[00:00:00] Speaker A: Foreign.
[00:00:07] Speaker B: To the cloud pod where the forecast is always cloudy, we talk weekly about all things aws, GCP and Azure.
[00:00:14] Speaker A: We are your hosts, Justin, Jonathan, ryan and Matthew. Episode 295 Recorder for the week of March 4, 2025. Skype files Chime to the grave. Good evening, Ryan and Matt. How are you doing?
[00:00:28] Speaker C: Hey, doing well.
[00:00:29] Speaker A: Good.
[00:00:29] Speaker D: How are you?
[00:00:30] Speaker A: I am back. I survived India.
[00:00:32] Speaker C: Congrats.
[00:00:33] Speaker A: Little jet lag. So if I fall asleep in the Azure section, just wake me up.
[00:00:37] Speaker D: That's normal.
[00:00:38] Speaker C: How's that different from normal?
[00:00:39] Speaker A: Yeah, well, normally I can make it through, but right about the time of recording is when I hit my jet lag wall.
I was in India just long enough to adjust to jet lag and then turn around and fly back to go back the other way, which is the absolute worst. Yeah, it was a good trip out there, but it was warmer there too than here. I got off the plane, I'm waiting for my car and it's like the wind's blowing and it's cold and I'm like, I want to go back for the weather.
But I only had to wait like a month. And then in California it's going to be hot as hell, so.
[00:01:10] Speaker C: Yeah, exactly.
[00:01:13] Speaker A: All right, well, we have some sad news. Play play taps in your head. I don't know the tune on my head, but sad.
[00:01:22] Speaker D: Happy.
[00:01:23] Speaker A: Yeah. It's basically on May 5th, Microsoft is shutting down Skype for good.
I swear this is the ninth time it's died though. Like I, I mean I did have to do some Googling on this and I did see that they did kill Skype for business a while ago with teams. So maybe that's what I'm remembering, but I swear also Skype has died multiple times.
[00:01:43] Speaker C: Yeah, I share that. We were discussing during the the pre read. I'm like, yeah, no, if that's a hallucination, it's shared between the two of us.
[00:01:52] Speaker A: Yeah, it's kind of like Berenstein Bears versus Berenstein Bears. Yeah, I'm pretty sure it was this way.
Skype existed for 21 years before it was, you know, 13 years ago. It was bought by Microsoft and has been under the board control ever since. And for real this time, apparently it's going to go away. May 5th be the last day and then they will short force you into teams, which I'm sure be a great way to get you into teams for all of your personal needs. You should be able to use your Skype login to actually log into teams. And I did try this yesterday which gave me a different interface for teams which I did not know existed. That it's basically the same client, but you log in with your Skype blog and it has basically an entirely different UI than what your teams for enterprise has. And it's sort of weird because it's all tied to the same icon, but it's two different screens. It's very strange, but. So you should be able to do that and to check it out. For those of you who do this, you'll see your existing contacts and chats from Skype now in teams and you can alternatively just export your data, specifically contacts, call histories and chats if you don't want to use teams. So you take it to something better. Current subscribers to Skype Premium services remain active until the end of those terms, but you'll not be able to sign up for any new Skype services at this time. And if you have Skype dialpad credits, which I Learned I had $8 when I logged into Skype for the first time in years and it said, these have expired in 2017. Would you like to reactivate them? I said, well, yes I would and let me reactivate my $7.18 worth of credits. Those will apparently appear inside the web interface and inside teams after May 5th, so I can finish using my credits to dial out with a number. I don't even remember what it is. So yeah, I don't know. Sad day for Skype, but long time coming.
[00:03:32] Speaker C: I mean that's a lot of effort to sunset and reduce that sort of impact to users, you know, something you don't normally see in sunsetting software. That's that, that age. So that's kind of, I'm trying to figure out if it explains like some of the badness in teams or if it's just like really going the extra mile for those Skype customers.
[00:03:52] Speaker D: I think there's a lot of people and you know, at least people I know in other countries that still use Skype like pretty heavily for like cross country communications, things along those lines. So I think it's a lot of that is that there probably is still a good amount of people using it and this is just, hey, they're trying to make it nicely. So how, you know, nice and clean cut over for people versus, you know, the Apple method of it just doesn't work anymore. Good luck.
[00:04:22] Speaker A: I mean I think it was really popular in Europe because I think part of the company was based there.
But I also think, I also think that there was, you know, it made sense in Europe because that's where you used to have to pay a lot of money for long distance calls between countries and so Skype became really popular there and that started to drive it in that direction. But yeah, crazy.
All right, well the other sad news this week, IBM has finished the acquisition of Hashicorp, which they announced last year.
The CEO Arman wrote a blog post reflecting on the journey that Hashicorp has been on and while that's nice if you want to go down memory lane, I was really more interested in the end of it where he talked about the future and his goal as Hashicorp is going to be in every data center and while they made strides as an independent company, he feels incredibly optimistic with IBM that they truly can get Hashicorp into every data center in the space. Now I like to point out that he seems to be a little dated since I don't know how many data centers are left in the world versus cloud.
He did say they will gain access to their global scale and increase R and D resources as well as he highlighted many of the integration opportunities of IBM and the Red Hat portfolio. Things like integrating Terraform for provisioning with Ansible for configuration management will enable an end to end approach to infrastructure automation as code, while integrating Terraform with Cloudability will provide you native finops capabilities to manage and optimize your costs as well as things like Vault integration with OpenShift, Ansible and Guardian will bring world class secrets management to those platforms and reduce the integration burden on end users. Which I just thought wow, Ivan is going to make a bunch of money because if they forced me to use Vault and Terraform Enterprise for all those capabilities because Hashicorp was never shy to charge you at least $400,000. Yes, that was the starting price for pretty much everything. So like there for quite a while.
[00:06:07] Speaker C: Yeah. So yeah, it'll be interesting see because they were also big on bundling, but the stuff that they couldn't sell it is interesting. Yeah, the data center comment I thought was funny too and I just assumed he was talking about, you know, because it's clouds running on data center somewhere just abstracted, but who knows.
[00:06:31] Speaker A: All right, Moving on to AI is how machine learning makes money. This week OpenAI has launched GPT 4.5, their largest and best model for chatting yet. GPT4.5 is a step forward in scaling up pre training and post training operations. OpenAI says early testing shows that GPT 4 to 5 feels more natural with broader knowledge base, improved ability to follow user intent and greater EQ or emotional intelligence for Ryan making it useful for tasks like improving writing, programming and solving practical problems as well as they expected to hallucinate less. Well, that's nice.
[00:07:05] Speaker C: I'm just happy to see the Death of the 4.0.
[00:07:09] Speaker A: Oh, the 4.0?
[00:07:10] Speaker C: The 4.0, yeah. Yeah, that'd be great.
[00:07:13] Speaker A: I'm definitely happy about that. Until they release 4.50.
[00:07:16] Speaker C: Oh God, you're right.
[00:07:19] Speaker D: You know it's coming.
[00:07:21] Speaker A: Yeah, they've established that the 4.0 is the reasoning model and so it has to be a chat GPT 4.50.
Yeah, this is interesting. It's only in Research Preview they did start rolling it out on one of the tiers for subscribers, but they talk about in one of the Twitter posts I saw that basically Sam Altman said that they don't have enough capacity, not enough GPUs to roll it out, and that's because it's the heaviest, most expensive model they've ever built.
Apparently it's big, expensive and slow, providing only marginally better performance than GPT4O at 30 times the cost for inputs and 15 times the cost for outputs.
[00:08:01] Speaker C: Wow.
