Release date: 2026-06-04
Package: query-doctor 0.5.0
Supported production diagnostic engine: Apache Impala
Release focus: deterministic diagnostic-loop hardening and raw-free evidence
handoff gates
0.5.0 is a safety and diagnostic-readiness release. It keeps Apache Impala as the only production-supported triage engine while hardening the local-first diagnostic loop, selected-case action contracts, release gates, and private preview evidence handoff paths.
- Recent Details recommendations now keep an analyst decision path: why the query deserves attention, where to inspect, what supported change direction to try, and how to verify with a comparable rerun.
- Query-shape and stats action cards now fail safe when older candidate data lacks comparable-rerun wording. Medium or High recommendations cannot render as EXPLAIN-only follow-ups.
- Stats optimization scoring now requires a complete evidence chain before a candidate can become Medium or High. Partial metadata plus estimate mismatch is no longer actionable unless supported missing or incomplete stats evidence is present.
- Representative Impala loop audits now include Details, trusted report revalidation, trusted optimizer artifacts, profile-evidence gates, diagnostic coverage, workload readiness, stats readiness, and optimizer funnel checks.
- Trusted optimizer artifact auditing now distinguishes trusted SQL drafts, trusted recommendations-only output, trusted no-rewrite output, and partial/untrusted artifacts without printing SQL, paths, filenames, or case identifiers.
- Workload diagnostics now share one action contract across queue and detail views, including verification metrics and local action-outcome feedback.
- Public release gates now include branch-history shape checks, current-tree public-safety scans, fixture provenance checks, README screenshot provenance, and stricter release-facing documentation audits.
- Trino remains private-preview groundwork only. 0.5.0 adds stricter raw-free handoff gates for sanitized evidence packages, one-query pruned QueryInfo imports, compact diagnosis JSON, source-version requirements, smoke summaries, and retained handoff-suite manifests. These paths do not create public Trino engine support, live Recent scans, live Query ID product diagnosis, Details or trusted-report output, optimizer behavior, metadata collection, or generated Trino SQL.
- Spark compact History Server and evidence-package paths remain experimental contract-shaping surfaces. 0.5.0 adds package readiness verdicts, compact diagnosis boundary drift checks, fixture export manifests, strict handoff audits, and optional raw-free machine summary JSON for handoff readiness. These paths do not create public Spark engine support, Recent scans, Details or trusted-report output, optimizer behavior, raw event-log handling, raw SQL/plan display, environment/log dumps, or Spark job execution.
- Normalized engine facts remain a raw-free contract seam. They are not the product engine registry and do not replace the production Impala analyzer, presenter, report, or optimizer facts.
- Apache Impala remains the only production-supported diagnostic engine.
- Cloudera Manager remains the full Recent discovery/profile/metrics/events provider for Impala workflows.
- Direct Impala supports bounded Recent scans, Running scans, and one explicit Known Query ID through impalad daemon endpoints, without Cloudera Manager events.
- Direct JSON profile,
/profile_docs, and/admission?jsonprobes remain optional compatibility surfaces. Missing old-cluster endpoints degrade to unknown or not-configured unless the user explicitly requires that source. - Optional Prometheus runtime metrics remain bounded direct-Impala context only when explicitly configured.
- Web Recent scan and Running scan workflows still do not auto-run reports, LLM narratives, or optimizer jobs.
- Browser-visible UI and trusted reports still must not expose raw SQL, raw profiles, raw metadata, local paths, process logs, secrets, model names, runtime internals, or raw artifact filenames.
- Query Optimizer behavior remains read-only. Trusted SQL drafts still require Python-owned recipes, deterministic execution, and strict validation.
- Report validators now have broader adversarial English/Russian coverage for unsupported causal wording, stale-statistics claims, stats-maintenance overclaims, row/cardinality estimate direction wording, and inline SQL-like prose.
- Browser/log host redaction now uses a linear scanner for free-text FQDNs,
host-like single-label names, and bare
host:portvalues, preserving stable aliases while removing a CodeQL polynomial-regex risk. - README screenshot currency was reviewed for this release. Existing public screenshots still come from the synthetic demo pack and match the documented first-screen demo path.
The 0.5.0 release candidate should be validated with:
PUBLIC_RELEASE=1 scripts/local_gate.shpre-commit run --all-filespython -m pytest -qpython -m pytest -q tests/test_pyproject.pypython -m buildpython -m twine check dist/*- public-release preflight with history scanning against the reviewable release branch
- synthetic demo pack generation and README screenshot provenance review
Focused pre-release Impala checks for this candidate included representative raw-free Details and stats audits over a retained sanitized Recent summary. The aggregate loop still tracks remaining calibration work for profile classifier parity, diagnostic coverage thresholds, and optimizer mixed-track grouping; that work remains post-0.5.0 diagnostic-loop hardening rather than a release blocker.
- Upgrade with
pip install --upgrade query-doctorafter 0.5.0 is published. - Regenerate the local demo pack with
query-doctor-demo --out <demo-dir> --overwrite; use the printedQUERY_DOCTOR_ACTION_OUTCOMES_PATHvalue when launchingquery-doctor-web. - Existing Impala workflows, configuration files, report safety rules, and optimizer trust boundaries remain compatible.