dbt-ci

A CI tool for dbt (data build tool) projects that intelligently runs only modified models based on state comparison, supporting multiple execution environments including local, Docker, and dbt runners.

How It Works

dbt-ci uses a cache-based workflow:

1. init - Downloads reference state from cloud storage (or uses local), compares with current code, and creates a cache of changes
2. run/delete/ephemeral - Use the cached state automatically (no need to re-specify state paths)

This design ensures:

• ✅ Consistent state across all commands in a CI run
• ✅ Better performance (no redundant state downloads)
• ✅ Simpler CLI (specify state once in init, reuse everywhere)

Installation

From PyPI (Recommended)

pip install dbt-ci

Extras

The base install deliberately carries only dbt-core and the CLI plumbing. The cloud SDKs and the Docker client are large and most projects need at most one of them, so they are installed on demand:

pip install 'dbt-ci[gcp]'            # BigQuery + GCS
pip install 'dbt-ci[aws,docker]'     # S3 state, Docker runner
pip install 'dbt-ci[all]'            # everything

If a feature needs an extra you haven't installed, dbt-ci says which one and how to install it rather than failing with an import traceback.

Supported dbt versions

dbt-ci needs Python 3.12 or newer. CI tests Python 3.12 and 3.13 on every dbt-core version above, and Python 3.14 on dbt-core 1.12, the first release that supports it.

pip install dbt-ci resolves the newest dbt-core 1.x (>=1.10.13,<2.0.0). To stay on an older minor version, pin it alongside dbt-ci, e.g. pip install dbt-ci 'dbt-core~=1.11.0'.

The published Docker images are tagged per dbt-core version (dbt-1.10.23, dbt-1.11.16, dbt-1.12.6); latest tracks the newest supported minor version.

From GitHub

# Install from main branch
pip install git+https://github.com/datablock-dev/dbt-ci.git@main

# Install a specific version
pip install git+https://github.com/datablock-dev/dbt-ci.git@v1.0.0

Local Development

git clone https://github.com/datablock-dev/dbt-ci.git
cd dbt-ci
pip install -e ".[dev]"

Or with pipenv:

pipenv install --dev

pyproject.toml is the single source of truth for dependencies — the Pipfile installs the project itself rather than restating them, so the two cannot drift apart. Add or change a dependency in pyproject.toml, then run pipenv lock.

After installation, the tool is available as dbt-ci.

Quick Start

The Workflow: Initialize once with init, then run commands that use the cached state.

1. Initialize State

First, initialize the dbt-ci state. This downloads/reads reference state and creates a cache:

dbt-ci init \
  --dbt-project-dir dbt \
  --profiles-dir dbt \
  --reference-target production \
  --state dbt/.dbtstate

With Cloud Storage (GCS/S3):

dbt-ci init \
  --dbt-project-dir dbt \
  --state-uri gs://my-bucket/dbt-state/manifest.json \
  --reference-target production \
  --state dbt/.dbtstate

2. Run Modified Models

After initialization, run commands use the cached state automatically:

# No need to specify --state again!
dbt-ci run \
  --dbt-project-dir dbt \
  --profiles-dir dbt

With Docker:

dbt-ci run \
  --runner docker \
  --docker-image ghcr.io/dbt-labs/dbt-bigquery:latest

Commands

All commands share a set of common options (listed in the Common Options section below). Command-specific flags are listed under each command.

init - Initialize State

Creates initial state from your dbt project. Always run this first. Downloads reference manifest from cloud storage (if specified) and creates a local cache for subsequent commands.

dbt-ci init \
  --dbt-project-dir dbt \
  --profiles-dir dbt \
  --state-uri gs://my-bucket/manifest.json \
  --reference-target production \
  --state dbt/.dbtstate

Flags:

All common options also apply.

