AWS Glue is a managed extract, transform, and load (ETL) service designed to make it easy for customers to prepare and load data for analytics. With it, users can create and run an ETL job in the AWS Management Console. Users point AWS Glue to data stored on AWS, and AWS Glue discovers data and stores the associated metadata (e.g. table definition and schema) in the AWS Glue Data Catalog. Once cataloged, data is immediately searchable, queryable, and available for ETL.
$0.44
billed per second, 1 minute minimum
dbt
Score 9.5 out of 10
N/A
dbt is an SQL development environment, developed by Fishtown Analytics, now known as dbt Labs. The vendor states that with dbt, analysts take ownership of the entire analytics engineering workflow, from writing data transformation code to deployment and documentation. dbt Core is distributed under the Apache 2.0 license, and paid Teams and Enterprise editions are available.
One of AWS Glue's most notable features that aid in the creation and transformation of data is its data catalog. Support, scheduling, and the automation of the data schema recognition make it superior to its competitors aside from that. It also integrates perfectly with other AWS tools. The main restriction may be integrated with systems outside of the AWS environment. It functions flawlessly with the current AWS services but not with other goods. Another potential restriction that comes to mind is that glue operates on a spark, which means the engineer needs to be conversant in the language.
If you can load your data first into your warehouse, dbt is excellent. It does the T(ransformation) part of ELT brilliantly but does not do the E(xtract) or L(oad) part. If you know SQL or your development team knows SQL, it's a framework and extension around that. So, it's easy to learn and easy to hire people with that technical skill (as opposed to specific Informatica, SnapLogic, etc. experience). dbt uses plain text files and integrates with GitHub. You can easily see the changes made between versions. In GUI-based UIs it was always hard to tell what someone had changed. Each "model" is essentially a "SELECT" statement. You never need to do a "CREATE TABLE" or "CREATE VIEW" - it's all done for you, leaving you to work on the business logic. Instead of saying "FROM specific_db.schema.table" you indicate "FROM ref('my_other_model')". It creates an internal dependency diagram you can view in a DAG. When you deploy, the dependencies work like magic in your various environments. They also have great documentation, an active slack community, training, and support. I like the enhancements they have been making and I believe they are headed in a good direction.
It is extremely fast, easy, and self-intuitive. Though it is a suite of services, it requires pretty less time to get control over it.
As it is a managed service, one need not take care of a lot of underlying details. The identification of data schema, code generation, customization, and orchestration of the different job components allows the developers to focus on the core business problem without worrying about infrastructure issues.
It is a pay-as-you-go service. So, there is no need to provide any capacity in advance. So, it makes scheduling much easier.
Slow load times of the dbt cloud environment (they're working on it via a new UI though)
More out-of-the-box solutions for managing procedures, functions, etc would be nice to have, but honestly, it's pretty easy to figure out how to adapt dbt macros
Amazon responds in good time once the ticket has been generated but needs to generate tickets frequent because very few sample codes are available, and it's not cover all the scenarios.
AWS Glue is a managed service. It was easier for us to integrate it into our stack since we are already an AWS shop. It saved us the headache of managing a 3rd part service.
Most ETL pipeline products have a T layer, but dbt just does it better. The transformation is on steroids compared to the others. Also, just allows much more Adhoc solutions for very specific projects. Those ETL tools are probably better on the T part if you don't need too many transforms - also dbt is pretty much free dependent on how you work it, also extremely scalable.
