Latest / The Data Flowcast: Mastering Apache Airflow ® for Data Engineering and AI / The Future of Airflow Telemetry with Bolke de Bruin

Transcript

The Future of Airflow Telemetry with Bolke de Bruin · The Data Flowcast: Mastering Apache Airflow ® for Data Engineering and AI

• 0:00
  SPEAKER_02: Telemetry is about having the right data to be able to steer the project into a particular direction.
• 0:06
  SPEAKER_02: Airflow is very complex, has a lot of different mechanics going on, a lot of different gears, but sometimes certain things get outdated, people are not using it anymore.
• 0:14
  SPEAKER_02: We might still put in the effort to maintain these kinds of things, but we're limited in resources too.
• 0:19
  SPEAKER_02: So we obviously like to know where people are using it.
• 0:23
  SPEAKER_02: You just stay relevant for the users.
• 0:24
  SPEAKER_02: It is an important factor, not just for companies deploying it like Astronomer or Google or AWS, these kind of things.
• 1:08
  SPEAKER_01: Don't miss your chance to connect with the Airflow community and get the inside scoop on what's next.
• 1:13
  SPEAKER_01: Register now.
• 1:14
  SPEAKER_01: The link is in the show notes.
• 1:18
  SPEAKER_01: Hi everyone.
• 1:19
  SPEAKER_01: Welcome to the Data Flowcast.
• 1:21
  SPEAKER_01: I'm your host, Kenton Danis, and today I'm joined by Bolka DeBron, who is a director at Metis and one of the Airflow PMC members.
• 1:30
  SPEAKER_01: And it sounds like I did get your name mostly right.
• 1:32
  SPEAKER_01: So that's good.
• 1:33
  SPEAKER_01: We're off to a good start.
• 1:34
  SPEAKER_02: Excellent, y'all.
• 1:35
  SPEAKER_02: Excellent.
• 1:35
  SPEAKER_02: I'm really happy about that.
• 1:37
  SPEAKER_02: People actually know me now.
• 1:38
  SPEAKER_02: Here it goes.
• 1:39
  SPEAKER_01: Okay, excellent.
• 1:40
  SPEAKER_01: I passed for a like not terrible American Dutch accent.
• 1:43
  SPEAKER_02: So no, it was pretty good.
• 1:45
  SPEAKER_02: I'm happy.
• 1:46
  SPEAKER_02: I'm happy with it.
• 1:46
  SPEAKER_02: I will listen to my name this way.
• 1:48
  SPEAKER_01: So it's really great to have you here and to see you again.
• 1:51
  SPEAKER_01: For our listeners for today's podcast, we're going to discuss telemetry and airflow, an exciting topic.
• 1:57
  SPEAKER_01: So we'll cover how it's been implemented in the past and how the project is moving forward.
• 2:02
  SPEAKER_01: But before we dive into that, I want to start with a little bit of background on you because you've been very involved in the Airflow project for a long time.
• 2:10
  SPEAKER_01: So maybe you can tell me more about what you do and how you got involved in Airflow and what your role is now.
• 2:16
  SPEAKER_02: Ooh, yeah, that's that's a story, obviously, that has been told quite some time before now, I think, quite regularly.
• 2:22
  SPEAKER_02: I'll keep that brief.
• 2:24
  SPEAKER_02: I started out actually somewhere in 2015-ish when Maxime started out with Airflow, and we joined pretty quickly when I was still working at the bank at the time.
• 2:33
  SPEAKER_02: Uh, one of my data scientists, when we were looking at orchestrators, said, I said, Hey, you should have a look at Airflow.
• 2:38
  SPEAKER_02: I was looking at Luigi.
• 2:40
  SPEAKER_02: That's old times this one.
• 2:42
  SPEAKER_02: And I looked exactly, I looked at uh Airflow and said, Oh, yeah, it looks great, but all the security stuff is missing.
• 2:49
  SPEAKER_02: And for a bank that I really want quite, so I started adding that, and I actually did my first Python code ever, and Maxime just merged it just like that, and I was like completely surprised.
• 2:58
  SPEAKER_02: What I don't think I ever told Maxime is that we didn't even deploy it in production ourselves yet before Airbnb dead dead, and Airbnb was definitely running at a different scale than we did in this bank with just 10 people using some parts of it.
• 3:14
  SPEAKER_02: So, Maxime, if you listen to it, I'm really happy that you were our beta tester for this.
• 3:19
  SPEAKER_02: Here we go.
