feat(cabq): cabq backfill implementation

[njones-icasa]

Why

This PR addresses the following problem / context:

• Need backfill for CABQ source to enable re-fetching historical data.

How

Implementation summary - the following was changed / added / removed:

• Implemented cabq/backfill.py.
• Made changes to REST API calls in cabq/dlt_pipeline.py to query based on a time range.
• Made changes to shared/backfill.py to enable a source to use different value types as location ids instead of just int.
• Unit tests for the above.

Notes

Any special considerations, workarounds, or follow-up work to note?

• None

[github-actions]

Your pull request is automatically being deployed to Dagster Cloud.

Location	Status	Link	Updated
aqueduct_dagster_defs_definitions		View in Cloud	Sep 01, 2026 at 07:37 PM (UTC)

[likithabommasani21]

The bare raise here is fragile and loses the underlying error info in the common case where _fetch_locations returns (None, err) without raising. In that case the bare raise produces RuntimeError: No active exception to reraise and err is discarded — confirmed by running it directly.

Minimal fix — fold the locations is None check into the same try, so the existing except Exception: client.close(); raise handles both cases (an actual raised exception, and this returned-None case) with one close() call instead of two, and the real err reason included:

try:
locations, err = _fetch_locations(client)
if locations is None:
raise RuntimeError(f"Backfill fetch failed to receive locations: {err}")
except Exception:
client.close()
raise

[njones-icasa]

adjusted the error logging as suggested

[likithabommasani21]

looks good

[likithabommasani21]

CABQ's _fetch_readings_for_location builds the encoded query string by hand; hydrovu uses httpx's params (client.get(path, params=...)). This divergence is the issue — please switch CABQ to use params= for safety and consistency.

Note: tests/sources/cabq/test_dlt_pipeline.py assert the exact encoded URL string and will need to be updated to check parsed params (e.g. via httpx.URL(...).params) instead of raw string equality.

[njones-icasa]

has been addressed, and I updated the call for fetch locations to do the same

[likithabommasani21]

Found and fixed one more issue in the params= migration — the query still joined clauses with a literal + (leftover from when the URL was hand-built), which params= now encodes as a literal %2B instead of a space. This broke every real query with a 400 from the API (and crashed with an unhandled KeyError: 'features' instead of a clean error, since the API returns HTTP 200 with an error body).

This wasn't backfill-only — _fetch_readings_for_location is also called from raw_cabq_readings, the daily production ingest asset, so this would have broken the daily pipeline too.

Fixed by joining clauses with real spaces instead of +.
Before merging this PR please Verify:

• Direct call against the live API — returns real data, no more KeyError
• Full cabq_backfill_refetch job run — succeeds
• Full daily asset chain (raw_cabq_readings → canonical_bundles_cabq → frost_load_cabq) — succeeds end-to-end
• Tests/ruff/mypy all clean

[njones-icasa]

ah, fair enough, honestly thought that the + was still needed. Tested the changes locally and can confirm that everything is working.

[likithabommasani21]

Looks good to me!
Verify all the CABQ pipelines works as excepted in dagster before merging