 16 or 4096. All the objects for the same ledger will be in one partition together, so we don't and we don't need to use a composite primary key. And if we search and order by ID. the values for a given ledger number will be sorted in created at order. And for uh event sourcing that's actually exactly what we want. What's that partitioning look like? Here's a little bit of a diagram. You'll see I've got a few of the partitions out of the 256 across the bottom. And the ID goes from 000

8:11

Speaker 4: dot dot dot dot dot. um and uh is in less than 010 dot dot dot da dot that's for the first partition and similarly for the other ones

8:20

Speaker 2: About 30 seconds, Tim.

8:21

Speaker 4: Oh. How do we query by ledger number and get partition pruning? You have to use a query set method like for ledger number here. Which I've described. And it basically works out what's the minimum and maximum UUIDs for a given ledger number. And if you do a query and look at the SQL you get out, you get in the where clause here. We 've got an ID in the where clause and what this all means that essentially for this query you get partition pruning and so we can uh evaluate this and touch just a single partition. And that's it

8:56

Speaker 2: Thank you very much. Next up we have Anzi And the

9:08

Speaker 5: it's Anjè Pichard.

9:11

Speaker 2: Sorry.

9:15

Speaker 5: Um I blame my mother for picking a name that nobody outside of Slovenia can actually pronounce. It's been the bane of my existence.

9:27

Speaker 2: Okay, maybe I should let you do this.

9:29

Speaker 5: Uh if you just press this button.

9:33

Speaker 2: Great. Okay, round of applause please.

9:36

Speaker 5: Okay. Thank you. So let's talk about speeding up your Django startup times using lazy imports. So as you might know Imports in Python can be slow. Why? Because when you run import something Python executes that something as code, which means that all the code that is in that module is going to be executed. And sometimes that just function definitions that you want to actually use, but other times that's there's a lot more going on, and that can cause some severe slowness.

10:22

Speaker 5: Uh so much so that one day you might wake up and running python manager. py check starts taking an awkwardly long amount of time. And that frustrates you so much that you spend a lot of time trying to figure out why is that so Luckily, uh Python has a built-in tool to figure out which imports are slow and you can invoke it with the minus x import time uh flag. And what this is going to do, it's going to print out all the time all the imports that happened during the execution of your program with times with uh that

11:07

Speaker 5: took uh how much each of them took. And I don't want to name shame any third-party packages here, uh but I think there is a pattern here Um now why exactly uh Google's um imports here were very slow is a separate talk, maybe s another time. But uh the way to fix it is to make sure not to import Google 's um Google's packages when your application is starting. Uh so this is a snippet from our code. Um this is uh what was e essentially causing the problem.

11:53

Speaker 5: You see the import st uh statement at the top. Now to fix this we had to of course make sure that the type annotations still work. So there's an if clause at the top And we had to move the actual import inside the function to make sure that it doesn't start or that it doesn't do all the extra work when the app is starting. And this gets very gross and disgusting, and you feel dirty when you do this because now you have all of these imports and all of your functions that need to use these packages Um really not nice. Uh so we got from thirteen seconds to

12:39

Speaker 5: three seconds by changing one hundred and thirty files and essentially adding more than almost six hundred lines of imports within our function. So not not nice at all Besides that, you now also have two problems. Uh once you moved all of those imports, you you now can very easily regress by just Putting an import back to the top. It might be you, it might be your coworker, it might be an LLM, doesn't matter. Your app start times are going to get slow again. Uh the solution for this is have configure your lint tool to make sure to complain

13:26

Speaker 5: to complain when this happens. And the second problem is if you're actually using those imports during the request response cycle You're going to cause a bad user experience because now if you're starting to load that slow module when the user clicked a button That user is going to have to wait two seconds for that module to actually load. Now to fix that, uh usually um the web server that you're using has a post worker in it or a similar method about where you can put

14:05

Speaker 2: one minute

14:06

Speaker 5: when you where you can put all of those imports in and they'll be loaded when your production server starts. So luckily this experience is going to get a lot better in the next version of Python, thanks to PEP 810. Which adds explicit lazy imports. So explicit lazy imports, what does this mean? We now have an explicit lazy keyword That we can put in front of our import statements, which makes the imports lazy , as we can see here. There's also an X flag to make all of your imports lazy.

14:55

Speaker 5: And you can also disable all lazy imports, so that's also super useful. And I don't know if I should be telling you this, but lazy imports also allow circular imports, so

15:06

Speaker 2: thank you very much Thank you. Be lazy. Next up we have I I'm I'm I'm I'm really sorry, I'm doing my best. I um Yeah, I'm not sure. Okay. She said it Okay Oh, but no, it's not possible to make it much there.

