KOPYA NG SALIN 🔗
WEBVTT
00:00:00.239 --> 00:00:14.000
If there is any way of like maybe separating secrets per teams in Airflow, of course, you would need to create the concept of teams or like domains inside Airflow to be able to create a secret per team.
00:00:14.160 --> 00:00:23.199
So those like minor things that actually if you want to use the Airflow in a multi-tenant way, right now you would need to deal with these things yourself.
00:00:23.280 --> 00:00:28.000
And uh if it comes with the next release in the future ones, it would be very good.
00:00:30.079 --> 00:00:38.000
You're listening to the Data Flowcast, where we unlock the secrets of mastering airflow for data engineering in Teya.
00:00:39.039 --> 00:00:42.719
Hello everyone, welcome, welcome to the Data Flowcast.
00:00:42.799 --> 00:00:50.719
I'm your host, Mark Marty, and today I'm joined by Alexandre Manu Limba Martins, senior data engineer at Teya.
00:00:50.960 --> 00:00:54.560
And the first thing that I would like to ask you, Alexandre, is how are you doing?
00:00:54.799 --> 00:00:55.840
Hey Mark, how are you?
00:00:56.000 --> 00:01:05.920
Yeah, I'm very happy to be here actually to talk a little bit about what we did in Teya and I don't know, help foster the Airflow community be a little bit.
00:01:06.239 --> 00:01:07.280
So very happy.
00:01:07.599 --> 00:01:08.000
Awesome.
00:01:08.079 --> 00:01:14.159
I'm very excited because I know that you did a couple of medium blog posts that are truly amazing.
00:01:14.239 --> 00:01:21.519
And I can't recommend everyone enough to check out those articles because they're truly amazing to learn how to handle Airflow.
00:01:21.760 --> 00:01:24.159
We'll talk about that later in the episode, right?
00:01:24.480 --> 00:01:32.159
So just before getting started, I would like to briefly explain the primary takeaways that you will get from this episode.
00:01:32.480 --> 00:01:37.120
So you will be able to you will know how to make airflow self-serve for users.
00:01:37.280 --> 00:01:39.519
We'll talk a little bit about that.
00:01:39.760 --> 00:01:43.359
Also, how to generate DAGs based on dbt models.
00:01:43.519 --> 00:01:47.280
I'm sure that there are many dbt users that are listening to us.
00:01:47.519 --> 00:01:53.599
And last but not least, some useful tips around optimization and executors.
00:01:54.239 --> 00:02:01.359
Just can you actually start by talking a little bit about your role at Teya and what you do?
00:02:01.760 --> 00:02:02.400
Okay, cool.
00:02:02.560 --> 00:02:09.520
I'm a data engineer at Teya, so I'm more like responsible to build a data platform.
00:02:10.319 --> 00:02:30.000
Actually, so what we are uh calling nowadays a data platform engineer, basically responsible for deploying and like maintaining all our data infrastructure components, so we're related to the extraction of data, storage, and also transformation of data as well.
00:02:30.400 --> 00:02:30.719
Okay.
00:02:31.120 --> 00:02:40.159
And just to have a better idea of how
Load full transcript
you run Airflow in your organization, like what kind of use cases do you need to manage today?
00:02:40.400 --> 00:02:43.360
The Airflow is our primary orchestrator, right?
00:02:43.520 --> 00:02:53.439
Our main use case, I would say, it's CBT, the orchestration of the DBT project that really powers all the analytics downstream.
00:02:53.599 --> 00:03:09.680
But also we have a couple of DAX running that is part of our machine learning uh initiative, fraud as well, so fraud prevention, and some custom like housekeeping workloads as well that we use to run in Airflow, but the primary use case would be DBT.
00:03:10.000 --> 00:03:10.879
DBT, okay.
00:03:10.960 --> 00:03:11.759
That's super interesting.
00:03:11.919 --> 00:03:15.759
And how many teams are working on Airflow in your organization?
