Skip to content

Latest commit

 

History

History
87 lines (81 loc) · 3.65 KB

problems.md

File metadata and controls

87 lines (81 loc) · 3.65 KB

Why you want to consider Data Manager?

Here are the reasons:

Without Data Manager With Data Manager
  • My data is messy, I have lots of JSON, parquet files distributed in hundreds of AWS S3 buckets, it is hard for me to find the data when I need it
  • My ETL pipeline code load data with physical path directly, once I move the data, many pipelines got broken and I have to spend whole day to fix it.
  • Data Manager provides Data Catalog with UI and APIs. I can easily find the data I needed. I can go to the Datasets menu, find the dataset, and click the dataset, I can see all the instances of that dataset. (TODO) Use the search menu, I can search dataset by name.
  • With the data catalog Data Manager provides, your ETL pipeline can reference dataset by "logical name" instead of physical location, if the data moves the location, your pipeline stay unbroken.
  • The cost for building ETL pipeline is high. I have to contact Data Enginer to create a pipeline, test it and publish it, it usually take days to get it done.
  • We have bunch of data applications, each of them do a specific thing, document is not clear, only Data Engineer knows how to use it.
  • For launch the pipeline, I have to ask Data Engineer to manually update the airflow DAG code to adding the new pipeline, and set the schedule in the code.
  • ETL Pipeline creation is fully self serviceable. Now I just click the "Pipelines" menu, 95% of time, I can create pipeline using SparkSQL with this tool, and for the rest 5% rare cases, our Data Engineer write highly reusable Data Application to help me.
  • I can also create pipeline with mixed steps through web UI, some steps uses the SparkSQL, and some steps invokes the in-house Data Applications.
  • Data Manager creates airflow DAG for me based on the steps I entered in Web UI, Data Manager also allows me to set the schedule for the pipeline, so I can create our daily trading data processing pipeline easily, and launch it quickly.
  • Pipeline dependency is horrible. I set a airflow job to run a 3:00 PM daily just because I think the prior job will finish by 2:00 pm (and adding some safe buffer). If the prior job didn't finish, this job will fail since the data it requires is not there.
  • Data Manager manages the pipeline dependencies. When you define your pipeline, you need to declare the asserts it depends on, and the scehduler will only invoke the pipeline once the asserts it requires are all ready.

Data Manager Server

Data Applications

Feel free to contact me at [email protected] if you have any questions.