Link Search Menu Expand Document

lakeFS Hooks

Like other version control systems, lakeFS allows you to configure Actions to trigger when predefined events occur.

Example use cases

  1. Format Validator: A webhook that checks new files to ensure they are of a set of allowed data format.
  2. Schema Validator: A webhook that reads new Parquet and ORC files to ensure they don’t contain a block list of column names (or name prefixes). This is useful for avoiding accidental PII exposure.
  3. Integration with external systems: Post-merge and post-commit hooks could be used to export metadata about the change to another system. A common example is exporting symlink.txt files that allow e.g. AWS Athena to read data from lakeFS.
  4. Notifying downstream consumers: Running a post-merge hook to trigger an Airflow DAG or to send a Webhook to an API, notifying it of the change that happened

For more examples and configuration samples, check out the examples on the lakeFS repository.


Hook types

Currently, there are two types of Hooks that are supported by lakeFS: Webhook and Airflow. Experimental support for Lua was introduced in lakeFS 0.86.0.

Supported Events

Event Description
pre-commit Runs when the commit occurs, before the commit is finalized
post-commit Runs after the commit is finalized
pre-merge Runs on the source branch when the merge occurs, before the merge is finalized
post-merge Runs on the merge result, after the merge is finalized
pre-create-branch Runs on the source branch prior to creating a new branch
post-create-branch Runs on the new branch after the branch was created
pre-delete-branch Runs prior to deleting a branch
post-delete-branch Runs after the branch was deleted
pre-create-tag Runs prior to creating a new tag
post-create-tag Runs after the tag was created
pre-delete-tag Runs prior to deleting a tag
post-delete-tag Runs after the tag was deleted

lakeFS Actions are handled per repository and cannot be shared between repositories. A failure of any Hook under any Action of a pre-* event will result in aborting the lakeFS operation that is taking place. Hook failures under any Action of a post-* event will not revert the operation.

Hooks are managed by Action files that are written to a prefix in the lakeFS repository. This allows configuration-as-code inside lakeFS, where Action files are declarative and written in YAML.

Terminology

Action

An Action is a list of Hooks with the same trigger configuration, i.e. an event will trigger all Hooks under an Action or none at all. The Hooks under an Action are ordered and so is their execution. A Hook will only be executed if all the previous Hooks that were triggered with it had passed.

Hook

A Hook is the basic building block of an Action. The failure of a single Hook will stop the execution of the containing Action and fail the Run.

Action file

Schema of the Action file:

Property Description Data Type Required Default Value
name Identify the Action file String false If missing, filename is used instead
on List of events that will trigger the hooks List true  
on.branches Glob pattern list of branches that triggers the hooks List false Not applicable to Tag events. If empty, Action runs on all branches
hooks List of hooks to be executed List true  
hook.id ID of the hook, must be unique within the Action String true  
hook.type Type of the hook (types) String true  
hook.properties Hook’s specific configuration Dictionary true  

Example:

name: Good files check
description: set of checks to verify that branch is good
on:
  pre-commit:
  pre-merge:
    branches:
      - main
hooks:
  - id: no_temp
    type: webhook
    description: checking no temporary files found
    properties:
      url: "https://your.domain.io/webhook?notmp=true?t=1za2PbkZK1bd4prMuTDr6BeEQwWYcX2R"
  - id: no_freeze
    type: webhook
    description: check production is not in dev freeze
    properties:
      url: "https://your.domain.io/webhook?nofreeze=true?t=1za2PbkZK1bd4prMuTDr6BeEQwWYcX2R"

Note: lakeFS will validate action files only when an Event has occurred.
Use lakectl actions validate <path> to validate your action files locally.

Run

A Run is an instantiation of the repository’s Action files when the triggering event occurs. For example, if your repository contains a pre-commit hook, every commit would generate a Run for that specific commit.

lakeFS will fetch, parse and filter the repository Action files and start to execute the Hooks under each Action. All executed Hooks (each with hook_run_id) exist in the context of that Run (run_id).


Uploading Action files

Action files should be uploaded with the prefix _lakefs_actions/ to the lakeFS repository. When an actionable event (see Supported Events above) takes place, lakeFS will read all files with prefix _lakefs_actions/ in the repository branch where the action occurred. A failure to parse an Action file will result with a failing Run.

For example, lakeFS will search and execute all the matching Action files with the prefix lakefs://example-repo/feature-1/_lakefs_actions/ on:

  1. Commit to feature-1 branch on example-repo repository.
  2. Merge to main branch from feature-1 branch on repo1 repository.

Runs API & CLI

The lakeFS API and lakectl expose the results of executions per repository, branch, commit, and specific Action. The endpoint also allows to download the execution log of any executed Hook under each Run for observability.

Result Files

The metadata section of lakeFS repository with each Run contains two types of files:

  1. _lakefs/actions/log/<runID>/<hookRunID>.log - Execution log of the specific Hook run.
  2. _lakefs/actions/log/<runID>/run.manifest - Manifest with all Hooks execution for the run with their results and additional metadata.

Note: Metadata section of a lakeFS repository is where lakeFS keeps its metadata, like commits and metaranges. Metadata files stored in the metadata section aren’t accessible like user stored files.