บทถอดความฉบับเต็มของ AWS Cost Analysis Without Dashboards: Cost Goblin

the, um, the sharp edges to avoid. And frankly, the prototype was done in a week. 31 00:04:56.720 --> 00:04:59.260 And, uh, since then I've been keep building on it. 32 00:05:00.060 --> 00:05:10.560 Uh, I can maybe make some announcement to you a bit, uh, a bit now because you, even you don't know about this, but, uh, I have some, some news to share with you because we are going to move away from the, uh, Cure, uh, from the Cure format. 33 00:05:11.480 --> 00:05:18.060 Um, but, uh, maybe, uh, on this a bit later. But the motivation was really to make it accessible to as many people as, as possible. 34 00:05:18.780 --> 00:05:26.400 And then when I decided to start from scratch, as said, I was like, "But why do we even have some server running, uh, remote to display this data to the user? 35 00:05:26.440 --> 00:05:37.340 Let, let's totally, let's totally turn the table and, and make it on the laptop." And, and I really like it. It's like it's extremely, extreme- extremely snappy experience. It's, it's very nice. 36 00:05:38.140 --> 00:05:49.240 No, I-it's like, uh, I think that's getting a bit old school, uh, like being, uh, having things in your laptop or in your, you know, in your local, uh, in your local disk. 37 00:05:49.360 --> 00:06:00.000 So it's, it's, it's amazing that you've been able to do that given that, you know, everything that we work on is, you know, pull this data from whatever, and then it takes a lot of time. 38 00:06:00.740 --> 00:06:12.840 And, you know, it's, it's super, super useful. And, you know, one, one question that I wanted to ask you is like, okay, how... what, what's been your, your journey till that? Because, you know, you... 39 00:06:12.880 --> 00:06:23.860 To make a break like, like the Cost Goblin, uh, that you made, it takes a lot of time, a lot of effort, and you mentioned too, like backstage, which I know from my, from my background on the DevOps and, and the platform side. 40 00:06:23.920 --> 00:06:34.200 So can you tell us something more about, you know, your journey till Cost Goblin happened? [laughs] Okay. Do you want the full picture? So, um- Yeah, yeah. For sure... first of all, I'm French. Uh- [laughs]... 41 00:06:34.380 --> 00:06:42.860 uh, everybody knows already since- Yeah, I think everyone can, uh, can get that. [laughs] That's the first. The second, I... But that's fine. So there is nothing I can do about my accent. Uh, some people say it's cute. 42 00:06:42.920 --> 00:06:56.120 Some people say it's annoying. You will have to decide. Um, but so, um, yeah, so born and raised in France. Um, at the end of, um, high school, I did Epitech. It's, uh, it's a software engineering school in five years. 43 00:06:56.180 --> 00:07:00.860 Super grateful about that. Then I joined SFR. SFR is a cellphone network provider. 44 00:07:01.560 --> 00:07:08.700 Um, and then I started to see what, what it is to deploy application not, not just running on your, on your computer, but running for millions of people. 45 00:07:09.240 --> 00:07:19.880 Uh, super thankful about the architect I had back then because they really taught me the things. Uh, and then I joined Solaris as a backend engineer. So I was for two years working on the backend product of Solaris. 46 00:07:19.900 --> 00:07:28.440 So Solaris is a bit of a weird bank. We don't have customer like you and me. We offer a bank via an API, so other company can then create their banking application on top of us. 47 00:07:29.100 --> 00:07:39.820 And so I was part of the digital banking, um, application, and, um, and back then, um, I was not super happy about the stability of the product. 48 00:07:39.900 --> 00:07:48.180 So I started to shift a bit away from the focus of the team to build dashboards and build, uh, telemetry and observability and start to do some load test. 49 00:07:48.680 --> 00:07:58.080 And basically, my product owner was not super happy because I was not working for him anymore. I was working in my little own world. But other people in the company were pretty happy. 50 00:07:58.580 --> 00:08:08.640 Uh, and so, um, one day the CTO came to my desk and literally say, "Étienne, you have to pack your shit." And I'm like, "What?" He say, "Yeah, you have to pack your shit. You are going to the SRE team." 51 00:08:09.159 --> 00:08:20.220 And I was like, "Okay, let me Google what's an SRE," because I had no idea. And so I Googled what was an SRE, and I was like, "Okay, that's cool. I'm an SRE now." And so the next day I was part of the SRE team. 52 00:08:20.260 --> 00:08:25.040 It was a, it was a playing field. Um, we were just starting the SRE at, at Solaris. 53 00:08:25.780 --> 00:08:35.240 And so yeah, it's a bit outside of the, of the, of the topic of this call, but basically, uh, then it was about doing what I was doing at the scale of a team for full organization. 