Deployment safety

Deployment jobs can be more sensitive than other jobs in a pipeline, and might need to be treated with extra care. GitLab has several features that help maintain deployment security and stability.

You can:

If you are using a continuous deployment workflow and want to ensure that concurrent deployments to the same environment do not happen, you should enable the following options:

Restrict write access to a critical environment

By default, environments can be modified by any team member that has at least the Developer role. If you want to restrict write access to a critical environment (for example a production environment), you can set up protected environments.

Ensure only one deployment job runs at a time

Pipeline jobs in GitLab CI/CD run in parallel, so it’s possible that two deployment jobs in two different pipelines attempt to deploy to the same environment at the same time. This is not desired behavior as deployments should happen sequentially.

You can ensure only one deployment job runs at a time with the resource_group keyword in your .gitlab-ci.yml.

For example:

deploy:
 script: deploy-to-prod
 resource_group: prod

Example of a problematic pipeline flow before using the resource group:

  1. deploy job in Pipeline-A starts running.
  2. deploy job in Pipeline-B starts running. This is a concurrent deployment that could cause an unexpected result.
  3. deploy job in Pipeline-A finished.
  4. deploy job in Pipeline-B finished.

The improved pipeline flow after using the resource group:

  1. deploy job in Pipeline-A starts running.
  2. deploy job in Pipeline-B attempts to start, but waits for the first deploy job to finish.
  3. deploy job in Pipeline-A finishes.
  4. deploy job in Pipeline-B starts running.

For more information, see Resource Group documentation.

Skip outdated deployment jobs

The execution order of pipeline jobs can vary from run to run, which could cause undesired behavior. For example, a deployment job in a newer pipeline could finish before a deployment job in an older pipeline. This creates a race condition where the older deployment finished later, overwriting the “newer” deployment.

You can ensure that older deployment jobs are cancelled automatically when a newer deployment runs by enabling the Skip outdated deployment jobs feature.

Example of a problematic pipeline flow before enabling Skip outdated deployment jobs:

  1. Pipeline-A is created on the default branch.
  2. Later, Pipeline-B is created on the default branch (with a newer commit SHA).
  3. The deploy job in Pipeline-B finishes first, and deploys the newer code.
  4. The deploy job in Pipeline-A finished later, and deploys the older code, overwriting the newer (latest) deployment.

The improved pipeline flow after enabling Skip outdated deployment jobs:

  1. Pipeline-A is created on the default branch.
  2. Later, Pipeline-B is created on the default branch (with a newer SHA).
  3. The deploy job in Pipeline-B finishes first, and deploys the newer code.
  4. The deploy job in Pipeline-A is automatically cancelled, so that it doesn’t overwrite the deployment from the newer pipeline.

Prevent deployments during deploy freeze windows

If you want to prevent deployments for a particular period, for example during a planned vacation period when most employees are out, you can set up a Deploy Freeze. During a deploy freeze period, no deployment can be executed. This is helpful to ensure that deployments do not happen unexpectedly.

Setting appropriate roles to your project

GitLab supports several different roles that can be assigned to your project members. See Project members permissions for an explanation of these roles and the permissions of each.

Protect production secrets

Production secrets are needed to deploy successfully. For example, when deploying to the cloud, cloud providers require these secrets to connect to their services. In the project settings, you can define and protect CI/CD variables for these secrets. Protected variables are only passed to pipelines running on protected branches or protected tags. The other pipelines don’t get the protected variable. You can also scope variables to specific environments. We recommend that you use protected variables on protected environments to make sure that the secrets aren’t exposed unintentionally. You can also define production secrets on the runner side. This prevents other users with the Maintainer role from reading the secrets and makes sure that the runner only runs on protected branches.

For more information, see pipeline security.

Separate project for deployments

All users with the Maintainer role for the project have access to production secrets. If you need to limit the number of users that can deploy to a production environment, you can create a separate project and configure a new permission model that isolates the CD permissions from the original project and prevents the original users with the Maintainer role for the project from accessing the production secret and CD configuration. You can connect the CD project to your development projects by using multi-project pipelines.

Protect gitlab-ci.yml from change

A .gitlab-ci.yml may contain rules to deploy an application to the production server. This deployment usually runs automatically after pushing a merge request. To prevent developers from changing the .gitlab-ci.yml, you can define it in a different repository. The configuration can reference a file in another project with a completely different set of permissions (similar to separating a project for deployments). In this scenario, the .gitlab-ci.yml is publicly accessible, but can only be edited by users with appropriate permissions in the other project.

For more information, see Custom CI/CD configuration path.

Require an approval before deploying

Before promoting a deployment to a production environment, cross-verifying it with a dedicated testing group is an effective way to ensure safety. For more information, see Deployment Approvals.

Troubleshooting

Pipelines jobs fail with The deployment job is older than the previously succeeded deployment job...

This is caused by the Skip outdated deployment jobs feature. If you have multiple jobs for the same environment (including non-deployment jobs), you might encounter this problem, for example:

build:service-a:
 environment:
   name: production

build:service-b:
 environment:
   name: production

The Skip outdated deployment jobs might not work well with this configuration, and must be disabled.