Git comparison and clone depth: the git and hybrid strategies diff origin/<base-ref>...HEAD — the merge base — so only the commits on your branch count as changes. Shallow clones often don't contain the merge base; dbt-ci falls back to a direct diff and logs a debug message when that happens. For an accurate change set, fetch full history (actions/checkout with fetch-depth: 0).

How git's file-level changes are mapped to nodes:

• Moved or renamed files are reported by git as a delete of the old path plus an add of the new one. dbt identifies nodes by unique_id, which a move does not change, so a node that still exists in the target manifest is treated as modified, not deleted and re-added. Only nodes that are really gone are reported as deleted (and dropped by delete).
• Macro changes select every node that calls the changed macro, directly or through another macro.
• YAML-only changes (config, contracts, columns in a schema.yml) select the model the YAML configures. With git, every model configured by a changed YAML file is selected; with hybrid, only those that dbt also flags as modified.
• --dbt-project-dir is matched as a directory relative to the repository root, so dbt, ./dbt, dbt/ and an absolute path behave the same, and files in look-alike folders such as analytics_dbt/ are ignored.

run - Run Modified Models

Detects and runs models that have changed. Uses cached state from init, and fails if init has not run.

dbt-ci run --dbt-project-dir dbt --mode models

Flags:

With --mode all, the steps run in dependency order: seeds → snapshots → models → tests, so models built on a changed snapshot read its new state and tests run once everything they check exists. The test step runs both data tests and dbt unit tests (dbt ≥ 1.8).

Limiting blast radius

By default a change selects its entire downstream graph. On a large project a change to a core staging model therefore rebuilds almost everything, which is the opposite of what change-based CI is for. --downstream-depth caps how far a change propagates:

dbt-ci run --downstream-depth 0   # only what changed
dbt-ci run --downstream-depth 1   # changed models and their direct children
dbt-ci run --downstream-depth 2   # two levels out
dbt-ci run                        # unlimited (default)

For a chain customers → l1 → l2 → l3 where only customers changed:

New and deleted nodes are always included regardless of depth — a new model has to run whether or not anything depends on it yet. Every seed, snapshot and model the run builds also gets its data and unit tests, even though a test sits one level below the node it checks.

Deleted nodes no longer exist in the target manifest, so run does not look for their downstream nodes there. A node that depended on a deleted one has changed itself (its ref or test arguments differ) and is picked up as modified or new. This also covers generic tests whose arguments changed: dbt gives them a new unique_id, so they show up as one deleted and one new test.

All common options also apply.

Examples:

# Run only modified models
dbt-ci run --mode models

# Run modified models with defer to production
dbt-ci run --mode models --defer

# Run all modified resources (models, tests, seeds, etc.)
dbt-ci run --mode all

# With Docker
dbt-ci run --runner docker --mode models

ephemeral - Ephemeral Environment

ephemeral has two subcommands: create builds the environment and destroy tears it down. Plain dbt-ci ephemeral (no subcommand) still means create, so existing pipelines keep working. Options go after the subcommand (dbt-ci ephemeral destroy --target ci).

ephemeral create

Clones changed models and their downstream dependencies into an isolated target schema using dbt clone, allowing integration testing without affecting production. Uses cached state from init.

Important: --target and --vars must match the environment you want to clone into. The clone operation reads your profiles.yml to determine the target database/schema — if these are wrong, models will be cloned to the wrong location or the command will fail.

dbt-ci ephemeral create \
  --target my-pr-env \
  --vars '{"use_production_data":"false"}'

How it works:

1. Reads the cached change set from init
2. Builds a selection of all affected models and their downstream dependencies
3. Runs dbt clone --select <nodes> targeting the specified environment
4. The cloned tables/views can then be used as the base for subsequent dbt run commands in the PR environment
5. Records the datasets it cloned into, so ephemeral destroy can drop them later

If dbt clone fails, ephemeral exits with a non-zero code on every runner.

Flags:

All common options also apply.

ephemeral destroy

Drops the datasets that ephemeral create cloned into (BigQuery). Datasets used by any node in the reference (production) manifest are never dropped, and a failed teardown fails the step.

