Chapter 19 — Teardown (the CLI half)¶

Goal: remove the two resource classes the Python SDK can't — your spaces and your location filter — using cdf commands driven by your own config.<YOURNAME>-training.yaml.

Teardown happens in two halves, and you need both:

notebooks/06_teardown.ipynb — pure Python SDK. Deletes your global resources: functions, transformations, workflow, RAW database, the classic OBJ file and 3D model, and archives the data set. Run that first.
This chapter — the two things the public SDK cannot delete, via cdf.

⚠️ Destructive & irreversible. Only ever purge spaces / filters that carry your YOURNAME (UPPERCASE, e.g. SEBASTIAN). Never touch another participant's.

19.0 [INFO] Before you delete anything — two rules¶

You have 72 hours. Deleting an instance does not destroy it. CDF soft-deletes: the instance is stamped with a deletedTime, vanishes from every normal query, and is permanently collected roughly three days later. Inside that window it is still visible through /sync, and re-running the transformation that created it restores it — the source row never moved. Chapter 14 section 14.8 walks through exactly that recovery.

Two caveats worth carrying:

Soft-deleted instances still count against your project's instance limits until they are collected. Deleting is not the same as freeing.
A space cannot be deleted while it still holds anything. That is why the order below is instances first, spaces last.

Delete edges before nodes. Deleting a node cascades to every edge attached to it, and restoring the node does not restore the edges. You would get your instance back with all of its connections gone — usually worse than the original mistake. It also avoids a long cascade on a well-connected node.

That matters directly here: the CogniteDiagramAnnotation edges you created in Chapter 08 are edges. They go first.

19.1 [ACTION] Purge your three spaces¶

🟢 [ACTION] Run these in a real interactive terminal — not a notebook ! cell, not piped. cdf data purge space requires you to type the project name to confirm, and the --yes/-y flag is deprecated (it does nothing now). Order matters: instance space first, then the schema spaces — purging a schema space first can orphan instances the purge then can't see.

# 1) instance space — nodes, edges, both PDF CogniteFiles + content, time-series datapoints,
#    diagram-annotation edges, the Equipment Health Profile node
uv run cdf data purge space isp_<YOURNAME>_TRN \
  --include-space --delete-datapoints --delete-file-content

# 2) enterprise schema space
uv run cdf data purge space ssp_<YOURNAME>_TrainingCore_edm --include-space

# 3) solution schema space
uv run cdf data purge space ssp_<YOURNAME>_MaintenanceInsight_sdm --include-space

When Toolkit asks you to confirm, type the project name exactly (not y):

<your-cdf-project>

⚠️ [COMMON MISTAKE] Typing y instead of the project name, or piping the confirmation (printf … | cdf …). Both break the confirm flow — run it interactively and type the project name. This is the reason the SDK teardown notebook can't do the space purge for you: a notebook ! cell isn't a TTY.

🚧 [LIMITS] cdf data purge space builds its delete plan from space statistics, which can lag and report 0 right after heavy writes — the purge then no-ops and --include-space fails with "contain nodes or edges". Wait ~1–2 minutes and re-run it interactively; if it persists, delete the space from the CDF UI rather than looping.

19.2 [ACTION] Remove your location filter¶

The location filter loc_<YOURNAME>_TRN lives under the CDF apps API, and the public Cognite SDK exposes no delete for it — which is exactly why the notebook leaves it to cdf. Clean it against your own config:

uv run cdf build --config-yaml training/config.<YOURNAME>-training.yaml
uv run cdf clean --cdf-project <your-cdf-project> --include locations

(If a confirmation prompt appears, type <your-cdf-project> again.)

19.2b [ACTION] The leftovers nothing warns you about¶

Two things survive a tidy teardown and will not show up unless you go looking.

Function zips. Deploying a Cognite Function uploads its code as a classic file. Deleting the function does not delete that file — you are left with one orphaned zip per function, named after the function's external ID.

🟢 [ACTION]

leftovers = [f.external_id for f in client.files.list(limit=1000)
             if (f.external_id or "").startswith(f"fnc_{YOURNAME}_")
             or (f.external_id or "").startswith(f"file_{YOURNAME}_")]
print(leftovers)
client.files.delete(external_id=leftovers)

✅ [VERIFY] Re-run the list — it comes back empty. Five function zips plus the classic OBJ from Chapter 09.

Your data set. Data sets cannot be deleted in CDF, ever. Archiving is the clean end state, which the teardown notebook already does for you — but know that it is a platform limit, not a step you forgot.

19.3 [VERIFY] Nothing of yours is left¶

Fusion → Data management → Spaces: isp_<YOURNAME>_TRN, ssp_<YOURNAME>_TrainingCore_edm, ssp_<YOURNAME>_MaintenanceInsight_sdm are gone.
Re-run the verify cell in 06_teardown.ipynb: your functions / transformations / 3D model lists are empty, and the data set reports archived (data sets can never be hard-deleted — archived is the clean end state).
Keep config.<YOURNAME>-training.yaml — do not delete it.

That's a clean exit. 🎉

← Chapter 18 — PR & Merge