text stringlengths 0 2.67k |
|---|
**Lili Cosic:** Happy birthday. |
**Justin Garrison:** Thank you. So Lily is on the show as a guest. Lili Cosic is a senior software engineer. Lily, what software are you responsible for, or have you been responsible for in the past? |
**Lili Cosic:** Sure. So I recently joined a month or so ago Influx Data, and right now I'm responsible for - or I'm learning to be responsible for - the cloud dedicated service. Essentially, it's the newest InfluxDB version three. So it's a columnar data store, and we run it in a managed, single-tenant cloud environme... |
**Justin Garrison:** One of the best things about open source is you can shift companies and be "Oh, you're doing the thing that I was already working on." |
**Lili Cosic:** Exactly. |
**Justin Garrison:** I'm very familiar with that. And I can keep contributing to the thing. But you just had a whole bunch in that initial "What is Influx?" that I want to ask more questions about. Because you said, a) it's single-tenant. So if I go and I sign up for a hosted Influx database, I get my own instance of i... |
**Lili Cosic:** It's both. So there's a serverless option for the v3, but this is a dedicated one, so you don't have noisy neighbors, and you control the costs, and essentially, we run it for you. And it's K3S -- but we also use Kubernetes for the management nodes, for example. And then we run it on a cloud provider, a... |
**Justin Garrison:** And the interesting part, that you you're doing this on top of a cloud provider, not using the cloud provider's managed solution for this, because you're carving up those VMs smaller with K3S... You're "We can run a bunch of Kubernetes clusters here." |
**Lili Cosic:** Yeah. Obviously, this entire architecture is the team's architecture. I just joined a month ago and I think they did a really great job, as it's extremely resilient. And what I really about it is that -- I haven't used K3S before in the past, but from the outside I can tell that it's not Kubernetes, obv... |
**Justin Garrison:** \[08:18\] SQLite. |
**Lili Cosic:** SQLite, that's the one. Yes, thank you. But yeah, I think they chose etcd, which was a really good choice, I think, as there's at least three nodes, essentially, always. It's the minimum size, I think. Again, I've been there a month, so... |
**Justin Garrison:** Yeah, no problem. |
**Lili Cosic:** If I make a mistake, it's on me. |
**Justin Garrison:** The other thing I wanted to ask about is you described Influx as a columnar store database, and I always describe it as a time-series database. Can you describe the difference, and why you describe it that way versus another? When I think of columnar, I'm thinking of Cassandra. And when I'm thinkin... |
**Lili Cosic:** So I think this third version is different than the previous version. So there was a version one, then there was a version two, and the third version, essentially - it's built from scratch, from the ground up, trying to maintain a really high cardinality... So yeah, time series is definitely an analogy ... |
**Justin Garrison:** I think I remember hearing or reading a blog post about Influx switching from Go to Rust as part of that rewrite, right? So that was part of the big "Not only are we shifting architecture, but we're shifting language completely", which is a huge task for a company to say "Hey, we want to go this ot... |
**Autumn Nash:** You see this...? First, he rubbed in that his 3D printer is there, and now he's \[unintelligible 00:10:22.09\] Rude. |
**Lili Cosic:** Yeah. But yeah, definitely. I think it's interesting... So the majority, I would say, of the companies is Rust. We do use Rust as well in our team, but writing operators and controllers in Rust is still a bit rusty; pun intended there. Sorry about that. |
**Autumn Nash:** That was great. |
**Lili Cosic:** You can do a terrible graphic, or something. |
**Autumn Nash:** It was so good, though. |
**Lili Cosic:** But we do use it for CLI tools, and we have been trying to -- in the beginning, the team was talking about how they did initially use Rust as a client to Kubernetes, and they had some problems... And it's just easier, because what we use is QBuilder to create the CRDs, and create the logic there, and th... |
**Autumn Nash:** Yeah. And there's a lot less people that know it. But I definitely think that Rust is gaining really great traction. |
**Lili Cosic:** Yeah, definitely. Yeah, I'm looking forward to doing some Rust... I have a small task in the next sprint, which is exciting. And I've been doing a tiny bit... I had to troubleshoot at some point some Containerd stuff and some Containerd runtimes, Nydus and stuff that, the snapshotter, so... Something wa... |
**Justin Garrison:** Now, how is Influx different, or how is the work you were doing on the Prometheus operator in that stuff? Because a lot of people will think of Prometheus as another sort of database time-series