Transcript: Django on the Med - Paolo Melchiorre

More in their services later in the show.

Hi, welcome to another episode of Django Chat.

I'm Will Vincent here with Carlton Gibson.

Hey Carlton.

And we're very pleased to have on our friend Paolo Melchiori.

Welcome, Paolo.

How did you join us?

Let me give let me give you a quick bio for those who don't know, so you don't have to.

But you are Django board member, Python Software Foundation fellow, prolific contributor to the Django code base, the Django website, prolific selfie taker at conferences where you speak.

Yes.

And you both just got back from Django on the Med, which is the topic of discussion for today.

So I'm super happy about this episode because I hope I won't have to speak as much.

But let me start with a question, which is how did Django on the Med come about?

Because it was originally in your head, Paolo.

For the first time, we had this great uh conference in San Diego after a long period of non-in-person uh conference for the pandemic.

So it was great to meet you all in for the first time in Django Kong US for me.

But at the same time, it was a very um uh a long trip for me from Pescara to San Diego and uh very tiring, tiring week.

So when we met the first day of sprint with uh I remember uh Simon, I remember Marius, uh you in the beach in the morning to have a few times for us.

I realized that we needed more to this um time as a team, as a small group of people, to hang out together, to network a bit, to chat about our life, and at the same time to sprint.

And based on the history I found in the in the past in other Python community to organize small sprints to have half time on working on something specific and half time into uh being together.

I uh I remember I proposed this idea, and from that moment I tried to force every every opportunity.

Uh yeah, and at the end Carton did uh uh did a good move and uh forced me in some way to to transform this idea in reality with the Zelda, of course, and uh other people.

So it took three years, but at the end we we did it.

And so DjangoCon Europe this year, right, and at the sprints in Dublin, we we we we sat down and schemed it, you posted somewhere, you posted a picture, we some notebook I picked up at the conference where I was making notes, we sort of signed in blood that we would uh we would do the do the event in Palpaget.

So if other people will join, uh good.

Otherwise, uh we spent together a few few days and enjoy the time together, but at the end there were more than two of us.

Uh yeah, yeah.

We sat in a cafe, the two of us.

We could have got house, you know, just one house group, but then you know, by the time we got 14, we had like three tables was right that were sprinting together.

So we will link to that, but it was well documented.

Yeah.

So I tried to aggregate all these posts together and uh with some automatic script to create the blog post.

And there's and just while we're showing stuff, I'll just show it all.

So this was the website that went up um about the event and then the forum to subscribe.

We've got a value map of some of the things that were achieved, um, or not, you know, so that we can demonstrate the value of this.

I mean, so I'll let me phrase it as a question, Power.

For me, um, we've we've really demonstrated the value of it.

Some amazing things were were done in the short amid of time by the small group of people.

And I think now we can go from look, isn't this a nice idea to look, and we did it and it's it's valuable, and can we now get behind it as a community?

There's we'll also put it in the notes, but there were five different PRs opened, one one closed, right?

Yeah.

So wasn't it wasn't all just drinking in the sun.

Um more than that, uh, some people contributed in some documentation.

I was working in a script to create the the metrics of um the feature flank for the URM.

So we are are putting together at a point that I think we'll share the next uh weeks.

Uh so not only the issue on on GitHub, but we discussed more things than that.

Uh, we uh created more value than what you can only see in GitHub.

And the the the the better part I think is the connection that we created each other.

So we we came back to having more connection because new people show up and people that you uh met only on online for a long time, you spent quality time together and was very, very interesting to do.

I was I was actually on the Met as well in Cyprus for a company event.

But I was following it every day.

And actually we can mention um so Mark Smith, one of the people who was there, uh just joined JetBrains.

Pipe I actually I don't know if it's I don't think PyCharm, but so now he's a colleague of mine.

So yeah, so

Down for I'm I'm I'm open to it.

Yeah, we were we were in Paphos, the so the entire PyCharm team.

So kind of same thing you were saying, like a lot of these people I work with, but I hadn't met in person, and there's a similar mix of you know, some fun stuff, some work, but nothing replaces in person.

So it's important to have

We had breakfast, coffee, croissants, you know, stuff to get us going, and then till two um o'clock when we broke for lunch, and then people had the afternoon to explore.