[00:08:02] Speaker D: I feel like they went the wrong direction.
[00:08:04] Speaker A: Yeah, it feels like they're supposed to.
[00:08:05] Speaker D: Be faster, cheaper and better, not more expensive.
[00:08:09] Speaker A: Yeah, well, the market seemed to agree with your take, Matt as Gary Marcus, who's an author of rebooting AI and founder and CEO of a company called Geometric Intelligence, which was acquired by Uber, called ChatGPT405 a nothing burger and Ars Technica called it a lemon. So that's. That's unfortunate. Uh, basically, you know, reading through these Gary Marcus says he predicted that chat GPT 405 wouldn't be that impressive and that the pure scaling of LLMs or adding more data and compute has hit the wall. And he claims that he was right. Hallucinations didn't disappear and nor did stupid errors. And he points out both Grok 3 and Chat GPT4.5 didn't fundamentally change anything and both are barely better than Claude 3.5. I wonder what he says about Cloud 3 7. Oof, that's rough. He quotes other AI forecasters who move projections for AGI to later from Eminent and even pointed to that post that Sam said about the GPU points as well. So yeah, overall the feedback has been sort of limited. I do look to see get into more of the ChatGPT tiers as well as in Microsoft's world. But yeah, early reviews not great.
[00:09:15] Speaker C: It's interesting because it's in the consumer space like you got to have flashy changes that dramatically change the user experience. Right. So it's like you always want to do incremental improvements, but if you're announcing large bottle stuff, you know it's going to have a huge effect on your stock value if the new stuff is just more expensive and more of the same. So it'll be fun to see as they navigate this because it's new business model and uncharted territory.
[00:09:43] Speaker A: Yep.
I mean, I'm still just excited for Cloud 3.7. I've been loving it. And that code assistant they gave out, even though it runs on Node js, which I hate, it's quite good. I quite enjoyed it and in fact I just got access to Cloud 3.7 through GitHub Copilot, so I'm excited to see what that looks like too. But yeah, GPT 4.5, we'll see how it gets adopted, especially as it leaves research tier. Maybe they fix some of the shortcomings right now as it's in this early stage. Maybe they can make it faster and cheaper. Hopefully.
Well, an interesting turn of event. Microsoft is urging the Trump administration to ease export restrictions imposed on AI chips at the very end of the Biden administration. Microsoft says the rules disadvantage allies including India, Switzerland and Israel, and limit the ability for US tech companies to build and expand AI data centers in those countries. Tighter US restrictions on exports of advanced AI chips to Beijing are keeping American chip makers and big tech from serving one of the largest markets for semiconductors, accelerating a global race for AI infrastructure dominance. Microsoft says this will force some allies to turn to the Chinese market in the absence of sufficient supply of US tech and left unchanged, the rule will give China strategic advantage in spreading over time its own AI technology, echoing its rapid ascent into 5G telecommunications a decade ago. And I don't know, this is a weird area. So if we give them our AI chips, then they could build better AIs like deep seq on them. But by not giving them our chips, they could also create Deep SEQ with old technology. So I don't know if there's a win, win, lose, lose here or what, but definitely this restriction has not seemed to do much to curb their ability to innovate. But we'll see.
[00:11:24] Speaker C: Yeah, I'm conflicted because it does. Like my first take on it was that Microsoft just wants to make more money by.
[00:11:31] Speaker A: Well, that's the anti imperialist side of us who were like screw those companies and their capitalism wanting more money.
[00:11:38] Speaker C: Exactly. But then it's sort of this. It's supply and demand too. Like there's not enough of these GPUs. So it's interesting there too because isn't this going to make this problem worse?
[00:11:50] Speaker A: You would think so, yeah.
[00:11:52] Speaker C: So but then you know.
Yeah, I don't, I don't know how, how that positions US market versus foreign markets and that gets challenging. Yeah, I don't know, I don't know what to do on this one. Like.
[00:12:09] Speaker D: The innovation side of me is like we cause a large group of people to not have something, they will maybe figure out a way to optimize it and make it be better and then improve everything as you know, as a whole.
[00:12:21] Speaker A: But they basically said well we can't get these chips so we're going to figure out a cheaper way to build a model and then cause everyone to pain. But the other reality is that I'm sure China is getting access to all these chips through some other country who doesn't have the quite same restriction controls. So they buy all the chips from the US and then they sell them on the dark market to China. I'm sure if they really wanted them.
[00:12:47] Speaker C: It's probably scale though. I mean that's harder.
[00:12:49] Speaker A: Yeah, I'm sure. I mean any tens of thousands to run the GPT4.5 model, I mean who knows how many China would need to run their models because they have double byte characters in their words.
So there's a lot of storage required for that. And you have a bunch of other things.
We'll see if that goes anywhere. But this administration, I can't make any predictions because it's unpredictable.
[00:13:10] Speaker C: It really is.
[00:13:13] Speaker A: Amazon Web Services this week has several things for us. First up, they're renaming AWS Chatbot to Amazon Q Developer. The new name recognizes the integration of Amazon Q Developer most capable generative AI Power assistant for software development in Microsoft Teams and Slack to manage and optimize AWS resources. With Q Developer customers can monitor, operate and troubleshoot AWS resources and chat channels faster. Customers completely retrieve telemetry and ask questions to understand the state of their resources. Now I have so many thoughts and I could probably rant for this so I'll see. What are your guys thoughts about this.
[00:13:47] Speaker C: First, I mean I, I feel like you know, they're going to really force the Q branding down, down our throats and Amazon's good at that, you know.
I don't know.
[00:14:00] Speaker A: Yeah, so this, so this is where I sort of get weirded out by it. So AWS Chatbot is a very simple. I'm going to make a Request and I have to use a certain syntax in the AWS chatbot to slack. And then it calls the API and it returns data from the API that Amazon provides that I've synchronized and I have authorized and it provides accurate data back to me. Amazon Q does not provide reliable data ever. It provides hallucinations. So if I ask it like how many graviton based computers computers am I running in this region? And it comes back and says 32, can I trust that there's 32 boxes running or do I have to go double check it now? Because you're using a LLM in the middle of this thing that doesn't know what the hell it's doing? So I have that concern, first of all. And then the second concern is if everything becomes Amazon Q, how do I know what any of it is? Because there's Amazon Q Developer, there's Amazon Q Business, there's Amazon Q for something else. There's a bunch of random things in this Amazon Q name. And so it doesn't. And also I don't. I knew what a chatbot was. AWS Chatbot, check. Got it. Amazon Q Developer. Do I know that's what that is? No, I think Amazon Q Developer is a tool for your IDE to help you be a better developer. So this is branding confusion to me for absolutely zero reason or for someone to build an empire inside of Amazon who's in charge of the Q team, who's like, I want that chatbot thing.
I think, and I just think this, like this is a problem because I think it puts something that was very known an API response and now puts it into a situation where I don't know that I can always trust what it says? And do I trust it to send alerts to me? Do I trust it to send the data that I need to know how my account is working? And my answer to that question right now is new. And that's just based on using Q Developer or whatever the Q thing is inside of the AWS console. Because that thing is garbage.
[00:15:50] Speaker D: I mean, here's the thing. Most people, I think, kind of blindly trust the output of, you know, any of the LLM tools out there. You know, they, they look at the results like, yeah, it's good enough. I assume it's correct. They don't really dive deeper into it. I think the average consumer doesn't. So here the average person isn't a cynics are like, wait a second, hold on. How are we going to validate that this is actually accurate? Cool. Because they had an issue last week. You know, we're gonna be far more cynical about it, but I think the average person that's just communicating with it is gonna just blindly say, yeah, 32 feels like the right number, even though you really have 332.