- name: Tear down ephemeral environment
  if: ${{ always() }}
  run: dbt-ci ephemeral destroy

• Run it in the same job as ephemeral create: the list of cloned datasets lives in the dbt-ci cache (see Cache Location).
• Use if: always() so a failed run step doesn't leave the datasets behind.
• It doesn't upload anything or clear the cache, so it can run before or after finalize, and a failed teardown can be retried by running it again.
• --target defaults to the one init recorded. With --dry-run it only lists the datasets it would drop.

All common options also apply.

delete - Delete Removed Models

Detects and deletes models that have been removed from the project. Uses cached state from init.

dbt-ci delete --dry-run  # preview what will be deleted
dbt-ci delete            # execute deletions

The tables dropped are the ones in the reference manifest (usually production), so run the real delete only after the removal is merged, and use --dry-run everywhere else.

A node is only dropped if it is really gone from the current project. Nodes that still exist in the target manifest (for example a model whose file was moved or renamed, which git reports as delete + add) are skipped, as are relations that another target node still builds (for example through an alias). delete fails if init has not run, instead of passing without doing anything. On BigQuery the connection comes from the profile named in dbt_project.yml and the --target output in profiles.yml (its project and location).

Flags:

Only common options apply — no command-specific flags.

migration - Migrate Partitioning Changes

Detects models whose partitioning configuration changed between the reference and target state, and rebuilds the affected tables with the new partitioning spec. Uses cached state from init.

BigQuery cannot change a table's partitioning in place, so each affected table is copied into a temporary table with the new spec (keeping the model's cluster_by, require_partition_filter and partition_expiration_days). The original is then renamed to <table>__dbt_ci_backup, the copy is renamed into place, and only then is the backup dropped — if the swap fails, the original data is still in the backup table. The table migrated is the relation dbt resolves for the model, including its alias. Only incremental models are considered — other materializations are rebuilt by dbt anyway.

dbt-ci migration --dry-run  # list the tables that would be rebuilt
dbt-ci migration            # apply the changes

Warning: this rewrites tables in place. Always review the --dry-run output first. Requires a connector implementing a migration strategy — currently BigQuery only.

Flags:

Only common options apply — no command-specific flags.

report - Summarise the Run

Renders the change set detected by init and the status of every command that has run so far. In GitHub Actions the report is appended to the job summary automatically, so no workflow wiring is needed beyond calling it. Each command's entry ends as completed or failed however it exits, including dry runs and runs with nothing to do.

dbt-ci report                       # → $GITHUB_STEP_SUMMARY, or stdout locally
dbt-ci report --output report.md    # → a file
dbt-ci report --format json         # → machine-readable
dbt-ci report --pr-comment          # → also posted as a comment on the pull request

The report covers:

• Change counts and node names, grouped as modified / new / deleted. Long lists fold into a collapsible block.
• Exposure impact — the exposures downstream of the change set, including those downstream of deleted nodes, which is usually the case worth catching.
• Command status and duration for each of init, run, delete, ephemeral and migration that has run.

Flags:

All common options also apply.

Posting the report as a PR comment

With --pr-comment (or notifications.pr-comment: true), report also posts the markdown report as a comment on the pull request. dbt-ci owns a single comment per pull request and edits it on every later run instead of adding new ones. The comment is always markdown, even with --format json.