They could, you know, walk in the woods or go to the beach, walk along the Cami de Ronda, which is the coastal paths we have here.

Um I think people seem to like it, no?

So I uh I was able to go uh swim in the in the sea, which is was great.

Uh I didn't do a long swim, but I saw a lot of the coast, uh the fish.

Uh, other people like Marius went for uh multiple runs around.

Uh people have a drink, people enjoy the beach or the uh the Parf Foucher in the town.

So everyone did what they wanted to do to relax and to compress after to work in the morning.

There was someone, I think it was on Fastodon.

Some um Marius documented his runs and one of them was with Simon, but someone called him the Paolo of Runs at Django events.

So that's high phrase.

I I saw Eric poster that you know Marius is the official selfie run.

But I want to ask so Django on the med, the idea was that this is what twice a year, right?

So it's been in Spain, but then it'll also be in Pescara.

Is that still we're gonna move it to Pescara next year?

Yeah, yeah.

Yeah.

No, brought it for me as a as a gift for this time.

No, I'm I'm joking, yes, this this is the plan to organize uh twice a year and uh in the in a period that is not um uh high season.

So this year uh the weather has been married in Palatero.

We were able to run, to swim and sea, but it was not so crowded.

Um at the same time it was sunny, and but uh the prices of the everything was not high, the flights and the hotel rooms.

And we want to try to do that the same things in Pescara.

More or less the weather uh is similar.

We are still Mediterranean, specifically in the Adriatic Sea.

And um the original idea was to organize in April, but we see if we organize there in the beginning of May we will let you update on that.

But the the goal is the same to organize in an area not so touristic but still a great area to explore in a period that is not so high season to let everyone not spend a lot of money to to reach us and to to be there together and try to explore different areas in the Mediterranean.

But we are open also to other other locations like Cyprus or other

The theme is very simple.

I think it even you said you know Plone had these events that has I think you even asked me has Django ever had them and I don't think so.

Is that was that the initial what or one of the inspirations is

still with good weather and the sea.

But they also organize a one uh another one in in in Iastbrook uh during the winter and they go skiing after um yeah but other others community organize very similar events like that and then there is also the the core python developer that's organized.

Usually see the last time they organized I think in in September and in Cambridge yeah it was at the army of this exactly exactly so they do some very similar things and I asked a lot of people but nobody remembers something like that.

So of course we had sprints after conference but these type of sprints uh are something different.

And people sometimes come for like stay for one day but not two um does that match your experiences?

I know Carlton you usually were you were fellowing so you were running around uh helping everyone

Django on the med we had um I think five people who were you know new contributors they I mean they had some awareness they had some they weren't totally new contributors but they they you know self-identified as new contributors and we managed to get all of them going on you know good projects which was just super um you know including um an 18 year old ticket on the cash framework which I would dearly love to see revolve as a result there's an there's an open PR for that which we hope to get in um and

some things about speeding up um yeah that one that one there at the top there

And these are from new contributors right these are from people that haven't aren't established that aren't you know known within the community already.

And so that was phenomenal for me.

But the yeah I think you're right the end of a Django con it's often you stay for the first day or and then people start flying out it's not enough time to sit down and get a dedicated piece of of work done.

Whereas in three days we had you know really good amount of time I mean you know Marish has taken on um into implementing um native DB native cascade.

So when you've got on del the on-delete clause instead of having it done in in Python so it collects all the objects and walks the tree backwards you'll be able to um use the database constraints there which for years has been one of those performance gotches when you start scaling up your project and you you know you have to delete a whole tree of objects and you've got thousands of objects being deleted all in memory that's going to be a massive optimization for a lot of people and Marish said quite openly there's no way he could have got that done without being at Djang

o on the med where you know Simon Charette was there, Liddy was there, um Jacob Walls, the new fellow was there and those the four of them together working on the ORM you know they're all ORM sort of um specialists and to have the four of them in the room working together.

great to speak uh with everyone I I share every in every talk this image with me Marcus and McCaml uh working my first um issue then I realized after uh conference by conference how different can be the same event the sprint because when I went in other sprints when I was supposed to deliver a talk I felt myself at the end of the

the conference very tired and I tried to join the the the sprints I tried to share what I learned but I realized also that uh I already had a few energy to share with the with other newcomers and and the worst is when I help organize in some conference uh I usually try to help organize Python Italian other conference in the case it's less energy you have at the at the end of the so it depends from what role you have in the conf

erence having the springs at the end can be something that you can enjoy or you can put a lot of energy or not.

