text stringlengths 0 2.29k |
|---|
**Lars Wikman:** So now you need monitoring for your monitoring. And that sort of loops forever. And that's just infinity servers. That's no good. |
**Gerhard Lazu:** Yeah. I'm still hung up on the CP thing that you mentioned, like how do you get those Erlang releases out there - just CP them. So it sounds very manual to me. |
**Lars Wikman:** Yes. And that's not a part I love about it. So I've been looking at different tools that might sort of fit the trade-offs I like... Because it's not -- like, I need to find a tool that can do this. There are infinitely tools that can do this. It's just like getting a file to a thing; you could do it wi... |
\[21:43\] Now, if you're on a dedicated server, how do you do a nice blue/green deploy, a rolling deploy? It gets a little bit more tricky then. And if you have two dedicated servers, like - okay, yeah, then you can do blue/green in sort of a traditional way. Something I want to explore is how to do a nice blue/green d... |
Another one, which is even sort of moving it one step further, is that I believe there are socket options you can use to share a socket... |
**Gerhard Lazu:** Yep. 'REUSEPORT'-- 'SO_REUSEPORT' that's the one, yeah. |
**Lars Wikman:** So the new one will simply start getting the traffic, and the old one can be faded back into the background. |
**Gerhard Lazu:** I think at that point you are writing your own orchestration layer in the app, right? Because that's what ends up happening. Like, how do you orchestrate a new release? And even before that, you still have to run your tests, you still have to get the dependencies, you still have to do a bunch of thing... |
**Lars Wikman:** Yeah, but that's on the build side, right? |
**Gerhard Lazu:** Right. |
**Lars Wikman:** So let's say you're using a CI, whatever CI you are, to do all these things; you end up with an artifact that's okay to get out to production. And then that's where the CP comes in. CP, as you say - okay, you could do that in the CI, to get it out there. You have a single host, so then you don't have t... |
So - okay, let's say the new version is running, and - what else needs to be aware of this new version being out? You may need to notify something. Again, before you know it, you have like a whole orchestration layer, which is split between your CI and whatever this thing is; some code in your app, for example. |
**Lars Wikman:** The whole notify something else is probably what I'd consider -- when you hit that point, like "Oh, but there are other services, and they need to be notified when this goes out", and yada-yada-yada. Then you are probably not a monolith anymore. And my approach definitely is aggressively monolithic. |
**Gerhard Lazu:** I think I think that's a good one. But again, a monolith - I think it's a good idea, and I can see a lot of premature optimizations; people going to microservices, people going to like even serverless... That in itself has like a whole load of things, operational concerns that people need to be aware ... |
\[26:19\] We have been successfully running a monolith for a long, long time. However, there are external systems that the monolith needs to interact with. Your monitoring, your logging, your exception system. The monitoring is both of itself, and an external system which monitors it. So do you notify there's a new dep... |
**Lars Wikman:** Yeah, I think that was one I was influenced by your conversation with Kelsey Hightower about, right? |
**Gerhard Lazu:** That's the one, yeah. |
**Lars Wikman:** You just want to bring up that you've spoken to Kelsey, that's what you want to do... \[laughs\] |
**Gerhard Lazu:** No, no, no, because I think... I'm trying to get to the readme, and there's something that you wrote, which I really liked; you wrote "Human-scale deployments." And I think that's a very good way of putting it. Because even though the system is complex, it's not crazy complex. A human, a normal human ... |
**Lars Wikman:** Yeah. That's generally what I aim for. Some people really, really get excited about trying to solve problems at scale. I really, really don't like what I see of systems at scale. All in all, it tends to be sort of a big challenge of making layers upon layers upon layers of people, and tech, and bureauc... |
For example, the post office system, or the power grid - there has to be large-scale coordination in place, and then there also needs to be a lot of smaller systems that play nicely within that larger one. But I'm not interested in solving like a 100-engineer or a 200-engineer problem, in general. I like small teams, I... |
And sometimes small teams can run large things. For example, WhatsApp is a pretty good example of that. Now, I bet they had a lot of orchestration going on, because they had to, because they were at an immense scale. But they also did a lot of things that are not commonly done. For example, \[unintelligible 00:29:29.23... |
\[30:12\] It's a little bit like -- I consider Elixir a competitive advantage for many companies, or a potential, at least, competitive advantage, compared to, for example, all the companies that run Java. You cannot win a competitive advantage by choosing Java, because that is not an outlier; it has no opinion, it is ... |
For example, Apple likes to ship half-finished features and services... "No, no. We just removed a lot of buttons, and it's so simple, and so straightforward." Yes, but you could also add some options, so it's more flexible. But they don't. And I think that's part of their plan for sort of shipping more things, even th... |