[00:16:33] Speaker C: Yeah, yeah.
[00:16:35] Speaker A: I mean, honestly, you might feel that way about your LLM until you catch it in a lie, and then you never trust it. And I know, like, how I felt about L. I mean, I really, like, these are really great and all that. Then, like, I was. It gave me an answer one time and I was like, yeah, I'm with you. That doesn't feel right. And then I like. Or it was, oh, I know what it was. It was. I was asking about if there was a defect for an issue in Amazon's knowledge base or something or in Microsoft's knowledge base about this product. Because, like, I was getting an error message and I was asking it and it goes, yes, there's a knowledge base article for that. And I'm like, cool, can you please provide the link to the knowledge base article for me? And it goes, hmm, sorry, there is no knowledge base article for this. I'm like, but you just told me there was. And so I challenge it and it comes back and it's like, well, unfortunately I made a mistake and there is no article for it. I'm like, so that's one. And then I've also had a lie about policies that exist at our day job where we have our own AI thing for that. And it's told you one thing, but you actually go read the policy, the policy says differently.
So I've had enough of those experiences now. Where. And also in code, even Claude, which I love, and anthropic, it writes code that doesn't work sometimes and you can fix it, but it didn't work the first time. So, like, how much do I trust it?
[00:17:51] Speaker C: Yeah, I had Claude guessing incorrect Iam permissions the other day, which is fun to troubleshoot because it's auto completing all these things. I'm like, oh, how convenient. Until that service just didn't exist.
[00:18:04] Speaker A: All of a sudden, that doesn't work.
[00:18:06] Speaker D: I just like how it made up random API calls for me that I was supposed to do ran out. I was like, oh, this looks about right. You look at the script, you're like, that looks right. I go run it. It's like, this command's not found. I was like, oh, you just made up. Get whatever it was because it was a PowerShell script. I was like, cool, cool, good chat. Guys, let me go double check everything I've now done. But I think that we are more of the outliers than the average consumer. Maybe I'm wrong.
[00:18:38] Speaker A: I think you have a natural tendency for these things to hallucinate in software coding because they typically follow very specific patterns.
So if the LLM can determine what the pattern is, filling in words for the pattern is super easy. Like, oh, you're asking about this thing. Well, if I take this pattern and I put that thing in this pattern, then I'm going to answer your question. And it thinks it's right even though it's not a valid call.
I think that's why we see it. I think all developers probably have this complaint because we all do very heavily pattern based things.
That's mostly what coding is, is learning how to take English and put it into the proper syntax of the computer understands you, you know, to simplify our jobs to way too much. But you know, like that's the reality of a lot of stuff we do. So I think we are, we're much quicker to catch a lie. But I think that's the risk of things like Grok and stuff inside of Twitter that freaks me out that you know, this thing is just making up answers and people are like oh yeah, so and so did you know was in the, you know, the files of that pedophile? I don't know. Yeah, you know, like it can lie really quickly and then how do you check it?
[00:19:43] Speaker C: So it is sort of it, it's. I, I find myself using AI in ways where even if it, it's not going to hallucinate because I'm asking it to generate its own content. And so like it's, it's either I like it what it generates or not. And then I find myself still using search for things where I need a specific answer even though the AI is now part of that search result. Like I like having, I like having that format where it's giving me all the links that use to base that on there. It's kind of interesting.
[00:20:11] Speaker A: Yeah, it's nice about the reasoning. AI is where they, they give you the sources that they quoted. Like that is kind of nice. I do like that.
[00:20:17] Speaker C: But yeah, no, it's, it's. I get really conflicted on these things because it is sort of like if you start asking it, you know, like more generic like process related questions or how do I do this? And it starts leading people down wrong paths. Like that's a huge productivity time suck which is the, you know, the whole thing we're trying to solve with these AI assistants. So.
[00:20:40] Speaker A: Yeah, so Amazon F for this article.
This idea is an F. Everything about this is a fail. I. I'm just not happy that you think this is a good idea. But we'll see. You know, Q Developer will probably get replaced by something else at Reinvent and then we'll be even more confused by the brand confusion they have caused.
Amazon ECS is adding support for additional IAM conditional keys. These are eight new service specific condition keys for iam. These new conditions keys let you create iam policies and SCPs to better enforce your organizational policies in a containerized environment. The IAM condition keys allow you to author policies and enforce access controls based on the API request context. And today's release has added those condition keys and those are ECS Task cpu, ECS Task Memory and ECS Compute Compatibility container privileges, ECS privileged network configurations or ECS Auto Assigned Public IP or ECS subnet and tag propagation ECS propagate tags and ECS Enable ECS managed tags for your applications deployed on ecs. And I struggle with this one. So I'm hoping the two of you have ideas. So I don't know how I would create an IAM conditional key based on task memory.
[00:21:53] Speaker D: Yeah, so we were talking before and I have a thought.
[00:21:58] Speaker A: Okay.
[00:21:58] Speaker D: They last week, two weeks ago, in the last four weeks, go with that. They released the 128 CPU limit.
[00:22:07] Speaker A: Correct.
[00:22:09] Speaker D: Is this saying you can create an IAM role that says you cannot do something stupid and you cannot create an EC2 sorry ECS container that contains it and you can limit.
[00:22:22] Speaker A: This is just an access control. So basically you'd be able to create a task with 64 CPUs, then have a conditional key says you can't access it because if it's over 32 you could do that. But I don't think it actually prevents you from doing anything.
[00:22:35] Speaker D: All right, I was trying.
[00:22:36] Speaker A: It was a good try. Applaud your effort.
[00:22:41] Speaker C: Nothing.
[00:22:42] Speaker A: Yeah, I don't know why you define policy for this because that's attributes attribute Attribute based access. That makes sense with tags, but like subnets. Weird, because I think you were saying, well, I only wanted to launch from private subnets, but I'm like, private subnets are not labeled private. There's not a checkbox for that. You just, you put a tag on them that says they're called private or you label them Private is the part of the subnet name. So basically you'd have to list all of the subnets that are considered your Private subnets by subnet id, if you want to use that for your access, which I guess that might work, but again, like, it's just, it's so weird. Like, I don't. I'd love to hear what someone's using this for in the real world, like task CPU and task memory. Like, please tell me how you're using that. That makes sense to me because I, I don't get it.
[00:23:24] Speaker C: And at what point are you granting, you know, those IM permissions so granularly like where you, you have an SRE team, you don't want them to change the task version, but they can change the sort of runtime configuration? Like, is that really a concern? You know, like, maybe, but I wouldn't want to gate something that closely.
[00:23:42] Speaker D: Okay, so the. I went into it, It's a subset of the Create service which says grant permission to run and maintain the desired number of tasks from a specified task definition via service.
So I think I might be right with the CPU task in there where you could say you can't create a CPU of a certain thing.
[00:24:08] Speaker A: Okay, yeah, you're right.
[00:24:10] Speaker C: It is. Yeah. You can't create a service.
[00:24:13] Speaker D: You can't create the service that. Or update it, if it's.
[00:24:16] Speaker A: Okay, so it's a sub task to a different item. Okay, that. That actually does make some sense then. Okay, that's helpful. We figured this out.
[00:24:23] Speaker D: Required a little bit of digging, but it has to do with the Create Service API.
[00:24:29] Speaker C: It does take an integer. So you could, you could enforce a limit with this. Absolutely. By using IAM permissions.
All right, now I get it.