The step needs a token that can write pull request comments:

permissions:
  pull-requests: write

steps:
  - run: dbt-ci report --pr-comment
    env:
      GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

The pull request is detected from the GitHub Actions event, and GitHub Enterprise works through GITHUB_API_URL. Posting the comment never fails the job: on a run that is not for a pull request, without a token, or when GitHub rejects the request (pull requests from forks get a read-only token), report logs why and carries on. Reports longer than GitHub's 65,536-character comment limit are cut, with a note pointing to the job summary.

config - Generate a Config File

Writes a commented dbt-ci.config.yaml skeleton with the common options pre-filled, so you don't have to memorise every flag. See Configuration File.

dbt-ci config                          # create ./dbt-ci.config.yaml
dbt-ci config --output dbt/dbt-ci.config.yaml
dbt-ci config --force                  # overwrite an existing file

Flags:

This command does not take the common options.

finalize - Finalize State

Run after run, delete, or ephemeral to upload artifacts and clean up the local cache for the next CI run.

dbt-ci finalize
dbt-ci finalize --artifacts-uri s3://my-bucket/dbt-artifacts/

Flags:

Uploaded artifacts land at <artifacts-uri>/manifest.json, <artifacts-uri>/cache.json and <artifacts-uri>/logs.txt respectively (a trailing / on the URI is ignored). With --dry-run, nothing is uploaded, the cache is kept, and --clean-ephemeral only lists the datasets it would drop. log uploads the run log that dbt-ci writes to <cache dir>/dbt-ci.log.

--clean-ephemeral runs before the upload, so a missing manifest or a failed upload doesn't leave the CI datasets behind. If the cleanup itself fails, the artifacts are still uploaded, the step fails, and the cache is kept so a re-run can retry the cleanup.

Note: the log file always records at DEBUG level, including the resolved configuration for each command. Values whose names look sensitive (webhook, token, password, secret, credential, api_key) are masked before being written, so uploading the log does not publish your Slack webhook or credentials passed via --docker-env. This is a name-based heuristic — review your artifact bucket's access controls before enabling --files log.

All common options also apply.

Runners

dbt-ci supports multiple execution environments:

Local Runner

Execute dbt commands directly on your machine:

# After init
dbt-ci run \
  --runner local \
  --dbt-project-dir dbt

dbt Runner (Python API)

Uses dbt's Python API (fastest, default):

# After init - uses dbt Python API
dbt-ci run \
  --runner dbt \
  --dbt-project-dir dbt

Docker Runner

Run dbt commands inside a Docker container. Requires pip install 'dbt-ci[docker]':

dbt-ci run \
  --runner docker \
  --docker-image ghcr.io/dbt-labs/dbt-duckdb:latest \
  --docker-volumes "$(pwd)/dbt:/dbt:rw"

When using the Docker runner, --project-dir, --profiles-dir, and --state are derived from the volume map — if the host path is inside a mounted volume (compared by path, so a dbt mount does not cover dbt_project), the corresponding container path is passed to dbt inside the container; with nested mounts the most specific one wins. DBT_PROJECT_DIR, DBT_PROFILES_DIR or DBT_STATE in --docker-env take priority. If no matching volume is found, --project-dir and --profiles-dir are omitted and dbt falls back to its own defaults (typically the container's WORKDIR), while --state is passed through unchanged so an absolute container path can be given directly.

For Apple Silicon Macs:

dbt-ci run \
  --runner docker \
  --docker-platform linux/amd64 \
  --docker-image ghcr.io/dbt-labs/dbt-postgres:latest \
  --docker-volumes "$(pwd)/dbt:/dbt:rw"

Docker Advanced Options

Platform (for Apple Silicon compatibility):

--docker-platform linux/amd64  # or linux/arm64

Custom Volumes:

--docker-volumes "/host/path:/container/path" --docker-volumes "/another:/path:ro"

Environment Variables:

--docker-env "DBT_ENV=prod" --docker-env "MY_API_KEY=secret"

Network Mode:

--docker-network bridge  # or host, none, container:name

User:

--docker-user "1000:1000"  # or leave empty for auto-detect

Additional Docker Args:

--docker-args "--memory=2g --cpus=2"

dbt-ci drives Docker through the Python SDK rather than the docker CLI, so --docker-args supports the subset of docker run flags that map onto SDK options. Both --flag value and --flag=value spellings work:

Anything else is logged as ignored rather than dropped silently.