• 3:20
  SPEAKER_02: And from then on, I helped out pulling this through the Apache Software Foundation, particularly actually because the for a bank in this case it was important to have a different kind of backer with it and that it was vetted like that.
• 3:32
  SPEAKER_02: This is, I think, where people or companies dealing with the Linux Software Foundation as well, and and of the Linux Foundation and the Apache Software Foundation, is how they perceive that there's a quality aspect to it and also some risk mitigation aspect to it if there's continuation after the project is maybe abandoned, for example.
• 3:50
  SPEAKER_02: So I did that and I was really involved until somewhere 2018, and then also life gets a little bit in the way, and other people get involved, and companies get involved, and and people that have exciting ideas beyond that to get involved.
• 4:04
  SPEAKER_02: So I stayed a little bit on the background.
• 4:06
  SPEAKER_02: I always, when it's a hobby, I like it.
• 4:08
  SPEAKER_02: When it comes very closer to being a job, then I start like feeling itches that that's not entirely the same thing anymore.
• 4:14
  SPEAKER_02: Yeah, happy that what I'm doing now and getting a little bit involved with the very sexy subject of telemetry, I would say.
• 4:22
  SPEAKER_02: There is a sexiness to it, but uh, how you introduce it is like a very exciting subject.
• 4:26
  SPEAKER_02: I don't know if people look at it that way, but to a certain extent, it is.
• 4:30
  SPEAKER_02: So here we go.
• 4:32
  SPEAKER_01: I think it's very exciting to the people who maintain the project.
• 4:36
  SPEAKER_01: So we'll we'll get into that.
• 4:37
  SPEAKER_01: But I uh thanks for giving that background.
• 4:40
  SPEAKER_01: We're going way back in Airflow history.
• 4:42
  SPEAKER_02: So way back.
• 4:43
  SPEAKER_01: I love that.
• 4:44
  SPEAKER_01: But as you said, we do want to talk about telemetry today.
• 4:47
  SPEAKER_01: And this is, I know, one of the things that you've been working on recently in Airflow.
• 4:51
  SPEAKER_01: So maybe backing up a bit for those who might not know what is telemetry in the context of something like Airflow?
• 4:58
  SPEAKER_01: What is it used for?
• 5:00
  SPEAKER_02: So, what we do or what telemetry is about having the right data to be able to steer the project into a particular direction, right?
• 5:08
  SPEAKER_02: Airflow is very complex, has a lot of different mechanics going on, a lot of different gears, but and sometimes certain things get outdated, people are not using it anymore.
• 5:17
  SPEAKER_02: We might still put in the effort to maintain these kind of things, but we're limited in resources too.
• 5:23
  SPEAKER_02: So we obviously like to know where people are using it.
• 5:27
  SPEAKER_02: Used to stay relevant for the users is a is an important factor, not just for for for companies deploying it like Astronomer or or Google or AWS, these kind of things, but also for our users, because in the end, we're putting effort in a project that we'd like people to use because that brings new new contributors in and uh new flows there, and would like to know where we're if we're doing the right things basically.
• 5:51
  SPEAKER_02: At the complexity level of Airflow, it becomes like flying blind because if you don't assign get that kind of data in somehow, and while you could go for user surveys on a user list, there is always bias involved then because those are actively using users, and they might be a bit conservative too, right?
• 6:11
  SPEAKER_02: That might not be the bleeding edge user, so you would like to understand that a little bit better.
• 6:16
  SPEAKER_02: So the those are the those are the major challenges or needs for having having telemetry in.
• 6:24
  SPEAKER_02: Nevertheless, we understand that comes with, let's say, especially in the era of data privacy, these kind of things, people start looking at with a magnifying glass on what you're actually doing at the moment.
• 6:36
  SPEAKER_02: So here you go.
• 6:37
  SPEAKER_02: It's a bit of a why.
• 6:39
  SPEAKER_01: Yeah, that's very helpful background.
• 6:42
  SPEAKER_01: I my team talks a lot about the airflow survey, and we help to run that every year, and it's obviously a very useful tool in helping choose like what features are developed.
• 6:49
  SPEAKER_01: But to your point, it's one pretty limited in scale.
• 6:52
  SPEAKER_01: Last year we had more than 5,000 people fill it out, which was amazing.
• 6:56
  SPEAKER_01: But that's still a drop in the bucket of total airflow users around the world.
