Fix infinite hang when Dremio job is cancelled (CANCELED vs CANCELLED)

[shanmugapriya5c]

Summary

Fix infinite hang on cancelled Dremio jobs by handling the CANCELED job state (single L)

Description

[DremioCursor._populate_rowcount()] polls Dremio job status in a while True loop and treats FAILED, COMPLETED, and CANCELLED (double L) as terminal states. However, Dremio's REST API returns the American spelling CANCELED (single L) for cancelled jobs — whether cancelled by a user, by the engine (e.g. low memory), or by hitting a workload-management queue timeout.

Because the loop only checked for CANCELLED, a real CANCELED job never matched any terminal branch, so the poller looped forever. dbt never raised, never failed, and never returned — the run hung indefinitely (and, for orchestrators like Dagster/Airflow, the step neither failed nor cancelled). This is reproducible by letting a query exceed its WLM queue timeout, or by cancelling the job from the Dremio Jobs UI.

Reference — Dremio Job API jobState enum (final states are COMPLETED, CANCELED, FAILED): https://docs.dremio.com/cloud/reference/api/job/

Changes in [cursor.py]:

Match both CANCELED and CANCELLED as terminal cancellation states and raise (previously only CANCELLED).
Surface Dremio's cancellationReason in the error message when present (e.g. queue-timeout reason), falling back to "Job was cancelled" for backward compatibility with the existing message.
Updated the stale loop comment (COMPLETE, CANCELLED → COMPLETED, CANCELED).
The fix is intentionally minimal and preserves the existing exception type/message contract.

Test Results

All tests are passing

Changelog

• Added a summary of what this PR accomplishes to CHANGELOG.md

Contributor License Agreement

• Please make sure you have signed our Contributor License Agreement, which enables Dremio to distribute your contribution without restriction.

Related Issue

NA

[CLAassistant]

All committers have signed the CLA.