pronounced as gimmick.

Manage Google Cloud Platform (GCP) infrastructure using migrations that describe incremental changes such as additions or deletions of resources. This work is inspired by MyBatis migrations for SQL database setup.

Introduction blog post

Your gmig infrastructure is basically a folder with incremental change files, each with a timestamp prefix (for sort ordering) and readable name.

/20180214t071402_create_some_account.yaml
/20180214t071522_add_permissions_to_some_account.yaml
/my-gcp-production-project
    gmig.json

Each change is a single YAML file with one or more shell commands that change infrastructure for a project.

# add loadrunner service account

do:
- gcloud iam service-accounts create loadrunner --display-name "LoadRunner"

undo:
- gcloud iam service-accounts delete loadrunner

A change must have at least a do section and optionally an undo section. The do section typically has a list of gcloud commands that create resources but any available tool can be used. All lines will be executed at once using a single temporary shell script so you can use shell variables to simplify each section. The undo section typically has an ordered list of gcloud commands that deletes the same resources (in reverse order if relevant). Each command in each section can use the following environment variables: $PROJECT,$REGION,$ZONE and any additional environment variables populated from the target configuration (see env section in the configuration below).

Information about the last applied migration to a project is stored as a Google Storage Bucket object. Therefore, usage of this tool requires you to have create a Bucket and set the permissions (Storage Writer) accordingly. To view the current state of your infrastructure related to each migration, you can add the view section to the YAML file, such as:

# add loadrunner service account

do:
- gcloud iam service-accounts create loadrunner --display-name "LoadRunner"

undo:
- gcloud iam service-accounts delete loadrunner

view:
- gcloud iam service-accounts describe loadrunner

and use the view subcommand.

NAME:
gmig - Google Cloud Platform infrastructure migration tool

USAGE:
gmig [global options] command [command options] [arguments...]

COMMANDS:
    init     Create the initial configuration, if absent.
    new      Create a new migration file from a template using a generated timestamp and a given title.
    up       Runs the do section of all pending migrations in order, one after the other.
             If a migration file is specified then stop after applying that one.
    down     Runs the undo section of the last applied migration only.
    plan     Log commands of the do section of all pending migrations in order, one after the other.
    status   List all migrations with details compared to the current state.
    view     Runs the view section of all applied migrations to see the current state reported by your infrastructure.
    force    state | do | undo
    util     create-named-port | delete-named-port
    export   project-iam-policy | storage-iam-policy
    help, h  Shows a list of commands or help for one command

GLOBAL OPTIONS:
-q                   quiet mode, accept any prompt
-v                   verbose logging
--help, -h           show help
--print-version, -V  print only the version

Pre-compiled binaries are available for download from the Releases section on this page. If you want to create your own version, you need to compile it using the Go SDK.

go get github.com/emicklei/gmig

Prepares your setup for working with migrations by creating a gmig.json file in a target folder.

gmig init my-gcp-production-project

Then your filesystem will have:

/my-gcp-production-project/
    gmig.json

You must change the file gmig.json to set the Bucket name.

{
    "project": "my-gcp-production-project",
    "region": "europe-west1",
    "zone": "europe-west1-b",
    "bucket":"mycompany-gmig-states",
    "state":"gmig-last-migration",
    "env" : {
        "FOO" : "bar"
    }
}

If you decide to store state files of different projects in one Bucket then set the state object name to reflect this, eg. myproject-gmig-state. If you want to apply the same migrations to different regions/zones then choose a target folder name to reflect this, eg. my-gcp-production-project-us-east. Values for region and zone are required if you want to create Compute Engine resources. The env map can be used to parameterize commands in your migrations. All commands will have access to the value of $FOO.

Creates a new migration for you to describe a change to the current state of infrastructure.

gmig new "add storage view role to cloudbuild account"

Using a combination of the options --do, --undo and --view, you can set the commands directly for the new migration.

List all migrations with an indicator (applied,pending) whether is has been applied or not.

gmig status my-gcp-production-project/

Run this command in the directory where all migrations are stored. Use --migrations for a different location.

Log commands of the do section of all pending migrations in order, one after the other. If migration file is given then stop after applying that one.

Executes the do section of each pending migration compared to the last applied change to the infrastructure. If migration file is given then stop after applying that one. Upon each completed migration, the gmig-last-migration object is updated in the bucket.

gmig up my-gcp-production-project

Executes one undo section of the last applied change to the infrastructure. If completed then update the gmig-last-migration object.

gmig down my-gcp-production-project

Executes the view section of each applied migration to the infrastructure. If migration file is given then run that view only.

gmig view my-gcp-production-project

Exporting migrations from existing infrastructure is useful when you start working with gmig but do not want to start from scratch. Several sub commands are (or will become) available to inspect a project and export migrations to reflect the current state. After marking the current state in gmig (using force-state), new migrations can be added that will bring your infrastructure to the next state. The generated migration can ofcourse also be used to just copy commands to your own migration.

Generate a new migration by reading all the IAM policy bindings from the current infrastructure of the project.

gmig -v export project-iam-policy my-project/

Generate a new migration by reading all the IAM policy bindings, per Google Storage Bucket owned by the project.

gmig -v export storage-iam-policy my-project/

Sometimes you need to fix things because you made a mistake or want to reorganise your work. Use the force and confirm your action.

Explicitly set the state for the target to the last applied filename. This command can be useful if you need to work from existing infrastructure. Effectively, this filename is written to the bucket object. Use this command with care!.

gmig force state my-gcp-production-project 20180214t071402_create_some_account.yaml

Explicitly run the commands in the do section of a given migration filename. The gmig-last-migration object is not updated in the bucket. Use this command with care!.

gmig force do my-gcp-production-project 20180214t071402_create_some_account.yaml

Explicitly run the commands in the undo section of a given migration filename. The gmig-last-migration object is not updated in the bucket. Use this command with care!.

gmig force undo my-gcp-production-project 20180214t071402_create_some_account.yaml

The Cloud SDK has a command to set-named-ports but not a command to add or delete a single name:port mapping. To simplify the migration command for creating a name:port mapping, this gmig util command is added. First it calls get-named-ports to retrieve all existing mappings. Then it will call set-named-ports with the new mapping unless it already exists.

The Cloud SDK has a command to set-named-ports but not a command to add or delete a single name:port mapping. To simplify the migration command for deleting a name:port mapping, this gmig util command is added. First it calls get-named-ports to retrieve all existing mappings. Then it will call set-named-ports without the mapping.

This repository has a number of examples of migrations.

© 2018, ernestmicklei.com. MIT License. Contributions welcome.