[00:24:38] Speaker D: They don't understand the network A1 still, so I might ignore.
[00:24:45] Speaker A: Maybe you're preventing the service from being created and you only allow it to be created in certain subnets.
[00:24:51] Speaker C: Yes. Array of string is the input, so that.
[00:24:53] Speaker D: Then you still have to have the strings.
[00:24:55] Speaker A: Like Justin said, that's what you do in Terraform too. To be fair, it's a conditional policy.
[00:25:01] Speaker C: Right. And so, yeah, you can. It's because it's an array of strings. You can wildcard it. You can do those things. So yeah, you know, if I want.
[00:25:08] Speaker D: To say only launching my private subnets and I have a hundred accounts, I have to list out all hundred account subnets.
[00:25:15] Speaker A: Yes, you would. Or. Yeah, so I mean, that's why in that case you'd want to have a tag on the subnet and then you just use a subnet tag as your attribute.
[00:25:23] Speaker D: That makes more sense. But this.
[00:25:25] Speaker A: But. But I mean, not everyone has that problem. You're stating. So some people maybe only have 10 subnets they care about and so for them putting 10 keys and it's not a big deal. I mean like, is that how you would do it? No. Is how someone else might do it. Maybe.
[00:25:38] Speaker C: Yeah. If you have a subnet per team or you know, subnet per environment that's, that's named appropriately, you can enforce it that way. Or you know, maybe you could just restrict access to deploy into public. Except for like production team.
[00:25:54] Speaker D: What's compute compatibility? Is that just like Fargate vs standard?
[00:26:00] Speaker A: I might be Fargate. I think I was thinking it was more intel vs AMD vs that was my thoughts.
[00:26:07] Speaker D: Required compatibility fields.
[00:26:10] Speaker A: Yeah.
[00:26:10] Speaker D: Oh, so that's also like if you're running like this, that's running Graviton versus Graviton. Yeah, intel versus you know, all that stuff. So that makes more sense then. Okay, yeah.
[00:26:21] Speaker A: So I only want you to. I only want you to be like. Yeah, because like I only want you to be able to launch Graviton based instances up to the size of memory and cpu. Like that makes sense. Like I can see that those set of conditions make sense in that context.
Okay, well I'm glad we figured that out because that article did not make sense.
All right. Announcing extended support for Kubernetes versions for Amazon EKS Anywhere. This is very similar to something that Google did recently for Anthos. Basically they're extending support for the Kubernetes versions. With extended support, you can continue to receive security patches for clusters on any Kubernetes version for up to 26 months after the version is released in EKS anywhere. Extended support for Kubernetes versions of Rena Sandwich is available for Kubernetes 1.28 and above currently. So if you are worried about long term supportability of Kubernetes and you don't want to upgrade it every month as you probably should, you can now get 26 months of support.
[00:27:18] Speaker C: I just think this is funny that it's a EKS Anywhere feature because it's like, oh yeah, no, the team that was not moving their Kubernetes clusters out of their data centers into the cloud like that. They're going to upgrade very quickly. Probably not. That makes sense.
[00:27:35] Speaker A: All right.
You now get insights from the multimodal content within Amazon Bedrock Data Automation now generally available. This is announced at Re Invent. It's a feature to streamline the generation of valuable insights from unstructured multimodal content such as docs, images, audio and videos, reducing the development time and effort to build intelligent document processing, media analysis and other multimodal data centric automation solutions. This capability is now available with the support for cross region inference endpoints to be available in more regions and seamlessly use compute across different locations. And based on the feedback during the preview, they also have improved accuracy and added support for logo recognition for images and videos. So if you have a video and you want to count the logos in it, you can now do so. I guess it's nice to see automation capabilities of it's like a very Gentek AI without having to do all the heavy lifting of a Gentex AI at the moment.
[00:28:27] Speaker D: Yeah, the logo recognition is a nice ad I feel like too.
[00:28:31] Speaker A: Yeah, there's a bunch of super bowl commercials. Can you tell me how many ads for each company there was?
[00:28:38] Speaker D: Okay, I was even thinking like within within videos, like flag any, you know, logos that you see on clothing that to blur out or whatever like movies or things like that.
[00:28:55] Speaker B: There are a lot of cloud cost management tools out there, but only Archera provides cloud commitment insurance. It sounds fancy, but it's really simple. Archera gives you the cost savings of a one or three year AWS savings plan with a commitment to shortest 30 days. If you don't use all the cloud resources you've committed to, they will literally put the money back in your bank account to cover the difference. Other cost management tools may say they offer commitment insurance, but remember to ask will you actually give me my money back? Achero will click the link in the Show Notes to check them out on the AWS marketplace.
[00:29:35] Speaker A: All right, move on to Google Cloud.
You now get coding help from Gemini Code Assist for free.
If you can't sell it to engineers, just give it to them and then go after them for licensing violations later.
So that's available to you now for free. You can set that up pretty quickly and easily. If you don't have a Google Cloud account, it's pretty fast to do so.
[00:29:59] Speaker C: Is this because they couldn't sell it or because they couldn't figure out the pricing model for it? Because it took them forever. And then the last time it was explained to me how much it was going to cost, it was so convoluted that I could not understand it. So it's a lot easier just to give away for free than actually define a pricing model that makes sense.
[00:30:20] Speaker A: Yeah, I'm sure they'll give you enterprise features for more unlimited APIs, but if you give them a free taste of Code Assistant and then you're like, oh I love this now you're willing to pay versus like I want you to pay a bunch of money. And you're like, yeah, I don't know anything about this and if I trust you, but I don't know that I, I think I tried to use Gemini coast once and I didn't care for its recommendations compared to GitHub Copilot. And then of course Claude, you all know I love so but it's interesting.
[00:30:49] Speaker C: As being part of the POC and our day jobs that you know, like to see the back and forth because like when I first tried Gemini it was total garbage.
And I started using chat with GPT and then I heard feedback from here that Gemini had gotten better. So I tried it again and it was much better.
So it's like it's one of those things where it changes over time and you know, now I hear rave results from both Gemini and Chet or sorry co pilot.
[00:31:15] Speaker D: So.
[00:31:18] Speaker C: I guess it matters maybe on context or just with time.
[00:31:23] Speaker A: Yep. I mean they all get better over time. That's kind of the reality. I mean I feel, I feel honestly because GPT 4.5 has been so long that the code capabilities of GitHub Copilot were falling way behind. What other clouds? I don't know if GPT 4.5 has been added into that yet, but I'm curious if that actually has an improvement to coding practices. They're not really touting that as it's big improvements. They're going more with the natural language feel of a conversation as their selling point, which is less of a selling point in code. So we'll see.
Google says if you've never worked in cloud, it can be hard to know where to start. Even if you're a seasoned cloud architect. How do you pivot to your next big thing? And once you find it, how do you pinpoint the career of your dreams? The biggest hurdle of all is knowing the skills and training that will help you get to the job that you want. If you're dreaming of a new direction in your career or a new one entirely, Google gives you Career Dreamer. It's an AI powered career solution that you can go and determine the skills and things you need to learn for your next dream role and personalize it to going through the questionnaire creates a custom prompt for you to use inside of Gemini and then Gemini will act as a career coach, helping you learn the training and benefits of the things you need to make your career change.
This also is a good timing with our other article which we didn't link to for our main show. But apparently Google Cloud certified in 2025 is a great way to increase your value in your company, per the research that Google paid for.
Nice. Yeah.
[00:32:49] Speaker C: I mean, this is way better than my usual method, which is complaining about something until they just give you that responsibility and make it your job to fix it, which is how I've advanced through my career.