Note: containers are removed once the command finishes, so repeated CI runs don't accumulate stopped containers.

Complete Docker Example:

dbt-ci run \
  --runner docker \
  --docker-image ghcr.io/dbt-labs/dbt-postgres:1.7.0 \
  --docker-platform linux/amd64 \
  --docker-env "POSTGRES_HOST=host.docker.internal" \
  --docker-network host \
  --docker-volumes "$(pwd):/workspace" \
  --docker-volumes "$HOME/.aws:/root/.aws:ro" \
  --target prod

Common Options

These flags are available on every command.

Configuration File

dbt-ci supports a dbt-ci.config.yaml file as an alternative to passing every flag on the command line. It is loaded before any other options, and a flag passed on the command line always wins over it.

Default location: dbt-ci.config.yaml in the current working directory (override with --config / DBT_CONFIG).

If the config file is not found in the current directory, dbt-ci will automatically look for it inside --dbt-project-dir (or DBT_PROJECT_DIR). This means if your dbt project lives in a subdirectory (e.g. dbt/), placing dbt-ci.config.yaml there and setting DBT_PROJECT_DIR=dbt is enough — no --config flag needed.

The file uses a nested style where top-level keys map to common options, and command-specific options live under their command's key (init, run, finalize, ephemeral, docker):

Add the following comment to the top of your dbt-ci.config.yaml to get autocompletion and validation in editors that support yaml-language-server (e.g. VS Code with the YAML extension):

# yaml-language-server: $schema=https://raw.githubusercontent.com/datablock-dev/dbt-ci/main/dbt-ci.config.schema.json

dbt-ci.config.yaml

# yaml-language-server: $schema=https://raw.githubusercontent.com/datablock-dev/dbt-ci/main/dbt-ci.config.schema.json
project_dir: dbt
profiles_dir: dbt
state: dbt/.dbtstate
runner: docker

notifications:
  provider: slack
  format: compact   # the webhook itself is a secret: pass it via SLACK_WEBHOOK
  pr-comment: true  # dbt-ci report also comments on the pull request

init:
  state-uri: gs://my-bucket/dbt-state/manifest.json
  reference-target: production
  comparison-strategy: hybrid
  base-ref: main

run:
  nodes: models
  downstream-depth: 2

finalize:
  artifacts-uri: s3://my-bucket/dbt-artifacts/
  files:
    - manifest
    - cache

docker:
  image: docker.pkg.dev/my-project/dbt:latest
  volumes:
    - "${PWD}/dbt:/dbt:rw"
    - "${GOOGLE_APPLICATION_CREDENTIALS}:${GOOGLE_APPLICATION_CREDENTIALS}:ro"
  env:
    - "DBT_PROFILES_DIR=/dbt"
    - "GOOGLE_APPLICATION_CREDENTIALS=${GOOGLE_APPLICATION_CREDENTIALS}"
  network: host

The legacy flat DBT_* key style is also supported:

DBT_RUNNER: docker
DBT_PROJECT_DIR: dbt
DBT_STATE: dbt/.dbtstate
DBT_DOCKER_IMAGE: ghcr.io/dbt-labs/dbt-core:latest

Other keys accepted in the config file include dbt-version, a report: section (format, output) and a notifications: section (provider, format, webhook, pr-comment), matching the flags of the same names.

Precedence (highest → lowest):

1. CLI flags
2. dbt-ci.config.yaml
3. Shell environment variables
4. Built-in defaults

The config file sits above shell environment variables: a value you commit to dbt-ci.config.yaml is deliberate, whereas the environment a CI runner happens to export is not. To override a config value per-run, pass the flag rather than setting the environment variable.

--filters on run is command-line only — it has no environment variable or config file key.

The config file is validated on load. dbt-ci will exit with a clear error message if it contains unknown keys, invalid enum values (e.g. runner: kubernetes), or wrong types (e.g. defer: "yes" instead of a boolean).

