dlt-saga 0.2.6

Config-driven data ingestion and historization framework built on dlt

dlt-saga

Config-driven data ingestion and historization framework, built on dlt.

Why dlt-saga?

dlt is an excellent Python library for building data pipelines. dlt-saga adds the operational layer that teams need to run dlt at scale:

What you get	How
Zero-code pipelines	Drop a YAML file in configs/ — no Python needed for common sources
SCD2 historization	write_disposition: append+historize turns any snapshot table into a full change history with _dlt_valid_from / _dlt_valid_to
dbt-style selectors	saga ingest --select "tag:daily,group:api" — union, intersection, glob patterns
Multi-environment profiles	profiles.yml with dev/prod targets, service account impersonation, per-environment datasets
Plugin architecture	Register custom sources and destinations via packages.yml or Python entry points — no framework fork needed
Cloud-agnostic	BigQuery today, Databricks and DuckDB included, more via plugins

If you are already using dlt directly and finding yourself re-implementing incremental state management, environment switching, or SCD2 transforms — dlt-saga is the config layer you are building.

Installation

pip install dlt-saga[bigquery]          # BigQuery
pip install dlt-saga[databricks,azure]  # Databricks on Azure
pip install dlt-saga                    # DuckDB only (no cloud dependencies)

Quick Start

# 1. Create and scaffold a project
mkdir my-pipelines && cd my-pipelines
saga init                               # prompts for destination and credentials

# 2. Authenticate to your destination (skip for DuckDB)
#    See: https://github.com/Glitni/dlt-saga/wiki/Getting-Started

# 3. List available pipelines
saga list

# 4. Run a pipeline
saga ingest --select "example__sample"

See the Getting Started guide for a full walkthrough, or browse example/ for a minimal runnable setup.

Local execution is the default. Use --orchestrate to fan out to parallel workers (requires orchestration: configured in saga_project.yml).

CLI Commands

All commands are subcommands under the saga entry point and share common options: --select, --verbose, --profile, --target.

Selectors (dbt-style)

Selectors filter which pipelines to run. They work across all commands.

Syntax	Meaning	Example
name	Exact pipeline name	--select google_sheets__my_pipeline
*glob*	Glob pattern	--select "*balance*"
tag:name	Filter by tag	--select "tag:daily" (schedule-aware — see Configuration → Scheduling tags)
group:name	Filter by source group	--select "group:google_sheets"
space-separated	UNION (OR)	--select "tag:daily group:filesystem"
comma-separated	INTERSECTION (AND)	--select "tag:daily,group:google_sheets"

Common Examples

# List pipelines
saga list                                        # All enabled pipelines
saga list --resource-type ingest                 # Ingest-enabled only
saga list --resource-type historize              # Historize-enabled only
saga list --select "tag:daily"                   # Filtered by tag

# Ingest
saga ingest --select "tag:daily"
saga ingest --select "group:api" --workers 8
saga ingest --full-refresh --select "my_pipeline"
saga ingest --select "group:api" --start-value-override "2026-01-01"  # Backfill

# Historize (SCD2)
saga historize --select "tag:daily"
saga historize --full-refresh --select "filesystem__*"

# Run (ingest + historize sequentially)
saga run --select "tag:daily"

# Update BigQuery access controls
saga update-access --select "group:google_sheets"

# Target a specific environment
saga ingest --target prod --select "tag:daily"   # production (with impersonation)

Adding a New Pipeline

Create a YAML config file in configs/<source_type>/ — that's it. The framework auto-discovers configs.

Supported source types out of the box: API, Database (PostgreSQL, MySQL, SQL Server, and more via ConnectorX), Filesystem (GCS, SFTP, local), Google Sheets, and SharePoint.

See the Pipeline Types guide for config examples for each source type, and the Configuration reference for all available fields.

Write Dispositions and Historize

The write_disposition field controls what operations are enabled for a pipeline:

Value	Ingest	Historize	Use Case
append	Yes	No	Raw event/log data
merge	Yes	No	Upsert on primary key
replace	Yes	No	Full refresh each run
append+historize	Yes	Yes	Snapshot → SCD2
historize	No	Yes	External data → SCD2

Historize transforms raw snapshot data into SCD2 tables with _dlt_valid_from, _dlt_valid_to, and _dlt_is_deleted columns. See the Historize guide for the full reference.

Community

• GitHub Issues — bug reports and feature requests
• GitHub Discussions — questions, ideas, show & tell
• Contributing guide — how to get involved
• dlt community — dlt Slack / Discord

Further Reading

• Getting Started — Full walkthrough: install, init, first pipeline
• Architecture — Three-layer design, plugin system, execution flow
• Pipeline Types — Config reference for API, Database, Filesystem, Sheets, SharePoint
• Configuration — Hierarchical config, all options reference
• Profiles — Multi-environment setup, service account impersonation
• Historize (SCD2) — Snapshot tables → slowly changing dimensions
• CLI Reference — All commands, flags, and the programmatic API
• Deployment — Orchestration, Cloud Run, worker setup
• Performance — Parallel execution, worker tuning, backfill
• Plugin Development — Custom sources, destinations, hooks

Origin

dlt-saga is derived from an internal data ingestion framework originally built by Glitni for Amedia, a leading Nordic media group, as the ingestion layer of Amedia's data platform. Amedia supported open-sourcing the project and continues to fund ongoing development through their partnership with Glitni, enabling the framework to be shared with the broader community.

Project Structure

dlt-saga/
├── dlt_saga/              # Main package
│   ├── cli.py            #   CLI entry point (saga command)
│   ├── pipelines/        #   Built-in source implementations
│   │   ├── api/          #     Generic REST API pipeline
│   │   ├── database/     #     Database source (ConnectorX)
│   │   ├── filesystem/   #     Filesystem / GCS source
│   │   ├── google_sheets/#     Google Sheets source
│   │   └── sharepoint/   #     SharePoint source
│   ├── historize/        #   SCD2 historization engine
│   ├── destinations/     #   Destination implementations
│   │   ├── bigquery/     #     BigQuery
│   │   └── duckdb/       #     DuckDB (local development)
│   ├── pipeline_config/  #   Config discovery and parsing
│   ├── schemas/          #   Bundled static schemas (dlt_common.json)
│   └── utility/          #   Shared utilities (CLI, naming, orchestration)
├── example/              # Minimal runnable consumer project (DuckDB)
├── wiki/                 # Documentation (synced to GitHub wiki)
└── .dlt/                 # dlt runtime config overrides