00:03:15.919 --> 00:03:18.000
Do you have many different teams working on it?
00:03:18.319 --> 00:03:25.120
Yeah, currently it's being used by the data teams or the teams that relate to data somehow.
00:03:25.199 --> 00:03:25.360
Okay.
00:03:25.520 --> 00:03:28.319
So it's not just the data engineering teams.
00:03:28.479 --> 00:03:32.800
You also have like data scientists, machine learning engineers also using it.
00:03:33.120 --> 00:03:42.240
We want in the future to actually expand this usage to the other engineering teams inside the e-content.
00:03:42.319 --> 00:03:51.680
That would be your traditional software engineer that has a use case for some sort of like workload orchestration.
00:03:51.840 --> 00:03:52.800
But we're not there yet.
00:03:52.960 --> 00:03:54.560
That's a plan for the future.
00:03:54.879 --> 00:03:55.199
Okay.
00:03:55.280 --> 00:03:56.080
Yeah, it takes time.
00:03:56.159 --> 00:03:56.719
It takes time.
00:03:56.879 --> 00:04:03.280
And actually, speaking of that, is there anything that you have achieved or implemented with Airflow that you are really proud of?
00:04:03.520 --> 00:04:10.159
Because, as you said, you have many different teams and at least like profiles that work on your Airflow instance.
00:04:10.479 --> 00:04:13.360
So I know that you are working on something pretty, pretty great.
00:04:13.520 --> 00:04:14.800
Can you tell us more about that?
00:04:15.120 --> 00:04:29.920
Yeah, so I would say that for us, the most exciting thing would be the self-service fashion that we did the whole thing, that we did the whole platform, especially when we talk about our main use case, which is DBT, right?
00:04:30.079 --> 00:04:35.680
We have a DAG generation pipeline that builds on top of the DBT project.
00:04:36.000 --> 00:05:04.560
So when data analysts or even project managers that know a little bit of SQL and can build DBT models, they can see their models being translated into Airflow DAGs, go into Airflow and run those models as Airflow DAGs without requiring writing the DAGs per se, like without requiring knowing what an Airflow DAG is, right?
00:05:04.720 --> 00:05:08.959
Besides like how to we trigger it, how to clear
a task.
00:05:09.040 --> 00:05:19.279
So this is also this interaction with the Airflow UI is very important, but they don't need to know like how to actually write a DAG, they don't need to even know Python.
00:05:19.600 --> 00:05:20.000
Wow.
00:05:20.240 --> 00:05:33.360
So actually, I truly need to know more about that because I know that DAG generation is something that many users struggle with or even wonder how to uh implement that in their uh company.
00:05:33.600 --> 00:05:37.439
So can you tell us more a little bit about how you generate those DAGs?
00:05:37.600 --> 00:05:44.800
For example, if I'm a data scientist and I know, as you said, I don't I don't know how to write a DAG in Airflow.
00:05:44.959 --> 00:05:47.040
What should I do with your uh solution?
00:05:47.519 --> 00:05:54.079
So, yeah, our data generation, our DAG generation pipelines more focused on the dbt project.
00:05:54.319 --> 00:06:03.839
For the other use cases, like data scientists, machine learning engineers, they are usually more comfortable with Python with the concept of YA DAG.
00:06:04.000 --> 00:06:15.279
So they can generate DAGs freely as they basically want to, but of course, we also enforce some guidelines in DAG development in our CICD flows.
00:06:15.439 --> 00:06:24.240
It needs to be tags, it needs to be the the DAG needs to have the appropriate tags, a description, the access, the control layers as well.
00:06:24.399 --> 00:06:33.360
So we enforce through CI CD checks a little bit of minimum things that the DAG really needs to have.
00:06:33.600 --> 00:06:39.680
But when we talk about the DBT project, we have an entirely automated flow of tag generation, right?
00:06:39.759 --> 00:06:55.360
So we basically read off from the project, we parse the dbt manifest, and we have a separate file that we created that lives inside the dbt project where people can like register their models to be deployed.