54 00:08:35.960 --> 00:08:46.160 Um, how do you deploy SLO across an entire org, across many teams? How do you support people to, yeah, bring the success rate up without disturbing the business too much? 55 00:08:46.920 --> 00:08:53.460 And, uh, and then, uh, someone came to me and say, "Hey, we are spending a bit too much money on AWS. Can we do something about it?" 56 00:08:54.180 --> 00:09:02.760 And, uh, and then it was, you know, building dashboard, but not anymore about success rate and response time and latency, but about where is the money going. 57 00:09:03.500 --> 00:09:15.740 Starting to look into AWS tagging, starting to look into how you can distribute cost following ownership, all of the pleasure about broken ownership, about, you know, tag quality, about teams that be... 58 00:09:15.820 --> 00:09:28.080 are renamed by HR because they say this team should not be named that, but that, and then instantly [laughs] all of your tooling goes, goes through the window. Uh, but I mean, yeah, we all know about this. So, um, yeah. 59 00:09:29.300 --> 00:09:40.000 And that's, that's a super in-interesting thing because I, I feel

like unless you are very into the, you know, the hands-on topic or you've been working on that, you are not... 60 00:09:40.540 --> 00:09:55.460 or it's not super aware for people that, uh, the SRE part, so like the telemetry, the observability, how much it plays a role within FinOps, like within actual- So it's connected. It's so connected. Yeah. 61 00:09:56.040 --> 00:10:02.100 Because if you start to rightsize something and then the product, like, collapse, you haven't, you haven't earned anything, right? 62 00:10:02.180 --> 00:10:10.280 So you need to be able to kind of gauge, okay, what is the risk/reward of what I'm, what I'm pushing people to do? 63 00:10:11.060 --> 00:10:18.220 And, and you also have to understand that, hey, this product, maybe it's not super well rightsized, but it's actually a critical product. 64 00:10:18.260 --> 00:10:28.920 So it's normal for this product to not be rightsized because if we have a spike of traffic on this, on this product, and now we have a, we have a SEV 1 incident, I mean, you will not be rightsizing much more things after that, you know? 65 00:10:29.020 --> 00:10:43.720 So, um, yeah. No, especially with, uh, like with things like that, you know, with, um, disasters, uh, and disaster recovery and, and all of that, and all the SLA, uh, complaints. 66 00:10:43.760 --> 00:10:58.020 Like, especially if you have like a, a service that needs to be up, uh, has an SLA, uh, agreement with, with your client, that will probably cost the business a lot of money if you, uh, you know, don't fulfill that agreement. 67 00:10:58.400 --> 00:10:58.560 Mm-hmm. 68 00:10:58.680 --> 00:11:22.420 And, and also it's because c- like as, as we've seen in the updates and everything and, and examples from, like, from the pure AWS team, like how relevant is CloudWatch to do FinOps applications and how much information, like under the hood and for real time, especially for real time kind of FinOps or real time or that takes or notice compared to, you know, core 69 00:11:23.320 --> 00:11:28.720 or things like that that tend to have like a more delayed, um, ecosystem. 70 00:11:28.800 --> 00:11:42.060 And that's a lot of telemetry, which to be honest, I admire all the people that are the case to telemetry because I, uh, you know, in my old times I started to play that role and I was like, "That's... 71 00:11:42.100 --> 00:11:54.100 This is not for me." Like, I don't know how people can start watching to, to the Prometheus data and understand anything or, uh, do the queries for Grafana. And I was like, "Dude, I don't understand anything." 72 00:11:54.100 --> 00:12:00.660 [laughs] You know, there is a g- I have a good news for you. Today, you don't have to do this yourself. You can just prompt an AI to say, "Hey, I want a dashboard-" Yeah... 73 00:12:00.960 --> 00:12:13.436 "that says this," and it will produce a very nice dashboard to you. So maybe you should get back to it Yeah. No, I don't think so. But [laughs] Uh, it was like I, like, uh, unless you've seen it, like the... 74 00:12:13.476 --> 00:12:23.816 When you see like all these Grafana dashboards that look super nice, and they are super clear, just to put an example. Or, and then you see the queries behind it, and it's like before AI- Yeah... 75 00:12:23.876 --> 00:12:32.816 you had to make that at hand or for example, like- Yeah... and the Prometheus queries and like the, the tracking. I was like, "You, you, you, you must be crazy for that." 76 00:12:32.856 --> 00:12:39.016 But, you know, uh, there has to be someone for, for each role in the world. So it's, it's great. 