[00:32:58] Speaker A: So I would recommend that's why you're now in security.
[00:33:03] Speaker C: Yeah, this is backfired horribly.
[00:33:07] Speaker A: Yeah.
This is fun, actually. And this is another use case I never thought about for my ChatGPT uses like, oh, a career coach. That's kind of cool. And so I did play with it. I did create a prompt, and then I changed our prompt completely because they didn't quite have what I wanted for my next career, which was basically man who doesn't work but owns an expensive bar.
And so I had to tweak it a little bit, but it basically advised me to give up my dreams and to move back to cloud work because it pay better, which is probably true.
[00:33:40] Speaker C: Yeah, it's in a bar for sure, where I'm going to drink all the profits.
[00:33:49] Speaker A: It's like Jonathan, but an AI form. Because once we dreamed about buying a winery and owning a winery, and then he did the math for us. He was like, okay, well, if you guys are going to drink this much wine, then we need to produce this much wine. And so that'll never break even and we'll go bankrupt in seven years. And we're like, wow, you really killed this dream. So now AI can do it for Jonathan.
[00:34:05] Speaker C: Yeah, exactly.
[00:34:06] Speaker D: Yeah.
[00:34:08] Speaker C: I'll never forget the spreadsheet of, like, cost per gallon of water and square foot of land and how he had it all broken down. It was just. Not only did he kill that dream, he came with the receipts.
[00:34:18] Speaker A: Yeah. It wasn't just a, like a wave of the hand, like, you're stupid, you're going to drink all the profits. And we're like, yeah, yeah, whatever. And he's like, no, no, let me show you in spreadsheet four, which is always, always the best thing about Jonathan. Jonathan is great with a good spreadsheet and good, good analytical data on it. So.
But we, you know, we're. Our health plans at work. We just ask him, which one should we buy, Jonathan, which one saves us money, which one doesn't. And he just, he has a spreadsheet already figured it out for us. So that's great. Yeah, that's to be replaced by AI someday. So Jonathan's. Jonathan's Days are numbered.
[00:34:50] Speaker C: See? We'll see which hallucinates more, Jonathan or the AI.
[00:34:54] Speaker A: That's a good question.
Could go either way.
All right, well, Google is introducing a new enhancement to help you get even more out of Vector Search and AlloyDB. First, they're launching Inline Filtering, a major performance enhancement to Filter Vector Search and Alloy DB being able to perform vector search directly in the database instead of post processing on the app side, Inline filtering helps ensure that searches are fast, accurate and efficient, automatically combining the best of vector indexes and traditional indexes on a metadata column to achieve better query performance. Second, they're launching enterprise grade observability and management tooling for vector indexes to help ensure stable performance and the highest quality search results. This includes a new recall evaluator or built in tooling for evaluating recall, a key metric for vector search quality. And you no longer have to build your own measurement pipeline and process for your apps to deliver good results.
[00:35:43] Speaker D: These all seem like really good things. I have no understanding of how to use them all.
[00:35:48] Speaker A: Yeah, when you guys are talking about vector search and databases and like combining the metadata of different column types, I'm like, I sort of know what you're saying because I read some documentation, but I don't know how to use this in the real world.
[00:36:00] Speaker D: Yeah, I'm like, I understand the general premise and that's maybe it.
[00:36:05] Speaker C: Yeah, I think it's a combining just, just to add in my own SQLitnessy.
I think it's a combining of two of different, like sort of database types. Right. So you have all your data in a giant relational database, but it's really difficult without doing a whole bunch of heavy queries to sort of report on statistical analysis of that data. And I think that technologies like LADB are trying to give you your cake and eat it too, in a sense where you can have all your data in that database, but do a vector search across that data set to get your analytics out of it.
[00:36:40] Speaker A: See, now I thought it was more about solving one of the challenges with databases and doing things like, similar to queries.
So like, I'm looking for the color red, but my database has all kinds of colors of red because I've got, you know, pink and yellow red and all these like protruse and like, I don't want weird red colors. I don't maroon, you know, all these sort of red colors that are technically red, but they're in the red family spectrum. And so vector databases basically takes the idea of red and then turns it into vectors. That way you can identify all the flavors of red without having to do heavy computational tasks. That was sort of my understanding of vector, which is probably a really simplified version of it. But I mean, that still fits with.
[00:37:30] Speaker C: My understanding of what I think it is too. Because to get those analytical results out of a typical relational database, you have to have a query that encompasses every single shade of red. Yeah, right. Versus a vector search to get those analytics. How many customers have red type things? Now you can create that same vector search for those analytics and generate the analytics out of that.
[00:37:51] Speaker A: Yeah. Okay. So we're talking the same but different.
[00:37:54] Speaker D: My favorite part of that was listening to Justin try to come up with red light.
[00:37:58] Speaker A: Color.
[00:37:58] Speaker C: Red.
[00:37:58] Speaker A: Yeah. I was like, why did I choose red? That was a terrible choice. Yeah, yeah. Blue. I gave you a bunch of blues. Red. Yeah. Red. I fell apart very quickly.
I'm like, what's the other types of red? Pink, maroon.
I needed AI to help me out on that one.
[00:38:14] Speaker D: Red. Yellow, I think it was.
[00:38:16] Speaker A: Yeah, yeah, it's a red. Yellowy red.
[00:38:19] Speaker C: He came up with more than I would have.
[00:38:20] Speaker D: So, you know, I believe in the Roy gpiv. Those are the only real colors.
[00:38:25] Speaker A: This is why I need a vector search because I don't know the name. I don't know the names of colors red either. So this is why I can't write.
[00:38:30] Speaker D: That query partially colorblind. I've solved that problem.
[00:38:33] Speaker A: Perfect.
Yeah.
Google is sharing the general availability of Terraform providers for Oracle database at Google Cloud. You can now deploy and manage Oracle Autonomous database and Oracle X database services using the Google Terraform provider. I've always dreamed of being able to bankrupt a company with Terraform apply for my Oracle X data use cases. So thank you for that Google. I really appreciate it.
[00:38:56] Speaker C: Yeah.
[00:38:57] Speaker D: This is where Terraform needs like, are you really sure?
Like, like do not allow dash dash auto apply. Like auto approve. Like you need to have a secondary person that approves this somewhere.
[00:39:12] Speaker C: Type in the full name of your CFO to apply.
[00:39:17] Speaker A: This is where I want that cloudability integration of Terraform so I can like. Let's get on that IBM. I need to know how much this is going to cost me before I terrify Apply Approve there.
[00:39:26] Speaker D: There are a couple tools out there. I know I did one like a POC of one like four years ago now for aws and I've had somewhere in my backlog at this job for since I started there too.
Azure has a corresponding tool like that where like on the apply it will say this will increase your spend by X dollars. I always thought that would kind of be fun just seeing a pr, but yeah.
[00:39:52] Speaker A: So the one that I used to use on AWS was infra cost and I think it's been expanded out to Azure and to GCP now as well. And then the other one that was pretty popular was Kubecost, which did it for Kubernetes tasks and services and pods.
And I believe that has now been bought by Cloudability as well. So yeah, there's definitely some options. And then of course Terraform Enterprise gives you one of the features out of the box. But yeah, I think being able to integrate it directly into your phone UPS tool is probably going to be better.
[00:40:21] Speaker D: Wow.
[00:40:21] Speaker C: It depends on where you want to put it, but you want to put it in front of everybody, so you want all the knobs.
[00:40:27] Speaker A: Exactly.
[00:40:28] Speaker D: Infra cost has gotten a lot better since I played with it like four years ago.