16:03

Speaker 2: I won't do this again tomorrow, I'm sorry. I mean I I'm I'm absolutely done with this now. Okay. Um I'm doing it. I'm doing it. Please don't don't look at me while I'm doing this, uh, because it actually makes me very nervous. I don't like being on stage. And uh I don't I I don't like any attention or anything like this. Um I'm sorry, I don't think I'll be able to get the notes to work. Yeah.

16:48

Speaker 2: So after after After Hojung, we have venelin. Is venelin here? Yeah? And and you don't have any slides, right? No, no slides. Excellent. Gives me a chance to sort it out. Thank you very much. Well have a round of applause.

17:08

Speaker 6: Thank you. Uh first before I start, how many of you uh know to solve a RubyCube? Good. Um I learned to solve uh Ruby cube last year. Uh late but uh still it was fun Uh it was not hard. Uh around let's see two weeks it was enough. And the during this process I was thinking about how we learn at all. And also some relations to the neural networks because we are you're a neural network. And in the neural networks you know that there are some stuff called tensors. And what is a tensor?

17:55

Speaker 6: Tensor is a uh we we we we think of of them like uh vectors, but vector is uh one representation of the of a tensor. And for example uh if we think of this cube we have this motion and this motion is the same Let's say if you do it like that. And the orientation of the coupe is different, but ultimately the The movement is the same, just reflects a different part of the coop. And what the tensor is is the knowledge about this movement that we know how to

18:41

Speaker 6: To move one piece from one uh space to the from one place to the other place and no matter how the cube is oriented it's the same moving. And In in our brain and also in the neural networks, uh we have a knowledge of things and how we apply it. Uh this knowledge depends on the context. And this is why the context is very important. uh we accumulate knowledge by uh combining uh different stuff together which are similar in a sense but a little different And we can learn new things by uh associate them with the knowledge that we already have.

19:27

Speaker 6: And this way we can learn faster Because we need to only learn the different part, not the the whole thing. Because it can look different, but it's the same. Um I'm Sorry. Yeah, uh with we need to practice a lot. How we store uh data in our brain is by repeating one thing many times

20:12

Speaker 6: And this way we uh we say to our brand that this thing is important. If we don't repeat stuff, they're not uh considered important and we never remember them. This is why we can't learn without repeating things. And in order to be better at our work we need to work hard To do one thing many times and after that it will be easy because there are no hard things All things are easy, some things we just don't know how to do them. Our brain functions in two speeds, slow and fast. There is a

20:58

Speaker 6: book called Thinking Slow and Fast. The author is Matridli, if I'm not mistaken. Uh no uh the David Kahneman, sorry, David Kahneman. And the LLMs work more like the fast way how we think. Um this is our intuition. And by doing things slowly we build our intuition To be better and to answer better because our intuition is always fast but not always right. Sometimes it's wrong And in order our intuition to be better, we need to practice more to teach it to be better.

21:50

Speaker 2: Thank you very much. Next up. Alex Gomez with a talk on Django or maybe not Django or why community is important. I assume it relates to Django, is that right? In some way? Maybe Oh yeah, no. No no no. You go attach, so

22:14

Speaker 7: Yes, I'm not a natural speaker. So during this talk I'll be looking at my phone mainly because that's where I have my notes and also because if I look at you people I'm going to be even more nervous. I also don't have slides because there's going to be tech issues anyway caused by you In forty-eight hours I'm going to start my next term as the treasure of Python Spain, which is the nonprofit whose objective is organizing PyCon Spain, as you may imagine. And this year it's in Barcelona in November, please come. But I'm not here I'm not here to talk about the conference. Like Being a treasurer is going to be a lot of work.

23:00

Speaker 7: Part of me doesn't really want to do it. Like I'm too tired to work. But I want to help. I want to make things better. So I'm going to do it. But what I'm here to talk about is like this conference is organized by hundred percent volunteers. Almost every Python or Django conference is organized entirely or almost entirely by volunteers. And this may be a surprise for some, but organizing a conference is, and I have to swerve here, a shit lot of work. And speaking from my experience as an organizer of Python Spain, every organizing team always needs more hands, always needs more people who want to dedicate their time to help. Going to conferences is great, but never take the organization for granted.

23:47

Speaker 7: For example, this year there is no PyCon UK because no one wanted to be the conference director. That's a community that needs help. So here is my ask to all of you. Get involved, contribute, volunteer. You don't have to start big and organize the next DjangoCon Europe or the next Python US or the next Europeon. Organise a meetup in your city Or if there is already a community, as did they need help finding skippers, finding venues, finding anything. And Or maybe even do a talk if you are brave enough, which I'm not. I look better behind the scenes, let's say. And let me finish with this. Volunteering in a community is a lot of work, but it's worth it.

24:33

Speaker 2: Thank you so much. Hi Young, round of applause, please. As I say, this man is significantly more competent than than me, which speaks to my competence, really. Look at that. Okay.