00:06:55.439 --> 00:06:59.279
So they can actually group models however they want, basically.
00:06:59.519 --> 00:07:08.720
You can, if you want to deploy a single model, so like a single model becomes a DAG, or you can deploy how many models you want inside a DAG.
00:07:08.879 --> 00:07:14.160
So you can like slice and dice your dbt project in any way you like, basically.
00:07:14.399 --> 00:07:20.160
What usually uh people do is they separate the project for domains, right?
00:07:20.240 --> 00:07:24.879
So each DAG, each domain will have its own set of DAGs.
00:07:24.959 --> 00:07:26.879
What we call model groups.
00:07:27.040 --> 00:07:35.120
It's like usually models that need to be run and scheduled together, like that they make sense as a single unit.
00:07:35.279 --> 00:07:40.879
And all the models inside this model group is going to run as a single DAG in Airflow.
00:07:41.199 --> 00:07:45.680
And inside this DAG, it will follow the normal DBT lineage.
00:07:45.839 --> 00:07:54.879
So you actually the tasks inside the DAG will follow the same dbt lineage for model the execution.
00:07:55.120 --> 00:08:03.439
And we also did a thing that Cosmos did as well, which is to uh split run and test into set into separate tasks.
00:08:03.519 --> 00:08:07.439
So every model will run followed by tests.
00:08:07.759 --> 00:08:08.079
Okay.
00:08:08.240 --> 00:08:22.879
So basically, if I'm a data analyst, I actually don't interact with DAGs directly, I just interact with a file to say that this dbt model should be exactly generally like should be generated through a DAG, and that's all I do.
00:08:23.120 --> 00:08:23.519
That's it.
00:08:23.600 --> 00:08:37.200
And you you also on that file pass like the access control layers and tags and like whatever you want to have a more friendly the experience when like I searching for this DAG in Airflow.
00:08:37.360 --> 00:08:48.639
So people can like you actually write the DAG description, you apply tags, and they can determine which roles have the access to view and uh trigger this DAG.
00:08:48.960 --> 00:08:52.000
So all in this deployment file that we call it.
00:08:52.320 --> 00:08:53.279
And this is super interesting.
00:08:53.519 --> 00:08:55.840
This deployment file, what it is it's a YAML file?
00:08:55.919 --> 00:08:56.960
Is it a YAML file?
00:08:57.039 --> 00:08:57.600
Okay, here we go.
00:08:57.679 --> 00:09:04.320
And so you have a script that goes over those deployment files, or is it a DAG that goes over those deployment files?
00:09:04.639 --> 00:09:05.679
Is it a script?
00:09:05.840 --> 00:09:09.600
Is it actually part of our DAG generation project, right?
00:09:09.679 --> 00:09:17.279
Because this is one thing as well that we needed to consider when I starting the project, like how to generate DAGs.
00:09:17.440 --> 00:09:23.919
And then of course we have those two methods, which is the single file method or the like multiple file methods, right?
00:09:24.159 --> 00:09:46.559
So we started with the single file method, but then we it didn't work very well because as the dbt project like was growing and more and more models were added to actually parse the dbt project at your runtime, it was like your flow was like struggling to do that because the manifest gets like bigger every time we need to read the manifest.
00:09:46.960 --> 00:09:55.759
So we actually changed that to a multiple file method where the DAGs are generated before pushing it to Gearflow.
00:09:55.919 --> 00:10:00.159
So DAGs are basically generated in CICD after merge.
00:10:00.320 --> 00:10:18.639
So after the checks are like applied and we basically validate, then we generate all DAGs based off from the DT project, and then we push them to a bucket where basically Gearflow is just going to read as a normal DAG, right?
00:10:18.799 --> 00:10:21.519
So we are using this multiple file methods.
00:10:21.759 --> 00:10:33.120
That's really amazing because so you actually saw that using the multi-file method was delivered a better performance than if you were using the single
file method.