[00:40:32] Speaker A: Well, that's what happens in open source. You stop using it, it does not stop getting innovated on.
[00:40:37] Speaker D: I know.
[00:40:37] Speaker A: If it's popular.
I want to mention all those projects I found have been touched for eight years. You're like, okay, that's not going to.
[00:40:44] Speaker D: Work well, There's a whole SaaS now around infra cost.
We're like, there's nice pictures that tells you where your spend is going to be and how you decide to break everything on Azure because you charge to imply. Yeah, normal things like that.
[00:41:03] Speaker A: Yeah, good.
[00:41:05] Speaker C: Yeah.
[00:41:05] Speaker A: I didn't realize I had a SaaS product. That's kind of neat actually. One more FinOps tool in the bin.
All right, let's move on to Azure, who just had a ton of announcements this week. They wanted to make up for lost time.
First of all, Azure AI Foundry is getting support for, of course, OpenAI's GPT4.5 in preview.
The research preview demonstrates improvements from scaling pre and post training. A step forward in unsupervised learning techniques as natural integrations with broader knowledge, higher eq as we mentioned before, and improvements in coding, writing and problem solving tasks. Accuracy and illustrations will be better with hallucination rates being 37.1% versus 61.8%. It was lying to you 61.8% of the time before. Wow. Wow. As well as a higher accuracy, 62.5% accurate versus 3.8 or I think it was 38%. Everybody does something there. Stronger human alignment improves the ability to follow instructions, understand nuance and engage in natural languages with you. So that's available to you@Android AI foundry. They've also launched the latest version of Microsoft Phi which continues to push the boundaries of what's possible with smaller and more efficient architectures. The 5.4 multimodal unifies text, speech and vision for context aware interactions. Retail kiosk can now diagnose product issues via camera and voice inputs using this 5.4 multimodal, eliminating the need for complex manual descriptions or support people. 5.4Mini Packs gives you impressive performance in just 3.8 billion parameters with 128K context window outperforming larger models on math encoding and increasing inference speed by 30%.
They also give you new stability AI models including disabled diffusion 3.5 large ultra and Core. This lets you generate all those fun images and creepy AI pictures.
Cohere enhanced retrieval expansion with Cohere rerank 3.5 and they now have support for the GPT 4.0 family expansion with audio and real time preview. Plus you get new customization tools like distillation workflows, reinforcement, fine tuning and fine tuning for Mistral as well as support for bring your own vnet for AI agent interactions and Magma Multi agent goal management architecture via Foundry Labs. So quite a few things in Foundry.
[00:43:14] Speaker C: Yeah, a lot of really useful enhancements in there.
[00:43:19] Speaker D: Feel like they were like let's see how many announcements and how many model updates can we do at once.
[00:43:26] Speaker C: I do like the idea of those mini packs because I think that that's. I'm more interested in that side versus the the GPT 4.5 model. Like because I think that you know you can have these giant mega models with all the information in them. But I mean maybe it's just my usage of AI is pretty simplistic too. But you know their example of you know, being able to sort of take you know, different sets of information where it be visual text and then come up with a like a repair program like that is, you know, like that's the use cases I'm more interested in versus giant things. That's kind of neat.
[00:44:04] Speaker A: A simple man requires a simple lm.
[00:44:06] Speaker C: Yeah, exactly.
[00:44:10] Speaker A: I like those five models mostly because they can run on LM Studio my Mac locally as well, which I quite enjoy just because it's nice to not always have to go to the Internet for everything I need to do.
Well if you're fine tuning your models like those GPT 4.5 or 4.0 models that you're using through Azure AI Foundry, you can now make your agents behave and speak the way you like them to with scaled up rag apps. Now customers want it snappier and more responsive and luckily with the OpenAI service offering in preview Provision deployments for fine tuned models gives your application predictable performance and predictable cost. Provision throughput allows you to purchase capacity in terms of performance need instead of per token. With fine tuned deployments it replaces both the hosting fee and token based billing of standard and global standard with throughput based capacity unit called ptus. If you're already using provisioned throughput units with base models, they work identical and fine tuned models and are completely interchangeable. The two models you can provision deployments for currently in the preview are GPT4O and GPT 4.0 mini in North Central US and Switzerland with more regions coming in the future.
If you want another region, you should definitely click on this blog post and click the Submit a request button to make sure that that region you care about is included in the GA of this feature later on. Yeah, it was a little interesting going through it and I was like I thought they had this and then it was like oh no, they were just reviewing provision throughput for the base model.
But now you get it for the fine tuned version of the base model which is. I assume this will GA pretty quickly. I can't imagine this is a huge departure from the base model throughput versus a tuned model. Throughput is just really where the source data comes from.
[00:45:45] Speaker C: Yeah, I'd love to see the pricing model move to this in term instead of the per token because it's really difficult to do adapt integration and then you know, beforehand understand your token usage. You're sort of in react mode.
[00:45:58] Speaker D: Well that's the problem is when you deploy your, you know, a new app with a new thing, you're like okay, do I do provision? Do I hit my limits? You know and in Azure and definitely some of the smaller regions or other regions than the primary ones of like North Central East US to you know those ones. You can hit those limits pretty easily and all of a sudden then you get token limits or other errors that occur. So it's like do you provision it pay up front until you figure it out, or do you risk a new feature of your app potentially having issues when you initially launch it? Here's what sharp edge do you want to run into and how fast you want your CFO yelling at you or your customer yelling at you? Choose which one.
[00:46:46] Speaker A: It's a tough one.
Which one do I want to be yelled at by more customer? CFO customer? Yeah, I don't know if I can answer that.
[00:46:59] Speaker C: Trick question. You're going to be yelled at both by both Every time?
[00:47:04] Speaker D: Well, yeah, because if your customers yell at you, they don't renew and then your CFO yells at you that you're the cause of it or you bankrupt the company and the customer stole it no longer exists. Which you like in trouble for.
[00:47:17] Speaker A: Yeah, perfect. Good. I'm looking forward to this.
[00:47:20] Speaker D: I really like our day jobs, just jobs in general, really. We chose a great field where we're like, okay, let's see how many people can get like this.
[00:47:28] Speaker C: Yeah, where's that Google bot?
[00:47:29] Speaker A: I think I have some questions to ask him. The dreamer.
I need a career in anything but this.
[00:47:36] Speaker D: Something where I don't get yelled at.
[00:47:38] Speaker A: I do, I do sometimes wonder like, you know, if you could, if you could send a message back to, you know, 15 year old you, like, would you, you know, because like, would you tell him to follow the career path that he thought he wanted to do? Like, I don't know, like, maybe I would have sold him to find something else.
[00:47:54] Speaker C: Goat farmer.
[00:47:55] Speaker A: Yeah.
You know what, you'd enjoy this much more. Like, trust me. Like, you might think you like computers, but it's just a world of pain because you, you go up through support and then you come a system admin and then a cloud guy and all downhill.
[00:48:11] Speaker C: You have to grow a beard. There's all kinds of, you know, requirements.
[00:48:14] Speaker A: You lose your hair.
There's a lot to it.
All right. Microsoft has launched Microsoft Fabric Quotas, a new feature designed to control resource governance for the acquisition of your Microsoft fabric capabilities. Fabric quotas aim to help customers ensure that fabric resources are used efficiently and help manage the overall performance and reliability of the Azure platform while preventing misus. Microsoft Fabric is a comprehensive service, of course, that offers advanced analytical solutions through the multiple workloads, all available in a single SaaS capacity model. Fabric is available to you in three SKUS fabric free trial, which is a time bounded per user trial. Imagine that Power BI Premium Office sold offers available as total subscriptions apparently don't still sell, but if you have them, you're good. And then fabric capacities which are Azure pay as you go offers available in multiple SKUs. The fabric quota limits the number of capacity units a customer can provision across multiple capacities in a single subscription. And the quota is calculated based on the subscription plan type and the Azure region available. So basically, stop stealing all my capacity. I have a quota.
