text stringlengths 0 2.35k |
|---|
**Mat Ryer:** Yeah, because it did use to be kind of an afterthought really, didn't it? Which is why I think SRE is short for "Sorry", right? |
**Nayana Shetty:** That's one way of looking at it. I always think of it like "Sorry, I don't understand why people don't think about site reliability in the first instance", or "Sorry, I don't understand why people would build this in such a way that it is half broken, or you don't think about the future of this produ... |
**Matt Toback:** Just even on a personal note, I'm excited that Nayana is here and joining us because we met in 2018 for the first time in an attic in Amsterdam... Which - when said that way, it doesn't feel weird at all, right? \[laughter\] |
**Nayana Shetty:** I mean, we were talking loads of monitoring, Grafana, and Graphite, and all of those things... So yeah, attic didn't make a difference then, so... |
**Matt Toback:** But it was wild, because we planned GrafanaCon EU in 2018, which was technically our third Grafana Con, but was kind of our biggest up to that point, and the most what felt well-produced. It was in the middle of an arctic chill... Do you remember that, that the canals had frozen over? |
**Nayana Shetty:** Yeah, I had struggles getting back home... It took me two hours, where it should have probably taken me only half an hour to get home. |
**Matt Toback:** Oh, man... |
**Nayana Shetty:** So yeah, I completely remember that. |
**Matt Toback:** And we were in this stage where Tom had just joined the company, we announced it on stage, the acquisition of Kausal; the entire company fit on stage, it was 25 people when we said kind of goodbye at the end; there's still this photo that circulates here where everyone is just kind of shoulder to shoul... |
But I do remember - you stand out to me, us being up there, kind of in that breakout room, talking about what you were trying to do with the Financial Times... And it feels like you kind of continued in this natural progression, in this natural journey. When you think back to you then, how did you see the world? |
**Nayana Shetty:** \[06:15\] At that point we were investigating -- we had quite a lot of monitoring tools at the Financial Times, and I was working on the team that provided monitoring as a service to other teams... And my head was going mad thinking "Okay, how do I as a team with 4-5 engineers be able to support thes... |
**Matt Toback:** You're still worried. No... \[laughs\] |
**Nayana Shetty:** I mean, I have moved on from the Financial Times, so I'm less worried about the Financial Times monitoring systems, but I still worry about the same use case. I see it here as well in the Lego Group, where there's different monitoring tools that we've got across the organization, and it's "How do we ... |
**Mat Ryer:** Yeah, something you said earlier stood out... This idea that you're like "Why did you build it like this?! If only you'd built it differently, we'd be in a much better position now." |
**Matt Toback:** \[laughs\] "If you only did it right", is that what you're saying? |
**Mat Ryer:** Yeah, but that's the question... When should we start caring about this stuff? When should we start worrying about how are we gonna operate this? |
**Nayana Shetty:** I think this kind of relates to how I've moved in the journey in my career, and stuff. I started off as a test engineer, just doing some manual testing, then moved on to doing more QA, more quality-related things rather than just testing... And over the years I've seen the transition in a lot of orga... |
And one of the quotes I've often used is being kind to your future self. How can you make your life easy in the future? Think about that today, when you're building whatever you're building. And that comes with -- if you're building a new product, think about "Do you even have to build it? Can you just look at what's i... |
I've often used this carrot and stick kind of approach in teams, to show the benefits of what you could get out of thinking about monitoring, observability upfront. Usually, the carrots are like "You build it in the right way, then you can actually forget about your systems, because they will take care of themselves." ... |
**Mat Ryer:** Yeah. I mean, you know, I would be kind to my future self, but I'm too busy dealing with all the stuff that my past self left me to do... \[laughter\] So I don't know. |
\[10:15\] So that's the thing - if you think about how it's gonna be, where it's gonna be running, the realities of that... If you think about that, the earlier, the better, almost, isn't it? |
**Nayana Shetty:** It is that. And it's also that - yes, you are fixing things from yesterday; but if you don't fix it and leave some goodies along with it, then tomorrow you're fixing today's problem. So you're still in that vicious cycle. To get away from that vicious cycle I think you need to actually step back some... |
I remember one of the tech principles we had in the Financial Times for ft.com was -- I've forgotten this... \[laughs\] |
**Matt Toback:** It's okay. As we're talking about the past self and future self - is it okay that I've completely forgotten about observability and now I'm just on a personal journey and I'm thinking about all the decisions made, and yet to make, and how to provide goodies for everyone? \[laughter\] Or for me? I don't... |
But Nayana, the carrot and stick, can I ask you - has there been a stick that you've seen people try to use, that just didn't work? Or not didn't work, but either was too harsh, or just kind of like -- not ill-intentioned, but ill-executed? |
**Nayana Shetty:** I think it's about the motivation factor behind doing something. That's how I saw the carrot and stick. So the carrots were the motivation factors that we were providing to teams, saying "If you did something right and if you thought about how do you monitor something, how do you add alerts in place,... |
But at the same time, we know that every team has these deadlines to meet, and there are product owners who would have their own feature set to build, so it's that kind of scenarios where you actually still need the stick to help the teams be like "Look - I mean, yes, we understand your pressures, but this is more impo... |
Basically, yes, you can go at 100 miles per hour today, but then if you don't build it in such a way that you have put those measures in place, then tomorrow you have to break and stop. But if you slowed down and went at, say, 60 miles per hour, you're there for the long run, and you would go on longer. So that's how I... |