${VAR_NAME} references inside the config file are resolved from the shell environment at load time.

Note: dbt-ci.config.yaml is ignored by git by default (it is listed in .gitignore). Use it for local developer overrides and commit a .example variant for your team.

Core

Docker Runner

Only used when --runner docker is set.

Pinning a dbt version or adapter

--dbt-version and --adapter install the requested packages into a cached virtual environment under ~/.cache/dbt-ci/venvs/ and run dbt from it. Only the local runner can execute an arbitrary dbt binary, so the pin applies there:

dbt-ci run --runner local --dbt-version 1.10.13 --adapter dbt-duckdb=1.10.0

Adapters may be given with or without a version (dbt-bigquery, dbt-bigquery=1.10.0, dbt-bigquery==1.10.0). Each version/adapter combination gets its own cached environment, and an interrupted install is rebuilt on the next run rather than reused.

The other runners resolve dbt from elsewhere and log a warning if a pin is set:

dbt v2

dbt v2 is a ground-up rewrite of dbt in Rust. It ships on PyPI as dbt (or dbt-oss) rather than dbt-core, as a self-contained binary with every adapter bundled. dbt-ci drives it through the CLI, so use the local or docker runner:

# dbt-ci installs dbt v2 into a cached virtual environment
dbt-ci init --runner local --dbt-version 2.0.9 ...
dbt-ci run  --runner local --dbt-version 2.0.9 ...

# ...or point dbt-ci at a dbt v2 binary you installed yourself
dbt-ci run --runner local --entrypoint /path/to/dbt ...

# ...or run dbt v2 from a Docker image
dbt-ci run --runner docker --docker-image my-registry/dbt-v2:2.0.9 ...

With the Docker runner, dbt-ci passes the dbt command (ls ..., run ...) as the container command, so the image's entrypoint must be dbt, as in the official dbt images. The default --docker-image is a dbt-core 1.x image, so pass a dbt v2 image explicitly. A minimal one:

FROM python:3.12-slim
RUN pip install --no-cache-dir "dbt==2.0.9"
ENTRYPOINT ["dbt"]

Things to know:

• Keep dbt v2 out of dbt-ci's environment. dbt-ci depends on dbt-core 1.x, and dbt v2 and dbt-core both install a dbt Python package and dbt executable, so installing them into one environment overwrites one with the other. Install dbt-ci on its own (e.g. pipx install dbt-ci or uv tool install dbt-ci) and let --dbt-version or --entrypoint supply dbt v2, or run dbt v2 from an image with --runner docker.
• The dbt runner is v1 only. It runs the dbt-core installed alongside dbt-ci in-process.
• --adapter is ignored for dbt v2, because adapters are bundled. A warning is logged.
• Deferral stays opt-in. dbt v2 defers to --state by default, while dbt v1 only defers when asked. dbt-ci passes --no-defer whenever it passes --state without --defer, so run, test and build behave the same on both versions. --defer (or DBT_DEFER, or DEFER_FLAG=--defer) still turns deferral on.
• Change detection works unchanged. dbt v2 still writes target/manifest.json in the same v12 schema dbt-ci parses. CI checks every manifest field dbt-ci relies on (node ids, resource types, project-relative file paths, configs, lineage) against dbt-core 1.10, 1.11, 1.12 and dbt v2, through both the local and the Docker runner.
• Upgrading mid-stream is supported. A reference state built with dbt-core 1.x can be compared against a PR that runs dbt v2, so CI keeps working while production is still on 1.x. Once production runs dbt v2, refresh the stored state so both sides match.

Bash Runner

Only used when --runner bash is set.

The bash runner gets the same absolute --project-dir, --profiles-dir and --state paths as the local runner.

Cloud Storage Support

dbt-ci supports storing and retrieving state files from cloud storage (GCS, S3), making it ideal for distributed CI/CD workflows.

