Build deletion fails due to buildartifacttype schema validation error

[sblair-metrostar]

On two ~1 month old Nebari installations, I've observed that conda-store build deletions triggered through the UI fail with the following error on the worker pod:

Inspecting the postgres database confirms that CONTAINER_REGISTRY value is missing from the type definition. It appears it should have been added based on the alembic migration here and the version reported in the alembic_version table:

I'm not sure where the gap might be. If I manually add the value to the type definition, deletions appear to work as far as I can tell:

ALTER TYPE buildartifacttype ADD VALUE 'CONTAINER_REGISTRY' AFTER 'DOCKER_MANIFEST'

[kcpevey]

We are still seeing this on the backend, though on the admin interface, the builds are showing up as deleted:

[iameskild]

@pavithraes @trallard is there bandwidth to include a fix for this in the upcoming conda-store release? This blocks users from deleting envs/builds.

[kcpevey]

We confirmed that the fix about fixes the error on the delete process. Now we see the delete process succeeding.

[2023-09-05 21:11:10,476: INFO/MainProcess] Task task_delete_build[d12572cf-694d-4341-86ab-34a21dac779f] received
[2023-09-05 21:11:10,482: WARNING/ForkPoolWorker-2] [CondaStoreWorker] deleting artifacts for build=89
[2023-09-05 21:11:13,712: INFO/ForkPoolWorker-2] Task task_delete_build[d12572cf-694d-4341-86ab-34a21dac779f] succeeded in 3.235292388126254s: None

However, our storage size is still the same. I am going to open a separate issue for that.

[nkaretnikov]

Metadata update (size) and misc Python errors related to deletion were fixed in #572.
Postgres migration issue was addressed in #574.

I've also verified on a8d3524, which is on top of 432eaa1, which has both fixes, that I can:

• build the default env in the filesystem namespace
• add another env in user namespace with the same spec (but different name)
• build one env, check its artifacts and blobs in minio
• remove the same env, data in minio is deleted
• same for another one in user namespace
• then build both envs and test each other separately (to test potential issues with pattern-matching), all data gone in the end.

There's another issue that we may potentially want to address, in delete_build_artifact CONTAINER_REGISTRY is skipped. I'll open a new issue since that's separate.

I think this issue can be closed.

[costrouc]

Closing issue as completed. cc @nkaretnikov