Is there a way to work with multiple workspaces in parallel?

user_118587 0
Viewed 411

In a Continuous Integration setup (Jenkins) I'd like to deploy (terraform apply) changes of the same Terraform configuration code using multiple workspaces at the same time using the same copy of the repository - i.e. the very same directory with .tf files.

Stages are executed in parallel (on same node/agent) and each of them consists of the same sequence of operations (terraform workspace select and then terraform apply).

It doesn't seem supported as the attempts fail with lock errors, e.g.

# terraform workspace select "workspace1"
# terraform apply [...] -input=false -auto-approve

 Switched to workspace "workspace1".
 Acquiring state lock. This may take a few moments...
 
 Error: Error locking state: Error acquiring the state lock: ConditionalCheckFailedException: The conditional request failed
 Lock Info:
   ID:        52973611-a892-deac-985b-5aa28172fdaf
   Path:      my-project/env:/workspace5/state
   Operation: OperationTypeApply
   Who:       @87ddd2118473
   Version:   0.13.5
   Created:   2021-01-11 15:20:06.635256155 +0000 UTC
   Info:

So it looks like lock for workspace5 gets in the way for applying for workspace1.

Is there any way out?

I use:

  • Terraform 0.13.5
  • s3 as the backend
  • DynamoDB for locks
terraform continuous-integration
3 Answers

Just tested this locally so not sure if it will work or a CI system.

Rather than using terraform workspace select xxxx use the environment variable set for workspaces. So in one session do:

export TF_WORKSPACE=one-environment
terraform apply -input=false -auto-approve

Then in parralel you can do:

export TF_WORKSPACE=another-environment
terraform apply -input=false -auto-approve

This seems to work and doesnt throw any locking errors.

user_81020250

One idea I have (which seems to work) is to copy the whole codebase to a separate directory. I.e. each workspace to be applied to each own. And run terraform workspace select and terraform apply from separate copies.

It works. But it's lame (inefficient).

user_1185870

The only way I see you able to achieve what you want is by defining a lock resource in Jenkins and then essentially locking that job/step so it will only every execute it one at a time, or by ensuring that the jobs are ran on separate nodes/agents.

user_142850210
Related