• 7:00
  SPEAKER_01: And to your point, about there's probably some bias in the people who are taking the time to fill that out and what answers they're giving.
• 7:06
  SPEAKER_01: So having a more automated way to collect information about how airflow is being used can really help drive the project in the right direction.
• 7:14
  SPEAKER_02: Absolutely.
• 7:14
  SPEAKER_02: Yeah, absolutely.
• 7:16
  SPEAKER_02: And maybe going into why now we are having this kind of conversation, just building a bridge there, is we did actually collect telemetry for a while.
• 7:28
  SPEAKER_02: And we did this through a tool called Scarf, which was deployed by the actual Apache Software Foundation, and there were things in place there.
• 7:37
  SPEAKER_02: However, I think that we lost a bit of trust with the community in the way we've done that because we did it as an opt-out scenario, because we did think, and it's true in a way, obviously, if you make this an opt-in scenario, then people won't turn it on, and that makes it really hard to get people convinced to do that actually, because people in the end will just deploy Airflow and they won't look at those options to turn it on, so it's really hard to get a sufficient body that again makes sense because otherwise you get that kind of bias again into the data, which you'd like to prevent, obviously.
• 8:12
  SPEAKER_02: Nevertheless, I mean I think one of the things we've learned there is that actually people's jobs are on the line, and so people were deploying Airflow and suddenly found out that we were the Airflow was sending out telemetry, and maybe people didn't look at it the proper way or just saw outgoing connections and and people got challenged on it.
• 8:31
  SPEAKER_02: Hey, why is this actually happening?
• 8:33
  SPEAKER_02: And people got actually sometimes a little bit scared.
• 8:35
  SPEAKER_02: This is obviously what we don't like.
• 8:38
  SPEAKER_02: We like people's jobs, we like people that use airflow actually to get a job rather than the opposite way.
• 8:44
  SPEAKER_02: Here, I think we should definitely do a better job in either, and there are several options for it, and these are the scenarios we have to go through and also bring our community along with us on hey, obviously, why are we collecting our uh telemetry?
• 9:02
  SPEAKER_02: Can we, in what scenario, when can we not?
• 9:05
  SPEAKER_02: How do you turn it on?
• 9:06
  SPEAKER_02: How do you turn it off?
• 9:07
  SPEAKER_02: But also just saying, okay, be very transparent about the telemetry that we are actually trying to collect, and and always always give the autonomy of the decision basically to the one deploying it in a very transparent way to the uh to anyone involved.
• 9:26
  SPEAKER_02: So there's no surprises when you're doing this kind of thing.
• 9:29
  SPEAKER_02: That's definitely what we don't like because then trust us out of the window and it will come back very slowly.
• 9:38
  SPEAKER_00: This episode is brought to you by Astronomer, creators of Astro, the leading fully managed airflow service designed for the modern data stack.
• 9:47
  SPEAKER_00: Whether you're orchestrating large language models, implementing a data mesh architecture, or streamlining complex workflows, Astro empowers your team to focus on innovation, not infrastructure.
• 9:59
  SPEAKER_00: To see how you can maximize airflows potential within your data ecosystem, visit www.astonomer.io.
• 10:07
  SPEAKER_01: I imagine that was very surprising to a lot of users to find out, especially when I think people are busy and they don't always pay attention to what's going on the dev list or in release notes.
• 10:18
  SPEAKER_01: So it would be very easy to not see an update like that.
• 10:21
  SPEAKER_01: And with huge companies, you mentioned working at a bank, like security for a tool like Airflow when it's used at companies like that is really important.
• 10:28
  SPEAKER_01: So, what where is the project at now?
• 10:32
  SPEAKER_01: Kind of what are you looking at to move this forward in a way that is more transparent, Johannes?
• 10:38
  SPEAKER_02: So, one to be absolutely clear, we're in the current releases of Airflow, we are not collecting any telemetry.
• 10:45
  SPEAKER_02: So there's no data being sent anywhere at this moment, so you can safely deploy Airflow without being afraid of your job.
• 10:53
  SPEAKER_02: Just to be clear.
• 10:54
  SPEAKER_02: What we would like to go to is to go to a situation where we can actually collect certain information, but we need to build up the trust within the community.
• 11:04
  SPEAKER_02: So this needs to be a community over technology approach, which is well, coincidentally, and I know Yadek will love me for it, saying community over code, but really the Apache way kind of thing.
• 11:16
  SPEAKER_02: But uh, and absolutely in this case, it is completely true.
