Problem

Currently, we create Deployment for all deployment jobs. Since Deployment is a stateful model, this looks fine at the glance, however, it doesn't make sense to create a Deployment for jobs that are not executable state yet. Because of this, there are some data integrity and UX issues caused, such as:

• Users seeing Deployment Approval even if it's not executable yet.
• Users seeing manual deployment jobs in Environment details page even if it's not executable yet.
• Irrelevant Deployment records in Environment page harms the readability on the page.

Technical side

Basically, there are two steps until a job becomes executable state:

1. Create a job. https://gitlab.com/gitlab-org/gitlab/-/blob/master/app/services/ci/create_pipeline_service.rb. Job status is created.
2. Process a job. https://gitlab.com/gitlab-org/gitlab/-/blob/master/app/services/ci/pipeline_processing/atomic_processing_service.rb. Job status transitions to pending (pickable by runner), manual (manually executable by user) OR stay created if the pipeline pointer hasn't reached the job yet.

Example

You have the following .gitlab-ci.yml:

stages:
  - stage-1
  - stage-2
  - stage-3

deploy-a:
  stage: stage-1
  script: echo
  environment: prd
  when: manual

deploy-b:
  stage: stage-2
  script: echo
  environment: prd
  when: manual

deploy-c:
  stage: stage-3
  script: echo
  environment: prd
  when: manual

Expected behavior

1. deploy-a (manual with deployment) => deploy-b (created w/o deployment) => deploy-c (created w/o deployment)
2. deploy-a (success with deployment) => deploy-b (manual with deployment) => deploy-c (created w/o deployment)
3. deploy-a (success with deployment) => deploy-b (success with deployment) => deploy-c (manual with deployment)

Current bug behavior

1. deploy-a (manual with deployment) => deploy-b (created with deployment) => deploy-c (created with deployment)
2. deploy-a (success with deployment) => deploy-b (manual with deployment) => deploy-c (created with deployment)
3. deploy-a (success with deployment) => deploy-b (success with deployment) => deploy-c (manual with deployment)

Goal

Create Deployment only for processable jobs.

Performance implication

This will slow down the table growth a lot because many jobs are never processed by some reasons (e.g. pipeline stage didn't proceed to deployment). If avoiding to create deployments for such rows is successful, we can proceed with deleting legacy rows with the same condition. This will reduce main database load as described in gitlab#396799 (comment 1328744004).

Related

• gitlab#383600 (comment 1185384822)
• PoC: PoC: Create Deployment only for processable jobs (gitlab!108659 - closed)
• gitlab#383600 (closed)