[00:49:14] Speaker C: I mean I still, I'm struggling to understand the whole fabric thing because it's, you know, it's, it's, it's sort of, you know, the, the BI tool on top of a data set. But then I don't understand the, the, the pay as you go model and how you would provision anything like this, like in terms of, I guess it's computational queries that you would put into like a dashboard or something.
[00:49:38] Speaker D: Well, you can also like share data between, you know, here's my data set. Pay me as you access it, like S3 as it for like, you know, the consumer pays for it. Like it's kind of expanded fabric into like all these different ways.
[00:49:54] Speaker C: Okay, so fabric is much more than just sort of the analytics tool on.
[00:49:58] Speaker A: Top of the storage set. Okay. Oh yeah, yeah. So I mean like, yeah, I was gonna say like if you ever look at the diagram of all things it has, you know, basically it sits on top of Microsoft Purview, which is their DLP solution with one lake. So that's where the data sits. But then the data is then accessed by a bunch of different applications. The data factory, Data Engineering tooling, Data warehouse, Data Science tooling, Real time intelligence databases, industry solutions, and Power bi, all available to you on that fabric. And as part of the unified platform they give you security and governance, all the computing, storage and all of the business model for billing out of that as well. So you can build usage of this thing. So it's a pretty comprehensive solution and it competes with Snowflake. That's really what it's for.
[00:50:45] Speaker C: Okay, that helps actually help me understand here.
[00:50:48] Speaker D: Hold on.
Google's thing. What is Microsoft Fabric? Microsoft Fabric is a comprehensive cloud based data analytics platform that provides unified environments for users to ingest, store, process, analyze data from various sources. Combining from different Azure data sets like Power bi, Data Factory, Synapse into a single platform for seamless data management and analysis, all while centralizing. Oops, sorry. I had to scroll data storage in a multi cloud Data Lake called OneLake with the embedded AI capabilities for streaming, for streamlining, insight generation, semicolon essentially acting as an all in one solution for data professionals and business users to collaborate on data projects across the entire data lifecycle.
[00:51:35] Speaker C: All right, so let me translate.
[00:51:37] Speaker D: So that's one workload.
[00:51:39] Speaker C: Your workload has become so big that your database team can't handle the database. And the resources you need are so expensive that there's a comprehensive solution that you can use that just will burn all the money in the world.
[00:51:54] Speaker D: Yes.
[00:51:55] Speaker A: Yep.
[00:51:56] Speaker D: Like that whole thing I read is one sentence, it's one period.
[00:52:01] Speaker A: And then like in Amazon and gcp, it's multiple tools that you had to buy separately like bigQuery and then looker On GCP over in Amazon world, it's redshift and whatever that reporting situation.
Thank you, Quicksight. Why haven't they not Q reports? Come on guys.
[00:52:21] Speaker C: Oh, they're gonna.
[00:52:24] Speaker D: Yeah, well, quicksight's all built differently because Quicksight's built through like you know, per user logins and all disasters. It's clearly they never like fully integrated in with the aws. Like setup.
[00:52:39] Speaker A: Yeah, yeah. I mean I just asked Claude, I was like, is my fabric on terms of Snowflake? It's like, yes.
And then it gives you a breakdown of like how it's slightly different. So it does say that fabric offers an end to end platform. Integrating Microsoft's data tools or. Snowflake is primarily focused on being a cloud data warehouse, which is true because you don't get a reporting interface on Snowflake out of the box like Power bi. Gotcha. But basically, if you're Azure Shop and you're looking at Snowflake, you should take a pretty serious look at fabric.
Or you could do like us and buy Snowflake and then Power BI and pay for both.
[00:53:17] Speaker D: And your CFO yells at you again.
[00:53:19] Speaker A: Not me, not on that one. No, that one's not mine.
Azure SQL Database the modern cloud database relational database services announcing the general availability of.
Sorry, general availability of availability metrics for Azure SQL DBA enabling you to monitor SLA compliance availability. The Azure monitor metric is a 1 minute frequency storing up to 93 days and typically the latency display availability is less than 3 minutes. So you can know in a 93 day 3 minute period if your database was down for a minute or basically a minute. You can visualize the metric in Azure Monitor and set up alerts too. Availability is determined based on the database being operational. For connections, a minute is considered downtime or unavailable for a given database. If all continuous attempts to buy the customer to establish a connection to the database within the minute fail. And I have to say why?
[00:54:09] Speaker D: Because so stable.
[00:54:11] Speaker A: Because if my database is down because I can't connect to it for a minute, all of my app has failed. I don't know that I need you to tell me that your availability was a miss because I think I know from other reasons personally but some customer somewhere must have just been like Microsoft, you have to tell us how available your database is. You promised this SLA and you don't give us a way to measure it. And that's bs and that's why this feature exists. And that's the only reason why this feature exists. Because no one needs this Unless you are being super pedantic on database availability to the SLA that Azure provides to you.
[00:54:44] Speaker C: So I have one valid use case which is a conditional alarm.
[00:54:47] Speaker A: Right.
[00:54:48] Speaker C: Like so it's, it's if, if the. Like you said, if the database is down, the whole app is broken.
[00:54:53] Speaker A: Right.
[00:54:53] Speaker C: So you don't want to have every alarm in the world go just the database alarm. Fix that. Everything else will come back.
[00:55:00] Speaker D: Did you guys look at the FAQ though? How is a downtime minute boundary aligned? The aligned wall clock time for example 9 colon 00 or 9 colon 01. I like how they had to specify how the minutes were specified.
[00:55:19] Speaker A: That's very much left of them. That's fantastic.
[00:55:22] Speaker C: The lawyers had some product insight in that one.
[00:55:27] Speaker D: You can get the metric via sports private link. Guys, I love this. And the availability works for named replicas. The question is, does it work for geo. Well, GEO replies all worked for name wrap clause. Yeah, that makes sense.
[00:55:40] Speaker A: Yeah, I love this one here too. What data does the availability metric show in the case of an outage? When no connection activity to the database is observed, the available metric will show 100%. Wait, but if it's down because I can't make a connection availability. So zero when only failed connections are observed the entire minute, apparently.
[00:55:57] Speaker D: Well, does it support the read only replica business critical tier?
[00:56:01] Speaker A: It doesn't.
[00:56:02] Speaker D: Do you have a read only replica but you can't tell me if it's available or not?
[00:56:07] Speaker A: Correct.
[00:56:07] Speaker D: All of our read only replicas queries which are most likely what's shown in an app. You know, if it's a read intensive app, doesn't work for you.
[00:56:16] Speaker A: Look, and look, this is, this is a customer complained really loud feature and those will be future enhancements. Just, you know, we could talk about on future episodes perhaps.
[00:56:26] Speaker D: I just want to know if it went through a private preview and a public preview before they ga'da.
[00:56:31] Speaker A: I mean I'm much more cynical.
[00:56:33] Speaker C: I think they had to pay out a huge SLA and had no metrics to sort of reduce that risk to.
[00:56:38] Speaker A: Prove they actually achieved it. And the customer's like no you didn't and we're gonna charge you. Yeah, maybe that's what it is.