• 11:20
  SPEAKER_02: So we need to build up the trust with the community that we're that if we are going to be collecting any telemetry, and we need to be able to accept a no here and saying, okay, this is too complex to execute, and people will not allow us to do it in the way that is workable for us, then then then it's basically building a platform over time that we say, Hey, we would like to do this, we will put it out on the user list, for example, by having this talk with you now.
• 11:51
  SPEAKER_02: Here is being starting to be even more sharing.
• 11:54
  SPEAKER_02: Okay, this is our intention, this is what we like to do with it.
• 11:58
  SPEAKER_02: This is maybe also how you get access to the results of it.
• 12:01
  SPEAKER_02: Because, hey, if we're assembling telemetry, why wouldn't we give out the results of that?
• 12:07
  SPEAKER_02: So that will take time, basically.
• 12:09
  SPEAKER_02: That's step number one.
• 12:11
  SPEAKER_02: So the actually, this is probably step number one, really.
• 12:14
  SPEAKER_02: Step number one.
• 12:15
  SPEAKER_02: So it's starting to say, Hey, we would like to collect telemetry.
• 12:19
  SPEAKER_02: What type of telemetry can we collect?
• 12:22
  SPEAKER_02: Can we collect the amount of DAX that you're running?
• 12:24
  SPEAKER_02: Can we say, okay, which providers are we are you using?
• 12:28
  SPEAKER_02: By sending us data in one way or another, if that's in automated fashion, we will see IP addresses which can be and can be privacy sensitive.
• 12:37
  SPEAKER_02: We would like to disregard them right away, but obviously, because that connection happens, the data exists at that moment when it's there.
• 12:45
  SPEAKER_02: So, what kind of guardrails do we need to put in place to ensure that the community has a trust in what we do?
• 12:53
  SPEAKER_02: Where does it need to take place?
• 12:54
  SPEAKER_02: A scarf because it was a third-party, uh, third-party contractor, basically for the software foundation, wasn't fully trusted, and with the combination of the surprise that was for people was in inside Airflow earlier, that didn't that didn't connect really well, particularly for collecting this kind of information.
• 13:18
  SPEAKER_02: We probably need to keep it in our own hands somehow.
• 13:22
  SPEAKER_02: But in a foundation way where a lot of companies are working together and individuals are working together, what's our hands is a bit of a question to you.
• 13:31
  SPEAKER_02: Most likely somewhere in the foundation, somehow.
• 13:34
  SPEAKER_02: But that's uh though those things are dilemmas we need to deal with over time.
• 13:39
  SPEAKER_02: And I hope to have this a regional wall standard and somewhere around the Airflow Summit, which will happen in your hometown, uh, Seattle October.
• 13:49
  SPEAKER_01: Yeah, it will.
• 13:50
  SPEAKER_01: Promo for the Airflow Summit if you haven't already registered, please.
• 13:53
  SPEAKER_01: Because I would love to see everybody there in Seattle this October.
• 13:57
  SPEAKER_01: But yeah, that's really interesting.
• 13:59
  SPEAKER_01: And so you're looking at this from kind of two different angles, one being the community aspect of implementation, and how can you do that in a way that makes people comfortable and that they have input and know what's going on.
• 14:11
  SPEAKER_01: And it sounds like the other is on the technical side.
• 14:14
  SPEAKER_01: So you're looking at not using like a third-party tool that you would integrate, but instead having the community or the developers who work on Airflow contribute it to the project itself.
• 14:27
  SPEAKER_01: Is that right?
• 14:28
  SPEAKER_02: Nope, not necessarily in that way.
• 14:30
  SPEAKER_02: There's it's more like what is the trusted party where we can collect that information, and that's what it's not a real technical question.
• 14:36
  SPEAKER_02: This is really a trust question where the foundation is probably a little bit more trusted than a commercial company, because hey, those those privacy statements might actually change over time, something like that, because that was what triggered us also looking at the Scarf privacy statement.
• 14:53
  SPEAKER_02: They basically said, Yeah, we can just change our privacy statement the next day.
• 14:57
  SPEAKER_02: That's yeah, okay.
• 14:58
  SPEAKER_02: That's stuff that won't fly, obviously, in in the context that we are then deploying airflow in.
• 15:05
  SPEAKER_02: Yeah, not it's truly not a technical question.
• 15:08
  SPEAKER_02: We can record all the providers that are there and send it over.
• 15:12
  SPEAKER_02: It's not the hardest thing to do.