25:16

Speaker 8: Hi, I'm Huaung uh Hua Yang, uh Django developer and uh co-organizer of PyLady. So before I start, I have a quick question. Is there anyone else who traveled for more than fifteen hours to get to Athens? Oh yeah. But I think uh it looks like I might be the winner, but I didn't mind the long journey at all Traveled uh all this way because I really wanted to meet you all and more importantly to oh sorry share uh our

26:01

Speaker 8: community And yeah, in twenty twenty-three another organizer of our team, Susan, uh known as Luna. uh attended PyCon Korea and saw a shocking reality. There were only three women speakers in the entire lineup. This was a wake-up call. So in 2024 she started gathering people to build a community and that's how we revived Pilady Zoo, which had been inactive for a long time. So we wanted to create a space where women in Texas

26:47

Speaker 8: could connect, share their experiences, and grow together. Since then, we've grown in our team of seven organizers hosting monthly on offline workshops or meetups. And uh we are already seeing real change like take Doyen for example. She joined our Meetup, landed a remote job as the CTO of an overseas company, and is now an organizer herself. And also, although it hasn't been officially announced yet, I'm so ner honored to be sure that I've been on I've been named an outstanding pilot

27:33

Speaker 8: this year for my country. Also, through my experience as a Django nut, I learned that with the rice support, anyone can start contributing. Now I want to give the same courage to beginners in Korea who might be hesitating. That's why, as soon as I return home, I'll be leading a Django open source workshop. We want to help our members take their very first step into the open source world and show them that they belong here.

28:22

Speaker 8: Yeah. Community has no borders, whether it's Pilates, Athens, Pilates, Dublin, or Pilates Seoul, we are all part of the same mission to make the Python world more inclusive. So I strongly encourage you to join your local pilot chapters. Let's support each other and grow this global network together. Yeah.

28:51

Speaker 2: Thank you very much. What a great talk. Thank you so much.

28:55

Speaker 8: Yeah, and if you like to sponsor this or give us some advice or collaborate with us, please reach out. I'd love to connect with you all. Let's build a global breach together. That's it for me. Thank you.

29:13

Speaker 2: Next up we have T-Bot, I think. T-Bot? Yeah, I said it correct. I think that's uh one out of seven. Uh no, I two out of seven. I did say Carlton correct, I believe. Look, this is the last time. I've been banned from this now. I'm not doing it again. Just just forget about it. There we go. Flawless. Okay, round of applause. That was uh that was uh his fault.

29:52

Speaker 6: All right. Thanks.

29:55

Speaker 8: Okay, thanks. So yeah, I wanted to present a little tool I built uh which is called Django FMT. Uh it's an HTML uh Django template formatter that's written in Rust. So uh I started building this around a year ago because I was struggling with uh DJ Lynch, which is another template formatter. Uh it was breaking some of my templates, uh it was struggling with custom attributes with uh HT uh HTMX Alpine. And it was also very slow sometimes. And I also wanted to learn

30:23

Speaker 6: Rust and was uh following Ruff and UV development, so I took a lot of inspiration from uh the architecture So yeah, the heavy lifting is done by another library which is called the Markup FMT. It's a formatter for View, Angular, Ginjar, which is by Dino and Gprint. I helped to make it more stable for Jinja and I did what was missing so that it worked with Django. The project was also for me a way to do a lot of side quests, uh like uh running uh creating a playground for the uh for the formatter with uh wasm, desi and uh datastar in around uh 800 lines of code um doing also some micro benchmarking with code speed um an ecosystem check uh which is running on every pr

31:10

Speaker 6: so that I can see the the changes

31:12

Speaker 8: on uh real project open source project uh using Django.

31:17

Speaker 7: And also rather involve a release pipeline, but it's mostly automatic using cargo

31:21

Speaker 6: disks. Usage is quite simple. I suggest the pre-commit hook because it's uh great for the workflow, but also uh you can set up uh some uh unsafe integration uh in your ID

31:42

Speaker 8: And um

31:43

Speaker 6: yeah thanks. Um right

31:45

Speaker 2: the next step for the project would be for me to try to add some leading

31:48

Speaker 6: rules with uh autofixes. For now I'll just have the infrastructure to do that, but it's not uh uh no uh no fixes for for now. And I would also like to add some sort of type checking in the templates, uh but that's uh still a big work in progress. Um so yeah, thanks. Um yeah, my social if you want to re shout.

Conference

DjangoCon Europe

April 15 - 17, 2026

Athens, Greece

Note: We understand that names change, people change, and bodies change. We respect each individual's journey and privacy. If you have any concerns about a video or need us to remove content, please don't hesitate to contact us. We will handle your request with care and promptly address any issues.

More videos from DjangoCon Europe