00:10:33.360 --> 00:10:36.320
You saw that in actually, how did you see that it didn't work?
00:10:36.559 --> 00:10:38.240
Did you see that in in your metrics?
00:10:39.120 --> 00:10:48.159
Yes, yeah, we were monitoring some airflow metrics, especially the bag parse time, like and the scheduler loop time.
00:10:48.399 --> 00:11:14.799
So we saw that the scheduler, the average scheduler loop time was like increasing a lot as the dbt project was growing, just because like the manifest was like getting larger, and like in every so like basically every time that the a flow it parses DAG, it had to run this uh script that generates all the DAG, so you had to parse it all over again.
00:11:14.960 --> 00:11:19.200
And when we changed to the multiple file method, the scheduler loop time just yellow.
00:11:19.440 --> 00:11:22.960
Wow, and the scheduler itself became more performed.
00:11:23.279 --> 00:11:23.519
Wow.
00:11:23.600 --> 00:11:35.279
Out of curiosity, did you try to trick some airflow settings first before going to this method, or did you just decide to use the multifile method?
00:11:35.919 --> 00:11:40.559
Yeah, one option was to use the separate DAG parser, right?
00:11:40.639 --> 00:11:46.799
That the Airflow has to not parse the DAGs using the uh scheduler, but to have a separate component for that.
00:11:47.039 --> 00:11:55.279
That could be one option, but we decided to go straight to the multiple file methods because that was the recommended method for this case.
00:11:55.440 --> 00:11:58.080
Yeah, we just went uh straight with it.
00:11:58.320 --> 00:11:58.639
Okay.
00:11:59.919 --> 00:12:08.960
This episode is brought to you by Astronomer, creators of Astro, the leading fully managed airflow service designed for the modern data stack.
00:12:09.120 --> 00:12:20.799
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.
00:12:21.039 --> 00:12:28.000
To see how you can maximize airflow's potential within your data ecosystem, visit www.astronomer.io.
00:12:28.879 --> 00:12:36.080
That's amazing to hear because again, some people advocate for the single file method, some people advocate for the other method.
00:12:36.320 --> 00:12:45.440
At the end of the day, truly, and as you exactly, as you said, you need to look at your metrics, test different things, and see what works best for you and your use case.
00:12:45.600 --> 00:12:47.120
So that's really good to hear.
00:12:47.279 --> 00:13:00.399
And just to circle back on the data on the DBT models, I imagine that when you have the DBT models, you had some challenges right along the way to generate those DAGs based on the DBT models.
00:13:00.559 --> 00:13:07.519
I know that you have an interesting story about the executor, and more specifically, the Kubernetes executor.
00:13:07.600 --> 00:13:13.519
So tell you, can you tell us more about
the Kubernetes executor, which is an extremely popular executor?
00:13:13.840 --> 00:13:18.240
But in your case, why it might not have been the best option?
00:13:18.559 --> 00:13:19.279
Yeah, sure.
00:13:19.360 --> 00:13:21.200
That was a fun one, actually.
00:13:21.919 --> 00:13:22.639
Oh yeah.
00:13:22.879 --> 00:13:26.559
So because like here we run Airflow on Kubernetes, right?
00:13:26.720 --> 00:13:36.799
So of course, our first option would was like to yeah, let's start by using the Kubernetes executor, as we are running in Kubernetes.
00:13:37.120 --> 00:13:39.840
It's the most logical sense, logical choice.
00:13:40.000 --> 00:13:54.399
And then we started doing that, but what happened was that as our dbt project became larger, we started to have like many tasks because like we orchestrate like more or less on average 2,000 models.
00:13:54.960 --> 00:13:59.600
So at our peak, we had like over 3,000 models.
00:13:59.679 --> 00:14:02.879
So you can imagine the number of tasks.
00:14:03.039 --> 00:14:09.600
And in if you use the Kubernetes executor, one task equals one pod, right?
