Map the SQL-tester framework and the end-to-end regression surface used to validate FE/BE behavior against a running StarRocks cluster.
cd test && python3 run.py -vcd test && python3 run.py -d sql/<suite> -vcd test && python3 run.py -d sql/<suite>/T/<case> -rcd test && python3 run.py -l
- Use
${uuid0}for objects that must be unique across runs. - Clean up created objects in each case.
- Prefer focused tags and filters over whole-tree runs during iteration.
T/files define statements;R/files define validation expectations.- Use
[ORDER],[UC], shell commands, and helper functions deliberately so expectations stay deterministic. - Cloud/native mode tags should reflect the actual runtime dependency of a case.
- Suite ownership and change-based selection are not encoded mechanically.
- Flake policy and retry evidence are not tracked in repo-local metadata.
- SQL-tester artifacts are not yet normalized into an agent-legible eval registry.