• 15:14
  SPEAKER_02: It is really okay.
• 15:15
  SPEAKER_02: Do people actually trust us to send that information somewhere?
• 15:19
  SPEAKER_02: And what is it?
• 15:20
  SPEAKER_02: Does it go to a fault or something like that?
• 15:22
  SPEAKER_02: Or do we do it like a very you know, in a uh they say Bitcoin is anonymized, it isn't, but do it like in a zero trust tree kind of uh thing?
• 15:29
  SPEAKER_02: Is it that much?
• 15:30
  SPEAKER_02: Is that much anonymization required?
• 15:33
  SPEAKER_02: What level of collecting information is acceptable?
• 15:37
  SPEAKER_02: There have been talks about hey, we can show where contributors are coming from or where users of Airflow are geographically.
• 15:44
  SPEAKER_02: But that can be sensitive, right?
• 15:46
  SPEAKER_02: That's that might pinpoint individuals in certain locations where they might not want to expose that kind of information.
• 15:53
  SPEAKER_02: They'd like to probably have that in their own hands, and you need to be very transparent about that and making sure that they have the option to say no or yes in the right way.
• 16:05
  SPEAKER_01: Yeah, absolutely, totally makes sense.
• 16:07
  SPEAKER_01: And yeah, it's a tricky subject, definitely a walk between maintaining that trust and and as you said before.
• 16:14
  SPEAKER_01: So helpful for the project, too, because I think about all of the big features that have come out recently, including in Airflow 3, we're very community-driven, and Airflow really takes seriously.
• 16:25
  SPEAKER_01: I think all of the PMC and the contributors and committers like take very seriously what the community is saying that they want and need from the project, which is great, but they have to have enough information to work from.
• 16:36
  SPEAKER_02: Yes, exactly.
• 16:37
  SPEAKER_02: And then the challenge beyond that actually becomes at a certain point is if you are sharing or if you're not sharing, can you still get access to the outcomes of what community brings?
• 16:46
  SPEAKER_02: And the same question goes for growth, goes for AWS and any other commercial vendor of airflow.
• 16:52
  SPEAKER_02: So that's a next level dilemma which comes comes with it.
• 16:55
  SPEAKER_02: Um so these are yeah challenges that we need to overcome in order to provide the one, the best way of dealing with it from a from an ethical perspective, but then also be able to steer the direction of the steer the direction of the project, which you know you mentioned we have great new features in Air Force.
• 17:15
  SPEAKER_02: Are you actually being used?
• 17:16
  SPEAKER_02: Question mark.
• 17:18
  SPEAKER_02: Probably you some people know a little bit, but they know a little bit, and and we just get that from trusting each other and saying, yeah, absolutely being used.
• 17:26
  SPEAKER_02: We see millions of downloads going up, that that kind of thing.
• 17:30
  SPEAKER_02: That's a proxy for other kind of information, and we like to be be a bit closer to the actual usage there and the way it's being done because yeah, hey, we just like to cuddle our users quite a lot.
• 17:41
  SPEAKER_01: Yeah, yeah, absolutely.
• 17:44
  SPEAKER_01: Okay, I think that's a great place to round out this discussion.
• 17:47
  SPEAKER_01: I really hope to hear more about this in the future, including at Airflow Summit, and see yeah, any progress that has been made in a couple months.
• 17:55
  SPEAKER_02: I will I will do a talk at the Airflow Summit and then hopefully have some of those results being built up.
• 18:00
  SPEAKER_02: So this is a starting point.
• 18:01
  SPEAKER_02: Hopefully, Airflow Summit.
• 18:03
  SPEAKER_02: It's not it's not the end because it will be continuing in discussion, but then definitely we'll have a check-in moment again.
• 18:09
  SPEAKER_01: Yeah, yeah, love it.
• 18:11
  SPEAKER_01: I'm excited to see you there and another promo for the Airflow Summit for anyone who wants to learn more about this and lots of other things.
• 18:17
  SPEAKER_01: Okay, I always end with the standard final question, and that is what would you most like to see from the Airflow project in the near future?
• 18:24
  SPEAKER_01: Maybe aside from added telemetry.
• 18:26
  SPEAKER_02: Oh my god, oh my god.
• 18:28
  SPEAKER_02: Oh jeez, I was teasing, I was teasing Jarek with it saying, Hey, I told you about the Muse thing on uh just a minute ago before we before we started this one.