77 00:12:39.076 --> 00:12:48.336 And it's great to see, like, how d- that role has, well, continued to work on, but also people that are being helpful on, on other side. And 78 00:12:49.616 --> 00:13:05.456 maybe to, to continue, it's like, uh, what are, you know, from, from your side, what are the, the recommendations that you have for people to, you know, get the most out of the, out of your talk, and also to, to, for attending these summits and everything? 79 00:13:05.516 --> 00:13:11.056 Like, how are your, your learning process to become, like... Because you are being very self-taught from, from what you said. 80 00:13:11.115 --> 00:13:21.196 So what are your tips on learning more about FinOps and, and how you can, you know, how would you encourage people to, to learn more? Okay, that's a tough one. 81 00:13:21.336 --> 00:13:36.456 Uh- [laughs] I think, I think for some companies that are maybe going to attend and be at the stage zero, it can be a bit like, you know, looking at a very, very tall mountain and be like, "Wow, I'm not sure where to start." 82 00:13:36.516 --> 00:13:45.056 You know? Let's imagine you don't have any tags. Like you, like you are basically on, on a playing field, and you, you cannot understand anything of what's going on. 83 00:13:45.496 --> 00:13:56.136 Maybe sadly, you decided to have everything into one AWS account, and you are like, "Oh my God." Basically, all of the things that we are going to talk about, people will be like, "It doesn't apply to me," right? Um, 84 00:13:57.356 --> 00:14:08.116 like I will say try to start very, very simple. Mm-hmm. And try to... The very simple you start with, try to enforce it somehow. 85 00:14:08.776 --> 00:14:17.896 So for example, for the company that may be using Terraform, let's imagine you are like starting from a very plain field, but you took- Yeah, yeah, yeah. No, no... a good decision to be on Terraform, you know? 86 00:14:17.956 --> 00:14:23.576 It's, it's, it's a lot of people, so- Yeah. Yeah... you, you probably need to- So you are actually, you are actually not on a playing field. 87 00:14:23.636 --> 00:14:33.496 You are already starting with a very big header, and you can already start to apply some very strict rules step by step to enforce- Mm-hmm... some tags on your CI. 88 00:14:33.836 --> 00:14:43.456 And so all of the resources that can be tagged, there is

actually some ways, uh, using Terraform and CI to identify resources that a- don't have tag and should have tags. 89 00:14:44.036 --> 00:14:52.116 And maybe you can start to crank up a bit the rules on this, so your engineering team can over time start to tag things and apply your tagging strategy. 90 00:14:52.796 --> 00:14:58.136 And it will already give you a lot of, a lot of visibility on where is the money going. 91 00:14:58.696 --> 00:15:12.396 If you are on a multi-account strategy and you don't have any tags, you als- already have a big headroom because you can define tags on an account level that you can then fall back on your resources if they don't have tags themselves. 92 00:15:12.796 --> 00:15:21.016 So let's imagine you are super lazy. You have like 10 different AWS account with your environments and your different maybe domains. 93 00:15:21.476 --> 00:15:29.516 You can already tag your AWS account, and you al- already have some tagging apply on all of your resources under, and it will already help you to better understand what is going on. 94 00:15:29.916 --> 00:15:46.396 So what I want people to go out after my talk is like, A, you don't need super crazy tooling. You can just download this, enable your, enable your Focus 1.2, uh, report, and, uh, and get started. 95 00:15:46.636 --> 00:15:56.656 And then start to click around, and, um, start to use the MCP server provided inside of Cost Goblin to talk with your LLM about your AWS cost, and, and just get, get started. 96 00:15:56.716 --> 00:16:09.116 And then, and over time in- increase your, your, your tagging strategy. Um, and it's a journey. It's not a sprint. It's very, it's very a long marathon. It's like a never ending marathon actually. 97 00:16:09.756 --> 00:16:19.896 Uh, and you should not disturb your entire business, uh, running this marathon. Uh, but ideally everyone in the business should feel, uh, part of this marathon. And, and sorry, I... 98 00:16:20.076 --> 00:16:30.856 Now it's a bit lengthy, my answer, but I think we need to shift left the, the focus. Like basically FinOps in a lot of company is the deal of the CTO, you know, the CFO, finance people. 99 00:16:32.816 --> 00:16:41.456 The, the shift need to focus to the desk of the engineers because they are the one incurring this cost. They are the one making the decision about our database size. 100 00:16:41.476 --> 00:16:48.816 They are the one deciding, deciding if we go, I don't know, multi-region or not, if we will use Athena or a classic LDS database. 