00:14:09.840 --> 00:14:17.120
So you can imagine the number of pods that we were uh scheduling per day in our Kubernetes cluster.
00:14:17.279 --> 00:14:19.519
That's not necessarily the problem.
00:14:19.679 --> 00:14:26.960
The problem was that our dbt models usually run very fast, they are like usually very short-lived uh tasks, right?
00:14:27.120 --> 00:14:31.200
They run for one minute, two minutes, and they are done.
00:14:31.519 --> 00:14:39.360
And we have some peak times that like group large groups of models need to actually be scheduled in that same time.
00:14:39.600 --> 00:14:50.720
So, what happened was that like in some specific moments, we had a large, large spike in the number of tasks, thus in the number of pods, right?
00:14:50.879 --> 00:15:01.039
So our Kubernetes cluster that was running with I don't know, like 100 pods, actually it blew up to a thousand pods in a matter of five minutes.
00:15:01.279 --> 00:15:12.720
And then what happens is that we also, the cloud, the engineering team in our company, they deploy a tool, which is called uh eCarpenter, which is Kubernetes resource management tool.
00:15:12.879 --> 00:15:30.960
So basically, what iCarpenter does is that it finds Kubernetes nodes that are yonder utilized and like reschedule pods on those nodes to actually lower the total the amount of nodes, saving you cost like at the end of the day, right?
00:15:31.360 --> 00:15:50.399
And as our tasks were very fast, we could see that we were like scaling the number of nodes in the cluster, but like very quickly, those nodes were becoming sub-utilized because all the pods scheduled on those nodes were like being completed.
00:15:50.559 --> 00:15:55.519
However, some jobs took more than one minute or uh two minutes, right?
00:15:55.679 --> 00:15:57.519
Some jobs were like long-living jobs.
00:15:57.679 --> 00:16:08.639
And what happens that iCarpenter, when trying to deschedule those nodes, were like killing those jobs in favor of like, I know let me schedule this pod in like the other node.
00:16:08.720 --> 00:16:12.000
But it was a running BBT job.
00:16:12.240 --> 00:16:13.039
It was a problem.
00:16:13.360 --> 00:16:23.360
It was a problem because we started to have like many jobs being killed because of these spikes in the number of pods and thus in the number of cluster nodes as well.
00:16:23.679 --> 00:16:27.679
And then to solve that, we decided to migrate to the salary executor.
00:16:27.759 --> 00:16:36.960
So in salary, we don't have this problem because we have like long-living worker nodes, and then they can just like accept new tasks, right?
00:16:37.279 --> 00:16:40.480
And those tasks can run very fast.
00:16:40.720 --> 00:16:43.360
We didn't have this problem when we migrate to salary.
00:16:43.600 --> 00:16:45.279
That's a super interesting story.
00:16:45.519 --> 00:16:52.879
And actually, so that's a very good point of why the salary executor is most of the time better than the Kubernetes executor.
00:16:52.960 --> 00:16:58.080
It's because it can run your tasks faster, especially if you have short-living tasks.
00:16:58.240 --> 00:17:01.120
It's much better to use the salary executor in this case.
00:17:01.360 --> 00:17:12.160
But then also a downside of the salary executor is that you have some workers waiting, consuming resources while waiting, and so they are actually doing nothing.
00:17:12.400 --> 00:17:17.839
Are you thinking of have you ever thought of how to fix this maybe issue?
00:17:17.920 --> 00:17:23.839
Maybe it's not one yet, but have you ever thought of doing something to improve this downside, maybe?
00:17:24.079 --> 00:17:30.240
Yeah, one way to actually improve that is by using the uh KEDA, the autoscaler.
00:17:30.319 --> 00:17:37.200
Yeah, so by using KEDA, you can like uh scale in or out the number of like salary workers.
00:17:37.359 --> 00:17:44.319
So in periods that you don't have as many jobs running, you can run with like just one or maybe two workers.