• 18:37
  SPEAKER_02: Yeah, yeah, I yeah, as a the role that I sometimes have within the Airflow project is you're writing reports to the Apache Shopper Foundation board on how the project is doing.
• 18:47
  SPEAKER_02: And I did this just after Airflow 3 was released, and I did it with a little bit of funny remarks, and we're really proud of it.
• 18:54
  SPEAKER_02: And then Jadek asked me, Hey, uh, what's next?
• 18:56
  SPEAKER_02: And I said, Oh, let Airflow 4.
• 18:58
  SPEAKER_02: And do you actually already have the things that you would like to see in Airflow 4?
• 19:03
  SPEAKER_02: When and honestly, you put me on the spot, but but but let's say more if we can get you like maybe two things, if I'm thinking about it now.
• 19:14
  SPEAKER_02: I would like to see declarative workflows being even better, etc.
• 19:18
  SPEAKER_02: We see more Databricks is doing it, and we see it in other areas as well.
• 19:23
  SPEAKER_02: It just feels a little bit more natural than maybe the imperative one.
• 19:27
  SPEAKER_02: And the second one is the the agentic, I guess, the popular version of the agentic workflows, because you'll just see that kind of assistance coming up in any of the other other orchestrators very quickly.
• 19:39
  SPEAKER_02: And I think that would definitely speed up anything with what Airflow is doing.
• 19:44
  SPEAKER_02: It is it's uh AI isn't going away.
• 19:47
  SPEAKER_02: You see that kind of speed ups are happening.
• 19:49
  SPEAKER_02: We can't deny it.
• 19:50
  SPEAKER_02: Let's then ride the wave and and make it a first-class citizen so that it works works out for a lot of people that then can still deploy Airflow and have their jobs being a bit accelerated with what they need to do.
• 20:03
  SPEAKER_01: Yeah, absolutely.
• 20:04
  SPEAKER_01: I love both of those answers, and I think that I may have heard that there's at least some stuff coming on that in 3.1 maybe later this summer.
• 20:13
  SPEAKER_01: Maybe a tea teaser.
• 20:15
  SPEAKER_01: I'm not entirely certain where what the status of that AIP is, but people can fact check me on that one.
• 20:20
  SPEAKER_01: But soon, I know it's I know it's top of mind for certainly astronomer, definitely a lot of others working on that.
• 20:25
  SPEAKER_02: I assume so.
• 20:26
  SPEAKER_01: Yeah, yeah.
• 20:27
  SPEAKER_01: Love that.
• 20:28
  SPEAKER_01: Okay, well, it was really great having you on.
• 20:30
  SPEAKER_01: I appreciate it.
• 20:31
  SPEAKER_01: What is the best way for folks to get in touch with you?
• 20:34
  SPEAKER_02: Yeah, they can reach me on the deaf list, but I'm not looking at it too often currently, but I will be more present there because of the telemetry stuff.
• 20:42
  SPEAKER_02: If you want to reach out to me.
• 20:44
  SPEAKER_02: And uh on a on the business site, you can reach me uh through bulker.debrun at metis.com, which is then my professional address.
• 20:53
  SPEAKER_02: LinkedIn is also fine.
• 20:55
  SPEAKER_02: This and just ping me a little bit, then then I'll uh probably get awake and and respond to your uh your inquiries.
• 21:01
  SPEAKER_01: Okay, we'll throw all of that in the show notes and I will add another plug for the dev list, which of course anybody can join, and it's a really great way to actually stay up to date with what's going on in the project.
• 21:12
  SPEAKER_01: And if you ever want to weigh in on development of new features or even jump in and help, that's a great way to it's part of how I keep up to date with everything.
• 21:19
  SPEAKER_01: If you want any of the inner workings of what's going on in Airflow, dev list is a good place to go.
• 21:24
  SPEAKER_02: It is a very nice place to be in.
• 21:28
  SPEAKER_01: All right, I love it.
• 21:29
  SPEAKER_01: Okay, thank you so much again.
• 21:30
  SPEAKER_01: It was really great having you on.
• 21:32
  SPEAKER_02: All right, thank you so much again.
• 21:36
  SPEAKER_00: Thanks for joining us for the latest episode of the Data Flowcast, mastering Airflow for data engineering and AI.
• 21:43
  SPEAKER_00: You'll find links in the show notes to any resources mentioned in today's show.
• 21:47
  SPEAKER_00: If you're enjoying our podcast, please subscribe so you never miss an episode and leave us a five star review.