[NA] [SDK] [GHA] test: stabilize Gemini + agentic-judge integration tests#7405
Merged
alexkuzmik merged 2 commits intoJul 8, 2026
Merged
Conversation
…ests Three related changes to stop provider-driven flakiness in the real-model integration suites: - Agentic LLM-judge tests: run the Anthropic lanes against claude-sonnet-4-6 instead of claude-haiku-4-5. On the agentic path (tools in the request, so response_format is best-effort) Haiku wrapped verdicts in prose/```json fences; Sonnet follows the structured-output contract reliably (the one-shot suite already used Sonnet). - GenAI + ADK test matrices: run only the oldest and newest supported Python versions (3.10, 3.14). These suites make live Vertex AI calls against the shared opik-sdk-tests GCP project; the full 3.10-3.14 matrix re-multiplied those calls into 429 RESOURCE_EXHAUSTED. Endpoints cover version-specific breakage at a fraction of the request volume; max-parallel cap kept. - Remove the broken GenAI rate-limit retry helpers: _is_rate_limit_error checked exception.response.status_code, which raises AttributeError on google-genai's ClientError (no such attribute), so the retry crashed instead of retrying. With the matrix trim as the 429 mitigation, the dead helper and its decorators are removed rather than repaired. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2 tasks
2 tasks
Contributor
⏱️ pre-commit per-hook timing
⏭️ 36 skipped (no matching files changed)
|
alexkuzmik
marked this pull request as ready for review
July 8, 2026 18:39
alexkuzmik
deleted the
aliaksandrk/NA-stabilize-gemini-and-agentic-judge-tests
branch
July 8, 2026 18:40
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Details
Stops provider-driven flakiness in the real-model integration suites, in three related changes:
claude-sonnet-4-6instead ofclaude-haiku-4-5. On the agentic path (tools are in the request, soresponse_formatis best-effort) Haiku wrapped verdicts in prose /```jsonfences and failed to parse; Sonnet follows the structured-output contract reliably (the one-shot judge suite already used Sonnet).3.10,3.14) instead of the full3.10–3.14matrix. These suites make live Vertex AI calls against the sharedopik-sdk-testsGCP project; the full matrix re-multiplied those calls into429 RESOURCE_EXHAUSTED. Endpoints still catch version-specific breakage at a fraction of the request volume; themax-parallelcap is kept._is_rate_limit_errorcheckedexception.response.status_code, which raisesAttributeErroron google-genai'sClientError(no such attribute) — so the retry crashed instead of retrying. With the matrix trim as the 429 mitigation, the dead helper and its decorators are removed rather than repaired.Change checklist
Issues
AI-WATERMARK
AI-WATERMARK: yes
Testing
ruff check+ruff format --checkon the changed genai test files → clean (no unused imports / undefined names left after removing the retry helpers).pytest tests/library_integration/genai/ --collect-only→ 20 tests collect (imports resolve;genai_errorsstill used by an unrelatedpytest.raises).python_versionresolves to['3.10', '3.14']for both suites.Not run locally: the genai/adk suites make live Vertex AI calls (require GCP creds) and the agentic-judge suite calls real Anthropic models (require
ANTHROPIC_API_KEY); those are validated by CI.Documentation
N/A — no user-facing or API changes.