00:17:44.559 --> 00:17:54.240
But like when you have like more and more uh tasks being scheduled, you can use that metric, which is the number of tasks in the queued state, right?
00:17:54.480 --> 00:18:02.079
So actually as your scale as your scale out metric in KEDA, so you like increase the number of workers.
00:18:02.240 --> 00:18:04.240
That's a very good way out as well.
00:18:04.480 --> 00:18:05.519
Yeah, no, I agree, I agree.
00:18:05.599 --> 00:18:07.279
Definitely Keda is one solution.
00:18:07.440 --> 00:18:16.160
Also, I I heard about I've never tried it yet, but I heard about the Celery Kubernetes executor, I believe, which is a mix of both worlds.
00:18:16.400 --> 00:18:18.160
I know that many people are looking into it.
00:18:18.240 --> 00:18:20.000
I don't know, did you look into it yet?
00:18:20.079 --> 00:18:26.000
Or it's on our backlog, but yeah, we want to do
that as well.
00:18:26.400 --> 00:18:36.480
That would require a little bit of investigation and testing as well, the side on the R side because it's uh a big change, but we know that it can be yet improved.
00:18:36.640 --> 00:18:54.079
Although, like right now, when we need to run very uh custom tasks that would require a very custom Python packages, usually related to data science or machine learning, we can actually and we do use the Kubernetes pod operator.
00:18:54.240 --> 00:19:01.039
So for those cases, we spin up a totally separate pod to have runtime isolation as well.
00:19:01.200 --> 00:19:02.720
So we also use it.
00:19:03.200 --> 00:19:05.119
Yeah, that's a pretty popular operator.
00:19:05.440 --> 00:19:19.839
Many actually, I don't know if you do that, but many companies have a layer on top of the Kubernetes pod operator so that this layer defines the amount of CPUs or memories and so on based on the use case.
00:19:19.920 --> 00:19:30.000
So you don't interact with, for example, as a data scientist, I'm not gonna interact with the Kubernetes pod operator directly, but I will interact with a custom operator built on top of it.
00:19:30.160 --> 00:19:32.640
So I don't have control over the resources and so on.
00:19:32.720 --> 00:19:35.359
Is it something that you are doing or not yet?
00:19:35.680 --> 00:19:36.640
That's a very good idea.
00:19:36.720 --> 00:19:44.880
We don't do that, but it makes a lot of sense because to create this layer of the abstraction right on top of the Kubernetes pod operator, it makes a lot of sense.
00:19:45.119 --> 00:19:55.440
We don't do that just because the use cases that we do need it, they're like very very custom, and the people that use it know what they are doing basically.
00:19:55.680 --> 00:19:58.079
They're like very comfortable with it.
00:19:58.240 --> 00:19:58.640
Yeah.
00:19:58.960 --> 00:19:59.680
Good for you.
00:20:00.319 --> 00:20:02.160
Good for you that they know what they do.
00:20:02.640 --> 00:20:09.680
Okay, just also very quickly, because obviously we talked about dbt and generating DAGs using dbt models.
00:20:09.839 --> 00:20:12.079
Why aren't you uh using Cosmos?
00:20:12.400 --> 00:20:14.319
That's a very fair question.
00:20:14.480 --> 00:20:18.799
So our project started like more than a little bit more than two years ago.
00:20:18.880 --> 00:20:27.279
So at that time, I don't think Cosmos was released, I can be wrong, but if it was released, it was not very uh popular.
00:20:27.440 --> 00:20:41.839
So we didn't even uh consider it as a choice, and as well, because as far as I know, like Cosmos doesn't have a way to actually deal with cross-dEG dependencies or cross-dig communication.
00:20:41.920 --> 00:20:47.359
You can split your DPT project into many DAGs using Cosmos as well.
00:20:47.519 --> 00:20:52.079
However, I don't see how you can connect those DAGs.
00:20:52.319 --> 00:20:58.000
And that's a thing that was a requirement for us when we started building the
project that we are running right now.