So the idea for this type of event is to uh don't have to have a lot of responsibility uh create a talk or other things maintain the the event the the less structured as possible so everyone can can enjoy Carlton did a lot of job finding for breakfast and other things so he worked it there but I I I hope it was not too much for him to to do and he was able to enjoy the the three days

But yeah I mean so I was working on what I've I've nicknamed the Fast Django project.

You know it's just a it's just a bench you know about benchmarking Django.

It came up in the roadmap so in the first morning we had it held a road mapping session where we you know we went through various things and we'll do a write up of that and the findings that we had but it was a really great session and the the the conversations again in the room which just don't work the same on the forum or on the you know on the GitHub issues they just people there's an understanding that you can reach between people that sometimes gets missed in those those written online environments.

But coming out of that there was this talk about you know can we benchmark Django can we profile it can we improve the poems okay so I didn't manage to get much actual coding done for that but I did manage to get a Google Doc worked out with an outline of things that I'm then come back to and work on.

And more I can open up to you know contributors I can say look here's an outline of work that needs to be done.

Can we, you know, anyone fancy putting this bit together or anyone fancy putting that bit together?

And hopefully then we can get some um some speed momentum for that.

But I did have time, I did have just enough time and energy to get that.

But then I'm I'm blanking on the name of it, but the um um that's the person you worked with on on Template Google Somewhere code, he's he's got a project.

Maybe he has two projects.

So one he he did a thing called um, I think it's called Django Rapid, which is um just a kind of example of a REST view layer um using what's called message spec serializers.

And then he profiled that against Fast API and showed that, you know, within you know, you're getting comparable performance already with Django using modern serialization techniques for compared to say Fast API.

And that's a nice benchmark because we always hear this story that Django's slow, Django's slow, Django's slow.

No, Django's not slow.

If you, you know, a lot of the benchmarks are comparing um say async to sync, and they're not running enough workers to really exercise the CPUs on sync because you've got blocking operation.

So, you know, you that thread is just sat there idle, so you have multiple workers in order to get good throughput on a sync deployment and things like that, and the problems like that.

And also, you know, a lot of this is from a few years ago.

Um, that's Django Bolt, the second one.

Lots of it from a few years ago where Django wasn't as fast.

And what we need to show is look, this is what Django was doing, this is what it's doing now, this is what's you're capable of.

That Django Bolt project there is a whole nother exciting thing again.

So that is using um a Rust front end in front of um in front of Django.

So the the why is why has Python historically gone slow?

Well, you've got this gill, and the whole excitement, one of the exciting things about Python 3.14 coming out is the free-threaded Python core, try and say that with a Lisp.

Um it which will remove the the gill so that you probably have to have proper concurrency with threads.

Whereas previously we haven't had that.

Now, what um the Django Bolt project there is using Rust threads, which are proper OS threads not bound by the gill, to run each um separate um Python ham handler.

And he's getting results like what does he say, they're achieving 60,000 plus requests per second, um, which is just lightning fast.

Like, I mean, that's

And there's no reason why they couldn't do the same trick, right?

But what you're what it's what do I think?

It's kind of a a trick, right?

Rust is much faster than Python.

Why is Rust much faster than Python 1A?

It's lower level code, it's it's compiled down code, but also you haven't got these guild problems and whatnot.

Well, if you take if you use Rust to take the guild problem out of it, you can make Python go quickly.

You've always been able to do that.

But what Fahan is doing is demonstrating this.

And hopefully with that and with some nice benchmarks and some blog posts and and other such things, put pulling all of these, these, all of this work together, we can help change them this narrative that we have in the community that somehow Django slow when really it's not.

From custom software development to consulting, team augmentation, or opening an office in Bulgaria, they're ready to take your project to the next level.

Right.

I mean, benchmarks, right?

If it's not measured, sort of doesn't count in a way.

