text
stringlengths
0
1.49k
**Joe Doliner:** Yeah, not very successfully. I don't know what Go would be like if you just limited yourself to structs and simple functions and stuff.
**Brian Ketelsen:** It'd be like Pascal.
**Joe Doliner:** Yeah, there you go, it'd be like Pascal.
**Erik St. Martin:** Wow.
**Brian Ketelsen:** So what's the biggest use case you've seen yet for Pachyderm? What sort of sizes of data have you seen in the customer base?
**Joe Doliner:** We've seen sort of pushing up into the hundreds of gigabytes - that's about the biggest we've seen right now. That's definitely a lot smaller than where we wanna be eventually for a big data project.
What we've found though that's pretty interesting is that there are a lot of people who are in that middle ground of less than a terabyte, still don't wanna just have like a bunch of scripts running on AWS machines or a big box in their closet, because it's too unwieldy. So they want an orchestration layer to control a...
We're doing a really big push right now; basically, our users are forcing us to as they hit larger and larger use cases and things start to break. We're really pushing the envelope on how much data it can handle, and I suspect within maybe the next three to four months we're gonna be hitting the multiple terabyte range...
**Brian Ketelsen:** Nice. What are the constraints now for size of your data.
**Joe Doliner:** There's a bunch of them. One of the things we've hit a lot is just figuring out exactly where to put the data within the Docker containers, so that things don't crash. Kubernetes unfortunately right now doesn't have a way to ask for disk space as a quota. You can ask for memory and CPU quotas when you ...
\[19:48\] The other things that we've just hit are like - we have sort of two interesting axes that people scale along. One is they have really big files; they'll have a hundred gigabyte file that they throw in Pachyderm that they wanna process. Then we'll have people with small files but millions of them. With million...
So these are all the standard edges that you start to sand down as soon as you actually are putting the system through its paces and really starting to get these workloads working.
**Brian Ketelsen:** Do you see a use case for any distributed file systems in the future? I know there's been a lot of activity on that front lately too for very interesting distributed file systems, and a lot of them were written in Go, too.
**Joe Doliner:** Yeah, so one of the major pieces of Pachyderm is the Pachyderm File System, which is a distributed file system. That's the main feature that sets that apart from other distributed file systems, as that's the thing that's got all the version control logic inside of it. Yeah, I definitely see use cases f...
The Minio guys actually came in and got Pachyderm running on Minio, which is really exciting, because before then you ran Pachyderm on top of an object store, so this is really nice if you're deploying on a cloud platform, because you've got S3 or you've got Google Cloud storage or you've got Azure Blob storage. But if...
Minio is super smooth to set up. It's all written in Go, and we now have direct support for it in Pachyderm, so that's our preferred on-prem solution right now.
**Brian Ketelsen:** That's brilliant. Speaking on Minio, we love the Minio people. They are the best open source project when it comes to supporting the community. They support almost every Meetup I can think of; they've been hosting things like Women Who Go and they've supported GoTime FM, they've supported GopherTime...
**Joe Doliner:** Yeah. They're supporting us, and we're another open source project. Normally, you don't see that level of support for open source projects, but they just reached out and they did it for us. So yeah, those guys are really great.
**Brian Ketelsen:** That's really cool.
**Erik St. Martin:** I think it's about time for our first sponsored break, and after we come back, I'd love to talk to you -- you brought up a good point in our email exchange before the show about talking about open source projects, like running a large open source project and building a company around that, and I'd ...
**Break:** \[22:46\]
**Erik St. Martin:** And we are back. We're talking with Joe Doliner from Pachyderm. Before the break you had mentioned maintaining an open source project and the human side of it and how to stay open source but build a company surrounding that. I'd love feedback on that, because it's an interesting problem - how do yo...
**Joe Doliner:** \[23:45\] It's a very interesting problem, and the short answer is I think it's a very hard problem. There are some people when I tell them that our software is open source, they're just like, "Wow, so does that mean you're never gonna be able to make any money off of it?" That's not one hundred percen...
Starting at the beginning, to get an open source project that's gonna get any sort of traction, that's gonna get people interested in it, you have to align a set of incentives of people in the outside world. You have to a) make something that's gonna be useful to people. It has to solve a real problem for them, it has ...
After that, you have to start getting developer's interest aligned, and a lot of that goes into positioning it next to the things that are interesting... For us, we're positioned very close to Docker. We've been very close to Docker for a long time and it was just exploding at the time we started (it's still exploding ...
For a lot of people, we were just an interesting new thing that they could do with Docker. They'd been wanting to play around with Docker, they'd been wanting to hack on Docker... This was a project that appealed to a different set of people - it appealed to data scientists and more data-oriented people, so we became t...
