text stringlengths 0 2.35k |
|---|
**Bartlomiej Płotka:** So there are many, many different methods of how you can process this information about healthiness. |
**Björn Rabenstein:** \[07:51\] I mean, what we got with the age of cloud or the age of distributed systems was also a notion of that just being up or down - it's not that easy anymore, right? Your server consists of many microservices, and every microservice has many instances... Some of them are always down, because ... |
I mean, this is what I call the founding myth of Prometheus. I have to talk about Prometheus, of course... I don't know if it's literally true. I was joining the Prometheus team very early in its history, but not from the beginning, right? So I also got this just from stories. So this is why it's all myth in the distan... |
**Mat Ryer:** Would you assume that was the Java one, on Java Time? |
**Björn Rabenstein:** I don't know. I think I will never be on Java Time if there's anything like that. \[laughter\] We'll see. So this was the initial spark of inspiration. There was an instrumentation library where you could instrument for metrics and it would expose in a separate endpoint - I think Bartek mentioned ... |
**Mat Ryer:** Wow. |
**Björn Rabenstein:** It's one of the most used Go packages in the universe, I think. |
**Mat Ryer:** Wow, in the universe. Wow. I mean, that's assuming Go only exists on this planet. We can't say for sure that-- |
**Johnny Boursiquot:** Or just in this universe. Because there is a multiverse, from what the movies tell me. |
**Mat Ryer:** Yeah. Still not bad though if you're the best Go package in this particular universe, even if-- |
**Johnny Boursiquot:** In this particular-- yeah. |
**Mat Ryer:** Yeah, even if there's more stars in other universes. |
**Björn Rabenstein:** Is there Go on Mars? There must be Go on Mars, right? |
**Mat Ryer:** Yeah, there must be by now. Yeah. That's great. |
**Björn Rabenstein:** And then there might be a Client Golang -- a Prometheus Golang might be on Mars as well. Who knows? |
**Mat Ryer:** Probably. |
**Björn Rabenstein:** It's running in all the German -- like, on the rail stations, they use it to monitor those. |
**Mat Ryer:** Do they? |
**Björn Rabenstein:** So it's running everywhere around me. It's pretty cool. |
**Mat Ryer:** Well, if only you'd instrumented it and it was reporting that somewhere, so you could see all the places it was running, that'd be awesome. Actually, I would like that as a feature, please, because for -- yeah, all packages, we ought to have that, so we can see how it's been used. |
**Björn Rabenstein:** The popularity of this library is actually -- for me, it was a huge surprise. And that's also, I think, an important topic about instrumentation. Back then - we're talking about the year 2012 - it was very uncommon for normal developers to even think about instrumentation. I mean, if they really t... |
\[11:48\] So they kind of did instrumentation, but they would never think about instrumentation for monitoring. This was completely an ops concern, and developers would never think about ops concerns. And teaching them that they have to instrument their code for things like monitoring was a big deal, right? I didn't ex... |
Also, developers might have pretty quickly realized that if you instrument your code for even more things, like metrics or tracing or profiles, that it even helps during development. That's also an important thing, that monitoring per se is everywhere in the stack, right? It's not just in the end when you add it as an ... |
**Johnny Boursiquot:** Different people have different needs, right? So when we talk about the operation -- sorry, if your organization is large enough to have a dedicated ops team, or the people who care about the VMs, and basically making sure that the CPU is not spiking, or there's enough memory, or whatever it is; ... |
But then I think there's an even higher level to that, where you instrument things and you can tell whoever asks that, "Okay, the service is up." I can hit the health endpoint or the readiness endpoint... Hopefully, it's not just a ping to get a TCP response, but maybe it's something you do and something meaningful tha... |
But beyond that, you're going to have the - can you answer "Is this thing doing what the business wants out of it?" So you're not just saying, "Is this thing up?" If your thing is up, but maybe you're failing process, 1 out of your 100 credit card transactions, and you can't explain why, at scale, that's millions of do... |
I saw people who use Nagios as a tool for instrumentation. Well, I'm not going to throw my developers into a Nagios dashboard and say, "Hey, go answer me some questions about business KPIs or something", right? But yeah, I'm going to use different tools. But I think the observability, really - I think it's an all-encom... |
