:hedgehog: PostHog is the leading platform for building self-driving products. Our developer tools – AI observability, analytics, session replay, flags, experiments, error tracking, logs, and more – capture all the context agents need to diagnose problems, uncover opportunities, and ship fixes. Steer it all from Slack, web, desktop, or the MCP.
Python ab-testingai-analyticsanalyticscdpdata-warehouseexperimentsfeature-flagsjavascriptproduct-analyticspythonreactsession-replaysurveystypescriptweb-analytics
fix: temporal bk (#124241)
Alex V8017fddf2afb
diff --git a/products/business_knowledge/backend/temporal/learning/activities/analyze.py b/products/business_knowledge/backend/temporal/learning/activities/analyze.pyindex 772b5abff439..8bbdc75174f0 100644--- a/products/business_knowledge/backend/temporal/learning/activities/analyze.py+++ b/products/business_knowledge/backend/temporal/learning/activities/analyze.py@@ -111,6 +111,11 @@ class LearningAnalysisError(RuntimeError):self.code = code+class LearningOutputTruncated(LearningAnalysisError):+ def __init__(self, stage: str) -> None:+ super().__init__(f"{stage}_output_truncated")++class _SupersessionUnavailable(Exception):"""The conflicting source is not one this job may replace, so the candidate must not be published."""@@ -248,7 +253,29 @@ def _invoke_structured_model(tool_choice={"type": "tool", "name": tool_name},extra_headers={"X-PostHog-Privacy-Mode": "true"},)- return _parse_tool_result(response.content, output_model, stage)+ try:+ return _parse_tool_result(response.content, output_model, stage)+ except (LearningAnalysisError, ValidationError) as error:+ if response.stop_reason == "max_tokens":+ logger.warning(+ "business_knowledge.learning.output_truncated",+ stage=stage,+ team_id=team.id,+ trace_id=trace_id,+ max_tokens=max_tokens,+ )+ raise LearningOutputTruncated(stage) from None+ if isinstance(error, LearningAnalysisError):+ content = response.content if isinstance(response.content, list) else []+ logger.error( # noqa: TRY400 — the error is re-raised, so Temporal records its traceback+ "business_knowledge.learning.invalid_output",+ stage=stage,+ team_id=team.id,+ trace_id=trace_id,+ stop_reason=response.stop_reason,+ block_types=[getattr(block, "type", type(block).__name__) for block in content],+ )+ raiseexcept LearningAnalysisError:raiseexcept Exception as error:@@ -545,6 +572,7 @@ def _finish_without_knowledge("low_confidence": "rejected_not_useful","learned_cap_reached": "rejected_learned_cap","stale_candidate": "rejected_stale_candidate",+ "output_truncated": "rejected_output_truncated",}.get(rejection_code, "rejected_not_useful")_increment_counter(metric_outcome)logger.info(@@ -698,7 +726,11 @@ def _analyze(run: KnowledgeLearningRun, input: AnalyzeLearningEvidenceInput) ->_resolve_learning_user(run.team)trace_id = str(run.id)- extracted = _extract_candidate(run.team, bundle, trace_id)+ try:+ extracted = _extract_candidate(run.team, bundle, trace_id)+ except LearningOutputTruncated:+ # Sampling runs at temperature 0, so a retry truncates again. A reply this long is not concise knowledge.+ return _finish_without_knowledge(run, rejection_code="output_truncated")rejection_code = _extraction_rejection(extracted)if rejection_code != "none":return _finish_without_knowledge(run, rejection_code=rejection_code)diff --git a/products/business_knowledge/backend/temporal/learning/schemas.py b/products/business_knowledge/backend/temporal/learning/schemas.pyindex a2d36372a34c..8a73ba56efc5 100644--- a/products/business_knowledge/backend/temporal/learning/schemas.py+++ b/products/business_knowledge/backend/temporal/learning/schemas.py@@ -21,6 +21,7 @@ RejectionCode = Literal["already_completed","learned_cap_reached","stale_candidate",+ "output_truncated",]LearningResult = Literal["knowledge_created", "no_knowledge", "ineligible", "superseded"]ExtractionRejectionCode = Literal["none", "case_specific", "not_useful", "unsupported", "low_confidence"]diff --git a/products/business_knowledge/backend/tests/test_learning_analyzer.py b/products/business_knowledge/backend/tests/test_learning_analyzer.pyindex 46efc15452b8..281af3fcce71 100644--- a/products/business_knowledge/backend/tests/test_learning_analyzer.py+++ b/products/business_knowledge/backend/tests/test_learning_analyzer.py@@ -577,6 +577,50 @@ class TestLearningAnalyzer:client.with_options.assert_called_once_with(timeout=_MODEL_TIMEOUT_SECONDS, max_retries=0)assert entered.messages.create.call_args.kwargs["extra_headers"] == {"X-PostHog-Privacy-Mode": "true"}+ @pytest.mark.parametrize(+ "content",+ [+ pytest.param([], id="no_tool_block"),+ pytest.param(+ [+ ToolUseBlock(+ id="toolu_test",+ name="ExtractedKnowledge",+ input={"canonical_topic": "Refund policy"},+ type="tool_use",+ )+ ],+ id="partial_tool_block",+ ),+ ],+ )+ def test_truncated_extraction_completes_without_knowledge_instead_of_failing(+ self, team: Team, content: list[ToolUseBlock]+ ) -> None:+ run, input = _setup_sync(team)+ provider = _Provider(EvidenceBundle(replies=("Refunds are available within 30 days.",)))+ client, entered = _gateway_client()+ truncated = MagicMock()+ truncated.stop_reason = "max_tokens"+ truncated.content = content+ entered.messages.create.return_value = truncated++ with (+ patch(f"{_MODULE}.get_learning_provider", return_value=provider),+ patch(f"{_MODULE}.build_ai_gateway_anthropic_client", return_value=client),+ patch(f"{_MODULE}.logic.create_generated_knowledge_document") as publish,+ ):+ result = analyze_learning_evidence(input)++ run.refresh_from_db()+ assert result.result == "no_knowledge"+ assert result.rejection_code == "output_truncated"+ assert run.result == LearningRunResult.NO_KNOWLEDGE+ assert run.status == "completed"+ assert run.error == ""+ assert entered.messages.create.call_count == 1+ publish.assert_not_called()+def test_missing_gateway_config_fails_before_the_model(self, team: Team) -> None:run, input = _setup_sync(team)provider = _Provider(EvidenceBundle(replies=("Refunds are available within 30 days.",)))