I mean, there's the same thing going on with LLMs, right?

Like with what is we we were discussing with Jeff Triplett.

I think that episode will comes out after this, but you know, 42% or something of the SWE verified is just Django tests.

And so, yeah, it's like we all love a number, but it's like, where did that number come from?

Oh, yeah, where where this what it comes out as is lightning fast and more than fast enough for any service level requirement you might actually have in the real world.

Um uh I remember um a very great talk of Antonio Puni, a core Python developer at EuroPython, he was demonstrating uh how Python is slow for certain things, but uh in a way that it it doesn't matter, for example, in the web context, because when you have to interact with other services, uh services like your cache, your database, uh the things that matter is not how the language is fast or not, because more of the

time you were waiting for others to give you content.

Uh the great things we saw during the jungle mat was to see great uh guru of the URM like Simon and Maris working on trying to find a way to improve.

Uh Maris was working to improve and speed up the delete uh phase on directly in the DB.

So the more we work on that on that uh topic, I think the more the world system will be uh faster.

And another interesting thing is the asynchronous connector in the DB.

There is another great pull request that is still waiting for to be completed and merged.

And the more we know about this topic, the more we can improve.

I what I tried to explain in an article I wrote just before the jungle undermates to explain.

I was gonna mention it.

Here it is.

Yeah.

The more feature we have for all the database, the better will be the experience for everyone with Django and the feeling that Django Corsair is not is not slow.

So that the language is not something that impact so much in the experience in Django.

What is your service level requirement?

How what is the throughput you need?

Most, you know, most websites, if it if it did a hundred requests a minute, that would be like amazing.

But you can do thousands of requests a minute without even breaking a sweat on the smallest of instance, and yet people are like, oh, I can't use Django, it's slow.

It's well, how fast does your service really need to go?

And it it doesn't.

Still, Paolo?

It's a disclaimer.

But you had within a couple of hours, you had you know proof of concept.

So uh and also we found a way to improve this documentation, but we the proof of content basically being made the first the first morning together with Jaco.

Yeah.

Yeah.

I was we did wish.

I um I did want to call out you did have at least one sponsor, right?

Um Python Espana.

So Python Espana very nicely sponsored the breakfast.

So thank you, Python Espana.

And the DSF um gave us some uh sponsorship as well.

So thank you, DSF, and also but uh uh button down um sponsored the newsletter, which you know it was important.

It all counts, and it I it was something that if I hadn't had that, I'd have been like, well, how am I gonna do this?

So thank thank you again, button down there.

And as well, the Junt Amenda Pala Vugel, um, which are the council in Palafugel, they provided the venue and you know, lots of help there.

So um, you know, I

wanted to thank them.

So should we should we switch to books, other things we wanted to mention?

What else?

Are you Paolo before we switch to Yes?

He uh was very, very effective in helping us uh learn how to track everything, uh to provide uh the value of what we did.

Sometimes we want to solve uh an issue and then we start thinking to the next one because it's what we we do, what we like.

And Yelling did a great job in helping us to track everything and maybe we'll share uh something after after the event.

It's a lot of knowledge on this feed, so it's been very, very effective in doing that.

Uh, I think also Carton appreciates that.

It, you know, there was this line on the on the Python blog about the core developer sprints there that the the sprints are the best bit of a conference, and the great thing about the core developer sprints was well, they're they're sprints without the conference bit.

Well, same on Django on the Med, right?

It's where it's where the work's going to get done ultimately.

So um to be able to demonstrate that is great.

And to for Danielli to bring his expertise there is something that we just don't have.

You know, it's a tool set of tools that we we're not equipped with otherwise.

I mean, we'll feature it in the Django news newsletter, but that'll be the announcement.

But um yes, basically keep your eyes peeled.

We will be pushing it forward for May next year in Pescara.

Yeah.

I think we're at the book section of the show.

Carlton, do you want to go first?

So I'm gonna um say I've been reading this book, which is How Infrastructure Works by Deb Deb uh Chaklara.

Deb Chakra.

Um it's a wonderful book.

It talks about um, you know, water supplies and other such infrastructure things, but how we f how they form systems and how we fit into those systems.