**Bartlomiej Płotka:** Yeah. And I would like to kind of add something to your points, both Björn and Johnny, about instrumentation. I think it's very underestimated how much work it takes to really build a solid instrumentation library... Because the amount of work you put, Björn, in this Client Golang that you were s... |
\[16:15\] So in Client Golang you are able to essentially expose really interesting information about your Golang process, and around garbage collection routine, and memory, and essentially heap allocations, like how many tiny objects and big objects you are kind of allocating; very low-level information. So we kind of... |
So the amount of thoughts that had to go through those instrumentation libraries is enormous. It's not only metrics, and it's even worse when you are talking about logging instrumentation and tracing instrumentation, because it's just even more data that you are passing through. You want to pay for this application com... |
**Mat Ryer:** Hm. So it really matters, doesn't it? It's amazing, actually, to think of -- I genuinely probably naively thought that that library was quite simple, that it would just have had some helpers that would collect buffering maybe, and then sending off batches, or making them available or something. But yeah, ... |
**Björn Rabenstein:** You definitely don't want to think, you don't want to consider if you use that or not; you just want to use it and it doesn't have a significant penalty. Similar to when I realize in a Go binary profiling is kind of always on, and you just have this endpoint if you just anonymously import this pac... |
**Bartlomiej Płotka:** Right. But there are configuration variables for those Golang profiling that increases the frequency of-- increasing the sampling or frequency of CPU profiling that can trash your application completely; or you can actually reduce it. So there's lots of things you can control here. |
**Björn Rabenstein:** And in what you just described, Bartek, with those new Go runtime metrics, now we also need to consider if we make this user configurable, because it's so many metrics that if you don't need them, it might be too expensive. So, of course, there are traders offs, but the ideal state in some utopia ... |
**Break**: \[19:38\] |
**Mat Ryer:** Bartek, you mentioned earlier that you made a mistake in the client. What was that specifically? |
**Bartlomiej Płotka:** Right. Let's talk about mistakes, yes. |
**Mat Ryer:** If you don't mind... |
**Bartlomiej Płotka:** Of course. Of course. |
**Mat Ryer:** That makes me feel better. |
**Bartlomiej Płotka:** Yeah, so I think that the mistake is really around -- we didn't have time to review exactly properly everything, and that change was pretty big. And as maintainers of Client library, I think we just get that from Bjorn's hands, so we are kind of pretty new. And essentially, usually, it's really a... |
Now, what happened in Go 1.17 (I think) was that runtime/metrics package was introduced, that kind of gave even more information about internal Go runtime, and we wanted to switch, essentially, to this new format... So totally go away from this memstats logic. |
However, what's happening with especially metrics is that we all care about cardinality. We make sure the metric is stable, because this is what allows us to aggregate over time, and also over different series, and kind of compare to each other. This is the key concept behind metrics. And also, the cost of the metrics ... |
So this is generally kind of what we try to achieve. So we have a community who wants to have this new runtime, amazing, granular metrics... So we kind of approve it and merge it, and then suddenly, the rest of the world was like, "Oh, I need to pay twice more. I don't understand really." So this is really the trade-of... |
**Mat Ryer:** \[24:20\] Yeah. That is a big responsibility. Not only because it took off and became popular, but also the nature of it, the fact that it's found its way everywhere. I mean, it really is probably everywhere now. So it really does matter the decisions you take. How do you balance that? Is it a tough balan... |
**Bartlomiej Płotka:** Definitely, yeah. After merging this feature, we got like 12, let's say, issues, and very friendly ones. So we were very, very -- I mean, it was amazing community contributions to just report some problems. So I think that's the pros and cons of such a popular module, is that you're going to get ... |
But generally speaking - yeah, I think there are basic rules when you are writing any Golang code, especially that is consumed by others, like library modules and packages that saves your life. For example, any new feature you should add under a new API, under new methods. You don't change the existing code. Whenever w... |
**Mat Ryer:** Yeah, yeah. I like that you've got nice issues though, at least; like, someone's like, "Thank you so much. I love the recent PR, but--" |
**Bartlomiej Płotka:** "But I need to pay millions more", yeah. |
**Mat Ryer:** "My kid didn't need to go to college anyway. He's probably smart enough, didn't need it." But that's nice that they're nice. Be nice on GitHub, everybody. That's the lesson there. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.