**Mat Ryer:** Yeah, that is so true. We actually built a little project before Grafana, we built a project management tool called Pace. And it was trying to get across that exact thing, which is that you feel great going at a thousand miles an hour, but there's important things to do along the way... And it's hard to r... |
It's a bit like how you design for failure as well. In the perfect world all the messages flow perfectly and there's no problem, but in reality it's way more messy, things fail... So idempotency and things come into play, where you may design expecting this is gonna fail. |
I do write Go code, and Go has error handling as a kind of explicit feature. There are values that are just returned as the second argument to functions, and things like this. And that frustrates a lot of people, because they're used to exceptions or something, that's just sort of automatic... But it forces you to thin... |
**Nayana Shetty:** \[14:14\] And I think it's a myth to think that your system won't fail... Like, always build your system in such a way that it will fail. If it doesn't, then you have a problem. \[laughs\] So make sure you add those checks in place, so when it fails, it can smoothly recover. |
**Mat Ryer:** Yeah. I know some companies that have that as part of the proper testing approach - literally, things will break on purpose; it's a first-class concern that they have. And it is that thing of -- I don't know, is it just ego, that people think "I'm so good. I'll write this, it's gonna be great"? What's goi... |
**Matt Toback:** It can't be... Honestly, it can't be. We've all known and experienced it enough... Do you think? |
**Mat Ryer:** Yeah, I don't know... But the thing is, when I'm writing code and it doesn't work, it's shocking how quickly I'm like "There's something wrong with the processor." \[laughter\] "The processor is not working. Or physics has changed." I'll go to "Physics has changed" before it's my fault. But it turns out I... |
**Nayana Shetty:** I've been in teams where they do pairing, and those mobbing sessions and stuff... They have kind of helped in sense-checking people's egos a bit, and being like "I'm not the best." And when two people talk about it, I think it does help think "Yeah, that is the reality that we live in, and this is wh... |
**Matt Toback:** Is there anything that you would -- even like you were saying before, that progression between being in manual testing, and then Q&A, and then moving to SRE... Was there like a moment where it clicked, where people just started incorporating testing into the code? Do you see the same progression happen... |
**Nayana Shetty:** I have seen it work in some teams, and... A lot of teams I've worked in are all autonomous teams, so they can basically build how they want, using whatever technologies they want. What has often helped teams like that is having some sort of guardrails which actually says -- and also being aware that ... |
But what could happen in these kinds of scenarios is people go all-in and they just say "Oh, I'm gonna monitor everything, have all my logs..." You don't need to go all board on this. There's a limit to how much you need to monitor as well. And understanding the criticality of your app and then building your observabil... |
**Matt Toback:** How would you -- if a team was listening to this and they were trying to understand the criticality of the app and then make decisions around it, if you had them sitting in a room, how would you explain it and say "Here. Start here. Do this"? |
**Nayana Shetty:** \[17:54\] I think it depends on the business criticality. If it is a highly business-critical application, which means if it went down for, say, more than 15 minutes, then we wouldn't be in business - if it's that kind of app, then you need to have your alerting in place, the right level of logging i... |
So there's probably like two levels of monitoring that we should think about. One is the application-level monitoring, and then there is the system-level monitoring. Being able to figure out where the problem is soon enough is something very critical when it's a 15-minute recovery thing. But if it is an application tha... |
So I would suggest teams to think about how critical their app is, and that is something the business should help them with, not something that the team just decides "Oh, this is the most critical thing." |
And once you know the business criticality of something, then it is coming up with some sort of checks saying "If it is a highly critical system, then we do both application, as well as system-monitoring." Otherwise just one of them, based on your use cases, and stuff. |
In the past I've spoken about the USE method, and RED method that we could use for these kinds of things... I prefer using RED method or Google's four signals; it depends on what your team's needs are and what fits into your use cases. So you would use the RED method, which is rate, error and duration for every single ... |
And the same with systems side of things - you would go with the USE method, which is utilization, saturation and errors, and you would do this for the CPU disk or network and all of those different areas, and you basically know where the problem is, and it's easy to find out. |
I would say it is hard, it takes time, so invest based on how much returns you would get on these when you put these checks in. That is something the teams should be mindful about when they are investing in monitoring and alerting. |
**Matt Toback:** Is the primary counter-balance in your mind the effort that it takes to keep this monitored well, or is it also cost? Do you think about the cost to operate, or the backend? |
**Nayana Shetty:** It is the cost, and at the end of the day it should be the cost to the business, as in "How much does having the system down cost us?" and you basically work backwards from there, saying "If this was down for 15 minutes, it would cost the business so much", which means we as a team should be investin... |
I would always focus on the business value rather than the team's individual product value, and stuff... But yeah, it depends on -- like, if it were an internal system... Again, one of the teams I was in, we were building monitoring tools for other teams. So we don't have real business value as such as our team, but we... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.