Requires the matching extra: pip install 'dbt-ci[gcp]' for gs:// URIs, pip install 'dbt-ci[aws]' for s3:// URIs.

GCS/S3 State Storage

Store your dbt reference state in cloud storage for shared access across CI runs:

# Initialize and download state from GCS
dbt-ci init \
  --dbt-project-dir dbt \
  --state-uri gs://my-bucket/dbt-state/manifest.json \
  --reference-target production \
  --state dbt/.dbtstate

# Run using cached state (no need to specify URI again)
dbt-ci run --dbt-project-dir dbt --mode models

Benefits:

• 🔄 Shared State: Download the same reference state across different CI jobs
• 💾 Cache-Based: After init, commands use local cache (no repeated downloads)
• 📦 No Git Commits: State files don't need to be committed to version control
• 🚀 Scalable: Works seamlessly in containerized and distributed environments
• 🔐 Secure: Leverage cloud IAM and bucket policies for access control

Configuration:

The tool uses cloud credentials from your environment. Ensure your bucket is accessible:

# For GCS
export GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account.json

# For AWS S3
export AWS_ACCESS_KEY_ID=your_key
export AWS_SECRET_ACCESS_KEY=your_secret
export AWS_DEFAULT_REGION=us-east-1

# Or use IAM roles (recommended in CI/CD)
dbt-ci init --state-uri gs://my-bucket/manifest.json

Supported URI Formats:

• gs://bucket-name/path/to/manifest.json (Google Cloud Storage)
• s3://bucket-name/path/to/manifest.json (AWS S3)

Environment Variables

All CLI options can also be set via environment variables:

export DBT_PROJECT_DIR=./dbt
export DBT_PROFILES_DIR=./dbt
export DBT_TARGET=production
export DBT_RUNNER=local

# After running init, just use:
dbt-ci run

Common Environment Variables:

• DBT_PROJECT_DIR - Path to dbt project
• DBT_PROFILES_DIR - Path to profiles.yml location
• DBT_TARGET - Target environment to use
• DBT_RUNNER - Runner type (local, docker, bash, dbt)
• DBT_CI_CACHE_DIR - Where the state cache and log file live (default: <temp dir>/dbt-ci)

Note: State management is cache-based. Run init once, then subsequent commands automatically use the cached state.

Settings Inherited From init

init records the --target, --vars and --state it ran with, and later commands reuse them, so they only need to be given once:

dbt-ci init --target ci --vars '{"use_production_data": false}' --state dbt/.dbtstate
dbt-ci run          # runs against target 'ci' with the same vars
dbt-ci delete       # likewise

The state path is recorded even when init chose it: with --state-uri and no --state, the manifest is downloaded to <dbt-project-dir>/.dbtstate, and later commands such as ephemeral pass that directory to dbt as --state (needed for dbt clone and other deferring commands).

Passing the flag explicitly still wins — the cache only fills in what was left unset.

Cache Location

init writes its cache (state comparison, manifests, run report and log file) to <temp dir>/dbt-ci, and later commands read it from there. Because the location is fixed, two dbt-ci runs executing at the same time on the same machine — for example two pull request jobs on a shared self-hosted runner — would overwrite each other's state.

Set DBT_CI_CACHE_DIR to give each run its own directory:

export DBT_CI_CACHE_DIR="/tmp/dbt-ci-${GITHUB_RUN_ID}"
dbt-ci init ...
dbt-ci run ...

All commands in the same CI job must see the same value, since that is how run, delete, ephemeral and finalize find the cache written by init.