And it's kind of perspective changing, you know, it's like everyone likes to think that they're this freestanding thing, but we're very much not.

We're very much nodes in a in a much bigger, um, a much bigger thing.

And I find I I'm really enjoying it, I recommend.

How how do these things cross your attention?

And then I Googled it, and then I was like, okay, wow, that looks great.

And then I read I'm, you know, I I ordered it and it arrived, and it's like, oh yes, this is this is a lovely perspective-changing way of looking at it.

Um she's some kind of um engineer, or maybe she does a teaching, teaching it, but you know, the way she places the human being in the built in their built environment, in their their social environment, not so much social, but in their physical environment, the built physical environment that we depend on so much.

Um, and is able to talk through the interconnectedness of it all.

It's it's eye-opening, really, I think.

Yeah, she's a professor.

Uh actually here you're near Boston.

Yeah, she talks a lot about Boston actually, and the water supply one.

What not?

Um Paolo, did you have one you wanted to mention?

I I have more time this time these days, so I'm trying to reread something I already read.

This was not planned.

I'm working on the update.

So he's showing he's showing Django for professionals, but yeah, I'm working on the update.

Well, yeah, I I mean Carlton, you've I mean, Carlton, Paolo, you've long talked about a book you might want to do.

Is there anything you want to say around that, right?

Is that can we say that publicly?

A book that you might write since you've written a lot around search databases.

So it's something that I I try to write down and to have a lot of uh a list of things to do or things we I experience and maybe uh it's something I can share in the future if I will have uh enough time to uh be there and write a write down on the things, which can say I think I'm thinking should.

You you have to you need a chunk of time to commit to it because you can't just 20 minutes here, 30 minutes there.

I mean, maybe maybe some people can, but you really need ideally a couple hours, ideally a couple days, because just you know, there isn't a one-to-one time to words on the page uh process.

Let like code, right?

You know, it all counts, but yeah, yeah, yeah.

You need these blocks, blocks of focus.

Well, I would I I would read it, of course, if it comes out.

Thanks.

Um and so for me, I've got this um more, everything forever, yet another book about AI, overlords, space empires, and uh Silicon Valley control.

Um I still haven't quite gotten enough of this.

I know you're a bit bored of the topic, Carlton, but I'm finding new ways to be enraged by uh uh what's happening.

So I'm

So uh I'm starting to see some repeated repeats and patterns.

Um yeah.

It all comes down to people.

And so if you just like in any organization, I find if you you look at the background of the top person in an organization and you can tell the the DNA of it.

So if it's a salesperson, it's a sales B2B organization.

If it's an engineer, like JetBrain's PyCharm, that's a different thing.

Both have strengths and weaknesses, but it all, yeah.

People, people are endlessly interesting.

I think that's what it is for me, Carlton.

The tech, you know, but the people and the motivations.

And also I'm and I'm sorry, I'm forgetting who someone at um maybe it was it Mario, someone at DjangoCon US mentioned this book in particular in one of the talks.

Yeah.

I was like, other one.

And you know, these ones are fun too, because I can read them so fast.

You know, they're they're not, it's not deep literature, you know.

I can blast through it in a couple sessions and pick something up, take some notes, and then yeah.

Um, what else?

I've got a project that I want to do.

I can put the link to now.

So we've got the Gzip middleware, right?

But there are much better um compression algorithms now.

There's um Z standard, which is now in um the Python 3.4 um standard library, Brotly G and Gzip as well.

Um and if you just swap and use one of the modern ones, you're gonna get better compression and better performance.

And it's it's just not great.

It's like, yeah, this was this is wonderful.

And you can configure this at say the Nginx level or whatever your web server is, but a lot of times your Django project is sat on a generic um platform as a service, and you want to use one of these middlewares yourself.

I think this is a super project.

It wraps it up, it demonstrates it, gives the pro ship concept, and I think this should be on the you know the path for to be includ for inclusion on Django in Django Core.

It's a precisely the kind of thing that Django to sort of package itself.

It's not, you know, it's generally useful, it's something that we can recommend to everybody.

Um sort of thing you don't really want to be implementing yourself.

Yeah, that's so battery.

I'd like I'd like to see this gain some uh gain some pickup.