You also need to navigate how easy it's going to be for your product to be deployed. There's always some cost to deploying a new product, and for some products this is just insurmountable. I think a good example of this is Urbit - have you guys heard of that?
**Brian Ketelsen:** Mm-hm.
**Joe Doliner:** Urbit is this very cool idea where they're building this cryptographically-secure, peer-to-peer server network where you can just get a server that you can run anything on, but it's really opaque how to use it; it's very hard to use. People do use it, and they know this is a problem, they're working a ...
We saw that people were switching to containers, people were switching to container orchestrators like Kubernetes, and in fact, Kubernetes really seemed like the cream of the crop in terms of container orchestrators... So we made the decision very early on to make our product just completely deploy on Kubernetes. What ...
The other very interesting aspect of this is how do you build a company around it, because one of the things that we've needed to do, we've gotten some developers of the open source community that have come in, but most of the people who do the heavy lifting of Pachyderm development are developers that we employ, and t...
For that, you need to have some way that you can eventually make money off of your product. I think for a lot of different kinds of open source projects, I just don't see any way that this can ever happen. If you're making an open source BitTorrent client for example, I don't see how anybody's ever gonna pay for that. ...
\[27:51\] For us, fortunately... Companies, when they invest in data infrastructure, it's a big investment. If they're running Pachyderm, they're probably gonna have 10, 20 engineers who are just using it every single day as a major part of their workflow. In those cases, companies are often very willing to pay for sup...
The other thing that can work in terms of a business model for open source is if you can turn it into some sort of a hosted model. The analogy isn't totally perfect, but GitHub is sort of a monetization strategy for Git, right? GitHub itself isn't open source, but you can see how that works.
We're planning to eventually build an equivalent for Pachyderm, which is tentatively called PachHub. You can imagine this will be a site much like GitHub, except instead of code repositories there will be data repositories, and there'll be pipelines that are processing those, and you can see what the entire community i...
**Brian Ketelsen:** One of the things that you mentioned early in that very long monolog - thank you very much - was how people get interested in open source projects and how you get engagement of the community and the developers. One of the things that it triggered me to realize is that I get less engagement out of th...
Even at the beginning of the project when it's that rough patch that you talked about, where things don't quite work as planned - or they don't work as you intend them to work in the long run - that engagement is driven by the way the project communicates to me where they want to go, what their vision is. I think that'...
**Joe Doliner:** Yeah, absolutely. You really need to sell this dream of the future world to people, and we've spent a long time iterating on that and changing how we describe our vision so that it really resonates with people. Right now, I feel like we've got something that's working pretty well for a large group of p...
**Brian Ketelsen:** But when you think about the open source projects that have been really successful in the past, many of them have charismatic leaders that make statements that are even sometimes controversial. DHH in the Rails community, Solomon Hykes in Docker - they both frequently make controversial statements, ...
I feel like part of the adoption curve in an open source project relates to the charisma of the leaders, too.
**Joe Doliner:** Yeah, absolutely. Invariably, that charisma and that personality trickles down into the project itself and into the community and how all of that works. There is a lot of value in being controversial. You shouldn't just be controversial for the sake of being controversial, but you should always be will...
**Brian Ketelsen:** Yeah, that really ties back into my statement earlier - you're selling a vision, and the aggressiveness and the means in which you sell that vision are what attract me to a project. I just realized that.
**Erik St. Martin:** Yeah, it's an interesting world, trying to balance how to stay profitable and how to keep giving back.
**Joe Doliner:** \[32:04\] Absolutely, and I was just gonna say, of course, I think the original example of the charismatic but controversial leader is Linus Torvalds, who has no qualms about taking a stand about anything he believes in, which I think is absolutely great.
**Erik St. Martin:** It's hard... There's days I wish I was born with whatever gene made him that way. \[laughter\]
**Brian Ketelsen:** I don't know, it's tough... A lot of those charismatic type leaders... I can't imagine myself not having the filters that they don't have.
**Erik St. Martin:** But you feel like you'd be happier that way, if you just said and did whatever you wanted, and didn't really think about the way it was perceived. You'd probably stress less about what you were doing or saying.
So in your email - we were talking about projects, other things in the Go world that you had some interest in, and you brought up... I think it's called Gitea. Is that how you pronounce it?
**Brian Ketelsen:** Gitea, yeah.
**Joe Doliner:** Yeah, which I guess is a \[unintelligible 00:33:02.26\] of git and tea like the beverage. Yeah, I was browsing around Go projects I want to do and saw that. It looked really cool to me just because it's like an open source GitHub, and I think one of the long-term untenable things in the open source wor...
There's also GitLab, and I'm actually friends with the founders there; I'd love to see them eventually upset the open source world. But even though I think that GitLab is great, I just can't not have Pachyderm on GitHub right now, because that just is where people are gonna look for open source projects, and it's kind ...