store of being able to scrape metrics and do this stuff. What is the difference there? How are you trans... |
**Lili Cosic:** \[12:06\] I would say that Influx is a really generic, or -- general purpose I think is maybe the better term. The database in Prometheus is really, really -- it's only suited for the specific time-series metrics format, as it comes with the time-series database already, DSDB already baked in. Obviously... |
And definitely, a lot of people use still Influx for long-term storage of metrics as well. There's a lot of customers that do that, and you can definitely have the entire stack and you won't even notice. But I think that for me, learning about Prometheus in the beginning - there's a lot of similarities. There's the wri... |
**Autumn Nash:** Could we maybe explain the use cases to each? Because I think a lot of -- like, when I first got into NoSQL, a lot of people didn't know what time-series databases do... And it is a really cool, unique use case. So maybe I think it would benefit the audience by maybe explaining what those two things do... |
**Lili Cosic:** So with Prometheus specifically, you primarily use it for anything that you measure. So anything that needs to measure at a certain interval, and then whatever you do with that data - whether you visualize it, whether you alert on it, or you collect analytics, any endpoint that can be scraped... And the... |
But in general, essentially, whatever has a time notion for it, to be able to write the data and then also query the data is much quicker if you have a time-based series database... Because being able to not having the notion of the columns, and having just the notion of "This is where I want to query from..." Because ... |
**Autumn Nash:** If I remember correctly, they're immutable too, right? |
**Lili Cosic:** Yes. |
**Autumn Nash:** So that way it can't be changed... Because I think it sometimes is used in certain production environments where they're used to audit, too. |
**Lili Cosic:** Yeah. That's an interesting use case. Yeah, definitely. With Prometheus you can do some stuff to delete the data, and there have been use cases for that... Definitely customers who don't fully, maybe don't understand Prometheus... We've had that at Red Hat. They demanded deletion... But because it saves... |
**Autumn Nash:** Yeah. Because isn't that the whole point, for it to be immutable, so that way you know all the data is there? That's interesting. See, I feel like a lot of times we try to explain databases, and people are like "I'm just going to use this", and you're like "Is that really the right database for that? B... |
**Justin Garrison:** I mean, sometimes going into it we don't know what the data is going to look like, or what the use case is going to be. And sometimes people just don't know that other databases exist for specific things, right? |
**Autumn Nash:** Yes... |
**Justin Garrison:** \[16:05\] And that's one of the hardest parts. My first experience with Influx was around 2017 or something we had this database that was for licensing for software. And anytime people were using certain software, it would report it, it would add a row of like "I'm using the software at this time",... |
And so we were doing that, we started querying it, and I loved the line protocol... Because I literally was testing this out with just curl statements. I'm just like "Oh, let me just curl this string... This works really well." But then I discovered the TICC stack, the full thing. Telegraph, Influx, Chronograph, and... |
**Lili Cosic:** Graphite, I think? Yeah... |
**Justin Garrison:** Capacitor. |
**Lili Cosic:** Capacitor. Okay. See, my onboarding does not go well, clearly... |
**Justin Garrison:** Well, these are old, because that's not what it does anymore. |
**Lili Cosic:** Yeah, I know. We've been through it, yeah. |
**Justin Garrison:** I don't know what's in V3 of Influx, but I know for V2 some of that got integrated directly in Influx, and so the TICC stack's not around anymore. But Telegraph was another one of those pieces where it's an agent that could report metrics out, and it was super-easy to extend, where it's just like "... |
**Lili Cosic:** Oh, no... |
**Justin Garrison:** And so people started noticing at the command line, they're like "You know what? Every once in a while, I get an error from my tool that I never used to get before." And it took us a couple of weeks where I'm like "Oh, you know what? That was me." |
**Autumn Nash:** It's crazy the amount of damage you can do to a database by connections... Because you can take a whole database down and people never think of the fact that your connections can take your database down until it's too late. |
**Justin Garrison:** And especially things that aren't consistent. It wasn't until later we were like "You know what?" |
**Lili Cosic:** Because it's on intervals, yeah. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.