And you know, if it's once the bugs have been worked out, there's no reason why this can't be, you know, added into Django might.

Well, I'll I'll mention one because Paolo, I don't know if we mentioned the the projects part to you, so I'll I'll give one, uh, which is based Pyrite, which one of my colleagues at PyCharm is a co-maintainer on.

So this is the Pyrite fork with type checking.

And so type checking in generally, it generally is an area I'm I need to learn a bit more about it.

You know, obviously I've done basic stuff, but there's all these new things coming out.

So Astral has TY.

There's was it Pyrefly that face I'm saying that wrong that that that Facebook Meta just put out.

Um typing, typing is big.

Um, and specifically on PyCharm, we've long had our own built-in type checker.

Um, and we're, you know, we hired Morgan, who's great, to figure out what the next uh generation looks like.

So yeah, I'm playing around with that.

Sorry, guys.

I say like once upon a time there was just um you know my pie, and then there was the Microsoft one, which it was good, but then it was closed source and didn't have plugins.

So, you know, all it's not or was it it or like maybe the type checker bit was open source, but then the the VS Code extension was closed source, and so it's a bit like well,

There's there's sort of a history of open source things and then Microsoft hiring people, and then

I mean, Python tooling in general, if we can lump that in, there's a lot of change.

Uh so yeah, it's an exciting time.

I mean, no one's quite sure.

I mean, so the again, Morgan, my colleague, he's got opinions and he's tasked with sorting out exactly what the next generation looks like within PyCharm.

Maybe we uh think I worked on that the with an article, maybe, was the new uh release for uh uh Python 314 that added the support for the UAD version 7.

Also that was being released a few days ago.

Uh also Postgres uh 18 released uh the last month, added this new feature.

So uh in theory, it's already been able, we have already been able to use that directly in in Django and we want to experiment on that.

We had a chapter about that with Simon and Lily.

Uh Django on the mat.

It's been interesting to see that Django can be already uh can be already used this new new feature that are very interesting in the the context of the database.

So I try to experiment on that and share what what I find.

If you can.

Yeah.

Awesome.

It's been uh a year of consultancy and I want to find something, something new, of course.

Uh if I can working with with Python and Django, the things I more uh experienced ahead, also with with team.

And if someone uh is it searching for uh an engineer, uh I'm hoping to super engineer.

Yeah.

And yes, and uh other than that, uh we'll start working on the next edition of Django, Django on the Mat for the sprints.

Yeah, we'll we'll see.

Yeah.

If anyone doesn't already know how to contact you, uh LinkedIn is the best way.

So we'll put a link to your LinkedIn profile.

But um I would hire you if I did.

So yeah.

So it's still just me and Mark.

We're still going, you know.

I I want to get us to the point where we've got a little pool to hire people because every six months or so somebody, you know, some superstar from the Django Sys ecosystem is on, you know, comes up and you're like, oh, I need to grab them, and I can't.

It keeps slipping through my fingers.

Um, I'll just end with I this is even though I wasn't there, I I was there three years ago when you first were mentioning this Palo, and I think I've been there through many of the steps, and so I'm very uh pleased that the two of you took the initiative and did the hard work to make it happen because it needed to happen, right?

I think both for the Django community, but also like the sprints weren't enough.

And so hopefully this sets a precedence that can be repeated.

Like it, you know, there are things that we weren't able to do because it was the first time and it was just the two of us and what okay, those things are totally fine.

even more.

And in other content wasn't able to seeing that people uh uh get something useful from that and they join was very, very great.

And I hope that in the future we'll be more initiative like that.

Not only maybe Django Drainman can organize also in that place in the world.

So yeah, Django on the blank.

That would be the goal.

Thank you for coming on.

Thank you for sharing this again.

I wanted I wanted to hear from Carlton and you how it went, and hopefully this shares it, shares the story with other people as well who weren't able to attend.

Thanks for having me to be able to share this this yeah, which which you all has been incredible to uh did after a long time we spoke for the first time together, and now we are talking in at past.

Yeah, yes.

Did you have something to say?

So we're at DjangoChat.com.

Thanks everyone for listening, and we'll see you next time.

Bye bye.

Bye bye.

Bye.

Learn more about their services in the link in the description.