[00:56:43] Speaker D: I didn't realize it's a 49 SLA for a sequel. It's interesting.
[00:56:48] Speaker A: Yeah, because you'd never achieve that from them.
So maybe you want to set this up, Matt, Maybe you might need this one.
[00:56:55] Speaker D: Not commenting on why I stated that.
[00:56:59] Speaker A: Oh, we found the guy who had the huge credit required.
You're the one who built this feature for. I knew it. Matt.
[00:57:09] Speaker D: There we go.
[00:57:10] Speaker A: That's the story.
All right. Azure is announcing the general availability of Native Window Principles and Azure SQL Managed Instances, which. This broke my brain a little bit. This capability allows the migration of Azure SQL managed instances and unblocks the migration of legacy applications tied to Windows logins. This feature is crucial to the SQL Managed Instance link. And while the Managed Instance link facilitates near real time data replication between a SQL Server and Azure SQL managed instances, the read only replica in the cloud restricts the creation of Microsoft Entre principles. And with this new feature, you have three authentication modes for SQL Managed Microsoft Entre, which is the default. This mode allows authenticating Entre users using Microsoft Entre user metadata, which makes sense Paired, which is what everyone loves, as SA is the default mode for SQL Server Auth, then Windows, which is this new mode. And this mode allows authenticating Microsoft entry users using the Windows user metadata within SQL Managed instances. So we're basically doing a mapping from Microsoft Entre to Windows user metadata to then basically map to a metadata of SQL Server Auth, which happens behind the scenes. So it's just turtles all the way down.
[00:58:21] Speaker C: Why would you want that? You've got your identity in Entre and it's somehow different.
[00:58:26] Speaker A: Your legacy application, on premise, you're trying to move it to Azure and because it was built in 2002, it doesn't know what Entre is. And so it needs you to use Windows user metadata and you would be unable to move that workload to the cloud.
[00:58:40] Speaker C: I get it. Oh yeah. Why would you just update the app? You're right.
[00:58:43] Speaker A: Yeah. Why would you do such a thing? Yeah.
[00:58:46] Speaker D: And not all of Azure SQL, not to be confused with Azure Management SQL Instances supports everything. So for example, you can't run, you're running ssrs, you can't run on Azure SQL, you have to run on Managed instances. There's even Microsoft apps that can't support us support the full blown Azure SQL too.
[00:59:06] Speaker C: Have I ever mentioned that the user management of Microsoft SQL Server is just the most abhorrent sin that I can think of?
[00:59:15] Speaker D: Because I have feelings about this that I will not share because this podcast would never end.
[00:59:23] Speaker C: Oh my God.
[00:59:25] Speaker A: Yeah, this one, this one hurt my brain a bit. I had to work through how this works. I was like, okay, wait, Windows User Metadata Server.
And like I was having this conversation actually when I was in India because someone was asking me about like, why are we going to use Azure AD in our other environment? And I'm like no.
And they're like, I don't understand how that works. I'm like, I don't understand how you work because Azure ad, why does that have to do with you just set up a server and you install Active Directory. Why would you use Azure AD for this purpose in this new isolated environment that's highly secure and doesn't have allow access to Azure AD without pulling it into a boundary?
Which is as much as I can say about it.
I was just like Azure ad, which is now Microsoft entre, I think is probably going to destroy all of my preconceived notions of ad. And I don't like it because I don't understand.
[01:00:17] Speaker C: I hope it destroys all the preconceived nations of Active Directory because I mean.
[01:00:21] Speaker A: We'Ll make them better. Let's be clear. It'll. I'll have new hatred.
[01:00:24] Speaker C: It's just going to change them to other bad. You're right. Yeah, yeah, yeah.
[01:00:27] Speaker A: Let's not be. That's not fundamentally Azure is a flawed concept for cloud. So Microsoft just trying to keep shoving that down our throats because that's the only thing they understand how to do for Windows off, but it's still a problem.
[01:00:38] Speaker D: So they are to their credit adding more managed ID AKA like IAM roles for authentication between stuff. To their credit, I have seen in the last two years a bunch of things that now support it.
[01:00:54] Speaker A: How much have you had to deal with Windows gmsa?
[01:00:59] Speaker D: Zero loaded question.
[01:01:01] Speaker A: I'm sure that's why you think that's a good thing because as soon as you understand Windows gmsa, you're not going to like what you just said, so I'll let you work that one out.
[01:01:09] Speaker D: No, no, I'm good. I'm just gonna go with ignorance is bliss.
[01:01:13] Speaker A: Yeah.
[01:01:15] Speaker D: As I'm slowly typing Windows GMSA like.
[01:01:17] Speaker A: You want, you want, you want your server credentials to automatically rotate. Well, you need to create a GSMSA for that.
[01:01:23] Speaker D: Oh, that. No, no, no, no, no, no. I don't do that. That terrifies me.
[01:01:28] Speaker A: Yeah, yeah, it should.
[01:01:31] Speaker C: I've seen, I've made that recommendation to like three different data teams and they've all looked at me like I was some alien from another planet.
[01:01:37] Speaker A: Yeah, you screw three heads. They that's good then our last announcement is that my beloved Cloud 3.7 is now available to me and GitHub Copilot for visual Studio Microsoft. Coming up last in the release of who's supporting Cloud 3.7. So thanks Microsoft, which runs on AWS Bedrock. Yep. We talked about this last week that the cloud 3.5 did. So I'm sure the 3.7 does as well.
And that is it for another fantastic week here in the cloud, gentlemen.
It was a journey. We had to get there. Some Microsoft stuff. It was rough.
I had to spend some money on Oracle through Terraform and had to talk about Skype dying and Amazon bastardizing their naming convention. So, yeah, it's a typical cloud week, actually.
[01:02:24] Speaker D: Yeah, I was just saying, what's so different than normal?
[01:02:27] Speaker C: And you stayed awake through your jet lag, so kudos.
[01:02:29] Speaker A: I did, I did.
[01:02:30] Speaker D: It was all that Iam conversation, I think early on.
[01:02:34] Speaker C: Really ate yourself blood.
[01:02:35] Speaker A: Really got me going. Yeah.
[01:02:37] Speaker D: Yeah.
[01:02:38] Speaker A: Well, you know, if you don't like the show, just, you know, tweet at us or Macedon at us or Blue sky at us. You know, it's 11 and we'll know that you didn't like the episode, so let us know.
[01:02:47] Speaker D: But this is why we need Jonathan here. Let's be honest.
[01:02:50] Speaker A: Yeah, for sure he'll be back soon.
[01:02:52] Speaker C: Hopefully he'll tell us he doesn't like the episode.
[01:02:56] Speaker A: He tells us exactly what he thinks when he listens to it. He has to listen to it first, though, which never happens. But when he does, you hear about.
[01:03:02] Speaker D: It when we know we actually listened to it. Because he's like, nope, you did this wrong and this wrong. And this is what you should have said.
[01:03:09] Speaker A: Yeah, yeah. And then we're like, well, would you like us to do follow up on that so you can talk about that? He's like, no.
All right, cool. Thanks. Thanks, Jonathan. Appreciate it. All right, gentlemen, we'll see you next week here in the Cloud.
[01:03:22] Speaker C: All right, bye, everybody.
[01:03:26] Speaker B: And that's all for this week in Cloud. We'd like to thank our sponsor, Archera. Be sure to click the link in our show notes to learn more about their services.
While you're at it, head over to our
[email protected] where you can subscribe to our newsletter, join our Slack community, send us your feedback and ask any questions you might have. Thanks for listening and we'll catch you on the next episode.
