Glossary#

Cosmos#: An open-source Python package that allows you to write data transformations using dbt and then use Apache Airflow®’s orchestration to integrate dbt projects into end-to-end workflows.
DAG#: An Airflow term derived from the mathematical structure called Directed Acyclic Graph. A DAG provides a model that includes everything needed to execute an Airflow workflow. See DAG.
DbtDag#: A Cosmos class that wraps a full Airflow DAG around a dbt project, rendering each dbt node as an Airflow task. Use DbtDag when the entire Airflow DAG is dedicated to running a dbt project. See Cosmos core concepts.
DbtTaskGroup#: A Cosmos class that embeds a dbt project as an Airflow TaskGroup inside a larger DAG. Use DbtTaskGroup when dbt is one stage of a broader pipeline (e.g. ingestion → dbt → reporting). See Cosmos core concepts.
dbt deps#: A dbt command that pulls the most recent version of dependencies listed in your packages.yml file.
Callbacks#: User-defined functions that Cosmos invokes after the dbt command completes, during post-execution handling and before the temporary project directory is cleaned up. Useful for custom logging, alerting, or side effects without modifying the operator. Configure them on Cosmos operators via callback and callback_args; when using DbtDag or DbtTaskGroup, these can be passed through operator_args. See Callbacks.
ExecutionConfig#: The Cosmos class used to configure the execution mode and related runtime options such as the dbt executable path and invocation mode. See Execution Config.
ExecutionMode#: Describes where and how Cosmos runs dbt commands (e.g. LOCAL, WATCHER, DOCKER). Configured via ExecutionConfig. See Choose an execution mode.
Interceptors#: An optional list of callables (new in Cosmos 1.14) that run before Cosmos builds the dbt command for each task. Each callable receives (context, operator) and may modify operator.vars and operator.env; the modified values are then used when building and running the dbt command. Useful for injecting runtime variables or environment values per task run. Configured via operator_args={"interceptors": [...]} on DbtDag or DbtTaskGroup. See Operator arguments.
InvocationMode#: Controls how Cosmos calls dbt: DBT_RUNNER imports dbt as a Python library in the same process (no subprocess overhead, lower CPU and memory usage), while SUBPROCESS spawns a separate process (better isolation when dbt and Airflow share a Python environment). Running as a subprocess roughly doubles memory usage compared to DBT_RUNNER. See Invocation modes.
LoadMode#: The method Cosmos uses to parse your dbt project (e.g. DBT_MANIFEST, DBT_LS). Configured via RenderConfig. See Parsing Methods.
Manifest#: A dbt artifact that contains a full representation of the dbt project’s resources, including all node configurations and resource properties. See Manifest JSON file in the dbt docs.
Partial parsing#: A dbt feature, enabled in Cosmos since v1.4, that skips re-parsing files that have not changed. Each Airflow node performs a full dbt project parse only once; subsequent task runs reuse the cached partial_parse.msgpack artifact, reducing parse time and CPU usage. See Partial parsing.
Profile#: The authentication information used by dbt to connect to your data warehouse. See profile.
ProfileConfig#: The Cosmos class that determines which data warehouse Cosmos connects to when executing dbt SQL. See Connect to your database.
Profile mapping#: A Cosmos-provided resource that translates Airflow connections into dbt profiles. See Use a profile mapping.
profiles.yml#: The file where dbt stores connection information for each data warehouse connection. See Connect to your database, Use your own profiles.yml file, or The profiles.yml file in the dbt docs.
ProjectConfig#: The Cosmos class used to specify where your dbt project is located and any project variables that should be used for rendering and execution.
node_converters#: A RenderConfig option that accepts a dictionary mapping a DbtResourceType to a callable, allowing users to replace how specific node types are rendered as Airflow tasks. See Render Config.
RenderConfig#: The Cosmos class that controls how a dbt project is turned into an Airflow DAG or TaskGroup, including node selection, test behaviour, and source rendering. See Render Config.
SourceRenderingBehavior#: Controls whether dbt source nodes are rendered as Airflow tasks. NONE (default) skips source nodes entirely; ALL renders every source; WITH_TESTS_OR_FRESHNESS renders only sources that have tests or a freshness check defined. Configured via RenderConfig.
TestBehavior#: Controls when and how Cosmos runs dbt tests. AFTER_EACH (default) adds a test task after each model; AFTER_ALL runs all tests in a single task at the end; BUILD runs tests as part of dbt build inside the model task; NONE skips tests entirely. Configured via RenderConfig. See Testing behavior.