fix: re-seed the elementary graph cache on access

[Daniel-Wiszowaty]

Fixes the crash class in #1058.

Problem

get_cache / set_cache / setdefault_cache all read graph["elementary"], which is written exactly once, by init_elementary_graph() from the on-run-start hook. When that seed is absent the accessors fall through to their none default, and the five call sites that chain a method onto the result take the whole run down:

call site	expression	error
macros/edr/materializations/test/test.sql:382	get_cache("elementary_test_results").update(...)	'None' has no attribute 'update'
macros/edr/materializations/test/test.sql:54	get_cache("elementary_test_failed_row_counts").update(...)	same
macros/edr/tests/test_metric_stability.sql:196	get_cache("tables").get("metrics")...	'None' has no attribute 'get'
macros/edr/data_monitoring/anomaly_detection/store_metrics_in_cache.sql:3	get_cache("tables").get("metrics")...	same
macros/edr/data_monitoring/schema_changes/store_schema_snapshot_tables_in_cache.sql:2	get_cache("tables").get(...)	same

The first one runs for every test in the project, so a missing seed means every test fails to compile.

Fix

Route the three accessors through one helper that re-seeds the full default structure when the key is absent, and have init_elementary_graph() share the same defaults macro so there is a single source of truth. A lost seed now degrades (empty cache) instead of raising.

This matches the two suggestions on #1058 — it is option (2), applied in one place rather than at each call site, so future chained call sites are safe too.

How this is reachable today

Under dbt-core's v2 (Fusion) parser, the package's on-run-start hook never executes, so the seed is never written at all. Reproduction on a stock dbt-duckdb project with elementary 0.26.0 and dbt-core 1.12.4 — same project, same command, only the parser differs:

$ dbt run
Elementary: Runtime data: {... "elementary_version": "0.26.0" ...}
1 of 1 START hook: elementary.on-run-start.0 ................... [RUN]
1 of 1 OK hook: elementary.on-run-start.0 ...................... [OK in 0.01s]
1 of 1 START hook: elementary.on-run-end.0 ..................... [RUN]
1 of 1 OK hook: elementary.on-run-end.0 ........................ [OK in 0.32s]

$ dbt run --use-v2-parser
(no hook lines at all — neither hook runs)

dbt build reports 2 project hooks on the classic path and omits them entirely under --use-v2-parser.

To be clear about scope: the hooks not running is a dbt-side problem, not something this PR fixes. It also means on-run-end artifact upload is skipped, which is a separate and larger issue. This PR only stops elementary from crashing when the seed is absent, whatever the reason.

Verification

Stock dbt-duckdb project, elementary installed as a local package, dbt-core 1.12.4 + dbt-core-experimental-parser 2.0.1.

Reproducing the exact failing expression from handle_dbt_test:

# before
$ dbt run-operation probe_cache
[ERROR]: Encountered an error while running operation: Compilation Error
  'None' has no attribute 'update'

# after
$ dbt run-operation probe_cache
PROBE ok, cache now = {'x': 1}

where probe_cache is just {% do elementary.get_cache("elementary_test_results").update({"x": 1}) %}.

No regression on a full build:

	before	after
dbt build (classic)	PASS=35, 2 project hooks	PASS=35, 2 project hooks
dbt build --use-v2-parser	PASS=33, no hooks	PASS=33, no hooks

Files formatted with the repo's sqlfmt 0.29.0 pre-commit hook.

I have not run the warehouse integration suite — no access to those targets. Happy to adjust if you would rather see setdefault_cache at each call site instead.

Summary by CodeRabbit

• Refactor
  • Standardized initialization of the elementary graph cache to use shared default settings.
  • Preserved existing graph values when already present.
  • Ensured default graph settings are restored when the parser rebuilds the graph.

[github-actions]

👋 @Daniel-Wiszowaty
Thank you for raising your pull request.
Please make sure to add tests and document all user-facing changes.
You can do this by editing the docs files in the elementary repository.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: e8f48c48-76f8-4a3f-8dee-634b3210c33e

📥 Commits

Reviewing files that changed from the base of the PR and between 9a7e2b5 and 8ea904b.

📒 Files selected for processing (2)

• macros/edr/tests/on_run_start/init_elementary_graph.sql
• macros/utils/graph/cache.sql

Included review availability: Your plan provides up to 4 included reviews per hour; 3 remain after this review.

---

📝 Walkthrough

Walkthrough

The change extracts the elementary graph defaults into a reusable macro. Graph initialization and cache access now use these defaults when the elementary graph entry is absent.

Changes

Elementary graph cache

Layer / File(s)	Summary
Centralize elementary graph defaults macros/edr/tests/on_run_start/init_elementary_graph.sql	The default elementary graph dictionary moves into elementary_graph_defaults(). init_elementary_graph() calls this macro while preserving graph.setdefault behavior.
Apply defaults during cache access macros/utils/graph/cache.sql	get_elementary_cache() initializes the elementary cache with the shared defaults. set_cache, get_cache, and setdefault_cache use this accessor.

Priority: ⬇️ Low

Estimated code review effort: 2 (Simple) | ~10 minutes

Change: Bug fix

Merge Risk: ⚪ Minimal · up to 8ea90

The cache accessors now restore the complete default cache shape when the elementary cache is absent. No current merge-blocking risk is identified.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely describes the main change: restoring the Elementary graph cache when accessed without its seed.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check. Docstring coverage is scoped to functions touched by this diff. Analyzed 0 functions across 0…
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches 💡 1

🛠️ Fix failing CI checks 💡

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}

[Daniel-Wiszowaty]

Filed the hook half of this upstream: dbt-labs/dbt#16310 — --use-v2-parser never executes on-run-start / on-run-end, for root-project and package hooks alike, silently.

That is why the seed is missing here, so if it gets fixed this issue's crash stops being reachable via that path. The re-seeding in #1060 is still worth having as defence in depth, but the hook bug is the one that actually restores Elementary under the v2 parser — on-run-end artifact upload is skipped too, which no change inside this package can work around.

Worth noting the v2 parser already reads on-run-start (it parses CREATE FUNCTION there to register UDFs), so it sees the hooks and just doesn't run them.