Deleting/recreating image fails (e.g. with terraform) - possible race condition

[Yenya]

Description When trying to create an image just after the image of the same name has been deleted (e.g. when using terraform), creating the image fails with "NAME is already take by IMAGE ". Waiting for a few seconds and retrying succeeds. So it seems the oneimage delete finishes and reports success prematurely.

To Reproduce

$ oneimage clone 1229 test-yenya
ID: 1290
(wait for the image to become READY)
$ oneimage delete 1290 ; until oneimage clone 1229 test-yenya; do sleep 1; done
[one.image.clone] Error allocating a new image. NAME is already taken by IMAGE 1290.
[one.image.clone] Error allocating a new image. NAME is already taken by IMAGE 1290.
ID: 1291

Expected behavior The image name should be available as soon as the image delete finishes.

Details

• Affected Component: core oned, Ceph datastore
• Hypervisor: KVM
• Version: opennebula-5.12.0.3-1.ce.el7.x86_64

Additional context I discovered this when trying to use an OpenNebula Terraform provider. As soon as Terraform decides it needs to recreate the image, it does delete immediately followed by clone, which fails. Reported this as issue in OpenNebula/terraform-provider-opennebula: https://github.com/OpenNebula/terraform-provider-opennebula/issues/116 but the maintainer suggests that it is a problem with core OpenNebula.

Progress Status

• [ ] Branch created
• [ ] Code committed to development branch
• [ ] Testing - QA
• [ ] Documentation
• [ ] Release notes - resolved issues, compatibility, known issues
• [ ] Code committed to upstream release/hotfix branches
• [ ] Documentation committed to upstream release/hotfix branches

[paczerny]

This is not a bug, this is a feature :-) On oneimage delete the oned evaluates if it's possible to delete the image: It's not used, not locked, ... If all checks passes, it set the Image state to Image::DELETE and sends asynchronous command to storage driver to delete the file, doesn't wait for execution and immediately returns success. The oned deletes the Image from DB, when it receives answer from the storage driver. If it fails to delete the file, the error is reported only to oned.log There is no option to force synchronous execution

[Yenya]

So, the correct way to solve this would be terraform polling for the image being deleted? Is OpenNebula API sufficiently rich for that?

[paczerny]

Something like this should work: oneimage delete test; while oneimage show test &>/dev/null; do sleep 1; done; oneimage clone 0 test