101 00:16:49.096 --> 00:16:57.756 All of these decision have cost, and there is some good reason to make this decision, and there is some bad reasons. And if you don't see the bill, if you don't think about the bill when you make this decision, 102 00:16:58.816 --> 00:17:05.416 you are not making the decision with all of the information you should. Yeah. No, that's, that's a very good advice. 103 00:17:05.476 --> 00:17:13.976 And I think what you mentioned is that it, it's an ongoing process, and, uh, uh, up until you are super advanced, you can always in... 104 00:17:14.336 --> 00:17:25.876 Like you will have new business that you need to onboard, and you need to, uh, put the systems in place to make that businesses, you know, uh, work with the systems for tracking the, the cost. 105 00:17:25.916 --> 00:17:32.316 And we have a full, uh, attack course thing, uh, that we, we created with, with Alfonso. 106 00:17:32.356 --> 00:17:43.516 And, and it was like super in-depth with, uh, rules for, like if you think about all the rules that you can have for Kubernetes, all the rules that you can have for the CI/CD, uh, with Terraform, with Azure. 107 00:17:43.616 --> 00:17:50.136 If it's on Azure, like if you do it with power, with the scripting, with... There is a lot of things to, to do. 108 00:17:50.236 --> 00:18:02.916 And especially like when you move to the world to, to Kubernetes, then it complex like even more, uh, because you have the shared cost and, and everything. So a lot of work to do, and [laughs] 109 00:18:02.976 --> 00:18:11.216 definitely start from, from the beginning. I have to, I have to be honest, Kubernetes is a thing still on my roadmap for, for Cost Goblin. I'm looking for a nice data set. 110 00:18:11.576 --> 00:18:23.356 Um, actually, I just arrived at EloFresh, and so I will have the n- nice data set. But I, I'm still on my onboarding, so I, I haven't had the time to, to, uh, to, to work on this, uh, for Cost Goblin d- yet. 111 00:18:23.396 --> 00:18:36.636 But, uh, no, uh, the, the support of Kubernetes is something I w- I really want to, to improve on, on Cost Goblin. It's still a bit Inexistent. [laughs] So it's, it's a tough one to- Yeah... to add support for. 112 00:18:36.776 --> 00:18:48.336 Uh- I know... it requires a lot of, a lot of code. [laughs] So now, uh, hopefully AI can help, but, uh, it still requires a lot of effort from, from your end to, to work upon. 113 00:18:48.376 --> 00:19:00.136 So, uh, well, I think that's, that's a very good intro. I think people should have enough information for, from you and to look forward to your talk in the summit the 20 and 21st. 114 00:19:00.576 --> 00:19:06.296 And today, um- May, may I close, may I close the announcement? Yeah. Because I was teasing about the announcement. So- Yeah, yeah, yeah. 115 00:19:06.356 --> 00:19:17.636 So Cost Goblin is switching from the CUR format, so custom user report format, to Focus 1.2, uh, because we are going- Okay... because we are going multi-cloud. So, uh, we are going to support... 116 00:19:17.836 --> 00:19:29.336 We are already, like, bringing GCP- Mm-hmm... and the plan is to also bring, bring Azure. Um, but GCP is something I'm actively working on at the moment, and, uh, and Azure is the next one. 117 00:19:29.736 --> 00:19:39.116 And we also added another functionality, which is profiles that are fully sandboxed. So basically, on a laptop with Cost Goblin, you can now work with many companies. 118 00:19:39.216 --> 00:19:44.396 You can have, like, one co- Well, let's imagine you are a consultant, and you are working with many business. Oh, okay. 119 00:19:44.436 --> 00:19:53.716 You can now create different profile for the different company you are working with, and each profile can have many other, many cloud providers, so be, uh, AWS or GCP. 120 00:19:54.616 --> 00:20:07.456 Um, and you can visualize all of this into, into your dashboards. Um, so I'm pretty excited. You are making it like, uh, an enterprise-grade product now, like. [laughs] We will see. We will see. Step by step. Right. 121 00:20:07.456 --> 00:20:17.816 [laughs] But it looks good, and congratulations on migrating to, to Focus. I think it's, you know, a, a very good step on, on the journey and very FinOps one. 122 00:20:18.416 --> 00:20:30.036 Uh, so looking forward to seeing a lot of support on, uh, for your product. I will leave the, the link to the, to the GitHub in the, in the description, uh, so people can take a look and, and see. 123 00:20:30.736 --> 00:20:43.776 Uh, so again, Étienne, uh, great time talking to you, and thanks for, for taking the time. Looking forward to, to your talk. Thanks a lot, Victor. Looking forward to it. See you. Okay. See you. 124 00:20:43.916 --> 00:20:55.296 [outro music]