00:20:58.240 --> 00:21:16.160
Because in our case, if one model from deck A references another model has a dependency on another model that is being scheduled on that B, we are going to create a sensor on that A to repoke for the latest state of the dependent model in that B.
00:21:16.319 --> 00:21:18.400
So we do this activity with sensors.
00:21:18.720 --> 00:21:19.359
Okay, I see.
00:21:19.599 --> 00:21:20.240
That's good to know.
00:21:20.319 --> 00:21:27.519
And that might be uh something that the community will work on for Cosmos to make it better, because uh that's the kind of feedback that we love to hear.
00:21:27.759 --> 00:21:34.240
As a final question, what would you most like to see from the development of the Airflow project in the near future?
00:21:34.400 --> 00:21:41.920
Like, is there any any any feature that you would like to see in Airflow in the next release of Airflow?
00:21:42.240 --> 00:21:46.480
Anything that you would like to see as a new feature or an improvement that will make your life better?
00:21:46.799 --> 00:21:57.440
Yeah, like for us, everything that would improve the ability of using airflow as more like a multi-attendant airflow, right?
00:21:57.680 --> 00:22:03.839
More use cases related to more teams using the same Airflow instance.
00:22:04.079 --> 00:22:05.119
That would be very nice.
00:22:05.279 --> 00:22:12.079
And ways to deal with it, like one very specific example is the usage of uh secrets.
00:22:12.400 --> 00:22:26.559
If there is any way of like maybe separating secrets per teams in Airflow, of course, you would need to create the concept of teams or like domains inside Airflow to be able to create a secret per team.
00:22:26.799 --> 00:22:40.960
So those like minor things that actually if you want to use the Airflow in a multi-tenant way, right now you would need to deal with these things yourself, and uh if it comes with the next release in the future ones, it would be very good.
00:22:41.200 --> 00:22:41.519
Agreed.
00:22:41.599 --> 00:22:46.319
Definitely multi-tenancy has been a topic for a while in Airflow, to say the least.
00:22:46.559 --> 00:22:47.359
Okay, awesome.
00:22:47.519 --> 00:22:49.279
Thank you for joining, Alexandre.
00:22:49.359 --> 00:22:52.559
Honestly, it was I keep saying that, but it was truly amazing.
00:22:52.720 --> 00:22:58.319
And uh, what's the best way for folks to get in touch with you if I want to stay in touch with you?
00:22:58.559 --> 00:23:02.079
Yeah, I guess LinkedIn is would be the best way.
00:23:02.160 --> 00:23:03.519
Yeah, it's just okay.
00:23:03.839 --> 00:23:06.400
Type my name there, you can maybe find it.
00:23:06.480 --> 00:23:07.039
I don't know.
00:23:07.119 --> 00:23:09.599
Yeah, and LinkedIn would be good.
00:23:09.839 --> 00:23:10.160
Awesome.
00:23:10.480 --> 00:23:11.039
Thank you again.
00:23:11.200 --> 00:23:13.599
Thank you for joining the data flowcast.
00:23:13.759 --> 00:23:14.960
Again, it was amazing.
00:23:15.119 --> 00:23:18.000
We a lot of insights, and thank you so much.
00:23:18.240 --> 00:23:21.519
Hope you are uh gonna have a great day and hope to see you soon.
00:23:22.400 --> 00:23:24.400
I enjoy it very much as well.
00:23:24.640 --> 00:23:25.279
Thank you, Mark.
00:23:25.519 --> 00:23:26.079
Thank you, Alexa.
00:23:26.240 --> 00:23:26.880
Bye-bye.
00:23:28.079 --> 00:23:34.480
Thanks for joining us for the latest episode of the Data Flowcast, mastering airflow for data engineering and AI.
00:23:34.640 --> 00:23:38.319
You'll find links in the show notes to any resources mentioned in today's show.
00:23:38.480 --> 00:23:45.200
If you're enjoying our podcast, please subscribe so you never miss an episode and leave us a five-star review.