If you make decisions, and you sort of take chances and go in particular directions, I think that's where you can find interesting things. I don't have what I consider a complete plan for my operations. I don't have all the tools figured out that I would like to. I've got some recommendations for some nice tools for so... |
**Gerhard Lazu:** Would you use something like fly.io for your own stuff? |
**Lars Wikman:** Sure. I wrote a newsletter just recently about opposing ideas. I can find cloud deployments and sort of the whole cloud-native space interesting, but also be more attracted to bare metal, dedicated server, keep it as simple and lean as possible. And I can't do both. I can never do both. I can try both,... |
And similarly, if I'm launching a business venture, like if I'm building a product, there are different schools. It's sort of, "Oh, do you do the whole biggest camp thing, and like build a really thoughtful product, and you host it very carefully, and you run it in this particular way, and you design it very deeply, an... |
**Gerhard Lazu:** \[laughs\] Okay... |
**Lars Wikman:** And if I was going for "I want to launch fast", I would probably pick Fly right now, for that kind of launch, because they feel like pretty much the new Heroku in that regard. I don't think they have quite as polished a system as Heroku, but they also are a lot more featureful than Heroku was. So they ... |
**Gerhard Lazu:** \[34:11\] And if you were to explore, what would you pick? |
**Lars Wikman:** I have what I consider to be an art project I would like to try... And I think this one would be bare metal, but it could also be done in sort of an elastic-cloudy way. But that would be exploring Erlang hot code updates. I would like to build a system that has no persistent data store; as I mentioned,... |
**Gerhard Lazu:** Yeah... Reality has data. Data is a pain to manage. |
**Lars Wikman:** But the thing is, I want data in the system. I want the system to be incredibly stateful, and I want people to join, and contribute, and everyone just have to deal with the fact that there is data in the system, and that mistakes cost. Content, data... It would be an interesting way of building somethi... |
**Gerhard Lazu:** Yeah. |
**Lars Wikman:** But of course, it's a very interesting thing. |
**Gerhard Lazu:** You just need a lot of discipline. You just need a lot of discipline. You have to write those transformations. "How do I go from this state to the new state?" And it's not just putting the new thing out there, it's like the transformations of whatever is running, it needs to migrate. The function call... |
**Lars Wikman:** Yeah. You've done a decent amount of that... |
**Gerhard Lazu:** Actually, no. |
**Lars Wikman:** Okay... |
**Gerhard Lazu:** But I worked with someone that did. I think this was episode nine, Jean-Sébastien Pedron, we talked about release engineering, and I think in that context he shared - maybe; I don't remember whether we recorded this, but he was saying how he used to work on a team where they did use Erlang, and they d... |
**Lars Wikman:** This sort of pins down some of the reasons I don't want to go into the whole Kubernetes land, and also why I'm probably never going to be doing hot code updates for a real project. Usually, you can just keep it simple, a lot simpler than sort of the recommended practices, perhaps; or recommended at sca... |
\[38:07\] Most applications I've built have had times in the day when no one is using them, because people go to sleep... And they've been national; so they've been limited to one country. And it's like "Okay, yeah, maybe we call some downtime for some person that's currently in Thailand and wanted to check something."... |
And I think one of the reasons why we generally do sort of a high-availability approach, and like blue/green deploys, and all that, is that that is near the level of comfortable trade-off. It's not that hard to keep the system up while performing an update, if sort of all of your state is source of truth in a single da... |
**Gerhard Lazu:** Yeah. Yeah. I know for me - and you've seen the first episode of this year, where we've been talking about Talos OS and the experiment which I'm running with it, is I'm choosing a different starting point. And I have been doing packages enough, and updating packages, and things getting messed up, and ... |
And then there's also the contributing MD, which we've added things around how to set up everything locally, so that you can do development. No automation, like no Docker, nothing like that. Just the plain description of what the components are... "This is how you would install them manually, on a Mac. And by the way, ... |
So going back to that, do you have something similar that describes how your systems are set up manually? What are the components? How do they interact? What to do, where...? |
**Lars Wikman:** My last efforts towards something like that was when I was trying to sort of "This is how I want to set up all of them." And then the idea was mostly like "Oh, Bash scripts are pretty close to just the documentation itself." |
**Gerhard Lazu:** That's what I thought of makefiles, by the way. I changed my mind... \[laughs\] |
**Lars Wikman:** Someone came on the show and changed your mind. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.