CI/CD Integration

GitHub Actions Example

name: dbt CI

on: [pull_request]

jobs:
  dbt-ci:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v7
      
      - name: Set up Python
        uses: actions/setup-python@v7
        with:
          python-version: '3.12'
      
      - name: Configure AWS Credentials
        uses: aws-actions/configure-aws-credentials@v2
        with:
          role-to-assume: arn:aws:iam::123456789012:role/GitHubActionsRole
          aws-region: us-east-1
      
      - name: Install dbt-ci
        # The gcp extra provides the GCS client needed for the gs:// state URI below
        run: pip install 'dbt-ci[gcp] @ git+https://github.com/datablock-dev/dbt-ci.git@main'
      
      - name: Initialize dbt-ci with cloud state
        run: |
          dbt-ci init \
            --dbt-project-dir dbt \
            --state-uri gs://my-dbt-state/prod/manifest.json \
            --reference-target production \
            --state dbt/.dbtstate
      
      - name: Run modified models
        run: |
          dbt-ci run --mode models

GitLab CI Example

dbt-ci:
  image: python:3.12
  script:
    - pip install 'dbt-ci[gcp] @ git+https://github.com/datablock-dev/dbt-ci.git@main'
    - dbt-ci init --dbt-project-dir dbt --state-uri gs://my-dbt-state/prod/manifest.json --reference-target production --state dbt/.dbtstate
    - dbt-ci run --mode models
  only:
    - merge_requests

Features

• 🎯 Smart Detection: Automatically identifies modified, new, and deleted models
• 📊 Dependency Tracking: Generates and traverses dependency graphs for lineage analysis
• 🔄 State Comparison: Compares current state against production for precise CI
• ☁️ Cloud Storage: S3 integration for shared state across distributed CI/CD workflows
• 🚀 Multiple Runners: Supports local, Docker, bash, and dbt Python API execution
• 🐳 Docker-First: Extensive Docker configuration for containerized workflows
• ⚡ Selective Execution: Run only what changed, saving time and resources
• 🔌 Adapter Support: Install specific dbt versions and adapters on-demand
• 💬 Notifications: Slack webhook integration for CI/CD alerts
• ♻️ Ephemeral Environments: Test changes in isolated environments
• 🧹 Cleanup: Automatically remove deleted models from target warehouse
• 🎯 Blast-Radius Control: Cap how far a change propagates with --downstream-depth
• 📝 Run Reports: Markdown summary of the change set, exposure impact and command status
• 🔀 Partition Migrations: Rebuild tables whose partitioning configuration changed (BigQuery)

Use Cases

Pull Request CI

Only build and test models affected by PR changes:

# Initialize with reference state
dbt-ci init --state-uri gs://bucket/manifest.json --reference-target production --state dbt/.dbtstate

# Run modified models with defer
dbt-ci run --mode models --defer

Distributed CI with Cloud Storage

Share state across multiple CI jobs:

# Job 1: Initialize state (downloads from cloud)
dbt-ci init --state-uri gs://my-bucket/manifest.json --reference-target production --state dbt/.dbtstate

# Job 2: Run models (uses cached state)
dbt-ci run --mode models

# Job 3: Run tests (uses cached state)
dbt-ci run --mode tests

Selective Testing

Run tests only for modified models:

# After init
dbt-ci run --mode tests

Schema Migrations

Clean up deleted models from production:

# After init
dbt-ci delete --target production

Multi-Environment Testing

Create an ephemeral test environment, and tear it down at the end of the job:

dbt-ci ephemeral create --target ci
dbt-ci ephemeral destroy

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Development Setup

1. Clone the repository
2. Install dependencies: pip install -e ".[dev]"
3. Run tests: pytest tests/
4. Run linting: black dbt_ci/ tests/

Commit Message Format

This project uses Conventional Commits for automated releases:

• feat: New feature (minor version bump)
• fix: Bug fix (patch version bump)
• docs: Documentation changes
• refactor: Code refactoring
• test: Adding tests
• chore: Maintenance tasks

Example:

git commit -m "feat: add Docker runner support"
git commit -m "fix: resolve path resolution on Windows"

See RELEASING.md for details on the automated release process.

License

See LICENSE file for details.

Links

• PyPI: https://pypi.org/project/dbt-ci/
• Documentation: https://datablock.dev
• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Release notes: GitHub Releases