repo stringclasses 454
values | file_path stringlengths 5 201 | extension stringclasses 1
value | content stringlengths 8 509k | num_lines int64 3 16.9k | size_bytes int64 8 511k |
|---|---|---|---|---|---|
posthog | posthog/temporal/ai_observability/test_run_session_evaluation.py | .py | import uuid
from datetime import UTC, datetime, timedelta
import pytest
from freezegun import freeze_time
from unittest.mock import Mock, patch
from asgiref.sync import async_to_sync
from temporalio.exceptions import ApplicationError
from posthog.schema import LLMTrace, LLMTraceEvent
from posthog.hogql.constants im... | 469 | 22,594 |
posthog | posthog/temporal/ai_observability/test_metrics.py | .py | import pytest
from unittest.mock import MagicMock, patch
from posthog.temporal.ai_observability.metrics import (
EVAL_ACTIVITY_TYPES,
EVAL_WORKFLOW_TYPES,
EvalsMetricsInterceptor,
ExecutionTimeRecorder,
)
class TestExecutionTimeRecorder:
def test_records_completed_status_on_success(self):
... | 135 | 5,680 |
posthog | posthog/temporal/ai_observability/run_trace_evaluation.py | .py | """Trace-level evaluation workflow.
The evaluation scheduler starts this workflow when the first $ai_generation matching a
trace-target evaluation arrives. The workflow id (`llma-trace-eval-{evaluation_id}-{trace_id}`)
plus the USE_EXISTING conflict policy make every later matching event of the same trace a
no-op whil... | 792 | 32,073 |
posthog | posthog/temporal/ai_observability/test_team_capture.py | .py | import time
from datetime import UTC, datetime
from posthog.test.base import BaseTest
from unittest.mock import MagicMock, patch
from parameterized import parameterized
from posthog.api.capture import CaptureInternalError
from posthog.models.team import Team
from posthog.temporal.ai_observability.team_capture import... | 90 | 3,371 |
posthog | posthog/temporal/ai_observability/run_session_evaluation.py | .py | """Session-level evaluation: fetch, payload caps, Hog globals and judge transcript.
Sits beside `run_trace_evaluation` rather than inside it because the shared emit activity lives
there, and this module already imports a constant from it; if `run_trace_evaluation` ever imported
something back from here, the two would ... | 632 | 27,896 |
posthog | posthog/temporal/ai_observability/llm_endpoint.py | .py | """ChatOpenAI client for the labeling/report agents, routed through the internal
Go ai-gateway when AI_GATEWAY_URL + AI_GATEWAY_API_KEY are set, else direct to OpenAI.
The raw-SDK builders for summarization live in ``posthog.llm.gateway_client`` (importing
from this package would pull in the temporal workflow graph an... | 94 | 3,513 |
posthog | posthog/temporal/ai_observability/coordinator_metrics.py | .py | """Shared Prometheus metrics for LLMA coordinator workflows.
Used by both the summarization and clustering coordinators to track
team discovery and per-team success/failure across pipeline runs.
"""
from posthog.temporal.ai_observability.metrics import get_metric_meter
def record_teams_discovered(count: int, pipeli... | 48 | 1,838 |
posthog | posthog/temporal/ai_observability/evaluation_types.py | .py | from typing import Any, Literal, NotRequired, Required, TypedDict
class EvaluationActivityResult(TypedDict, total=False):
"""Result produced by evaluation execution activities.
`total=False` is used as the default so individual fields opt in via `Required` /
`NotRequired`, making the contract honest abou... | 46 | 2,131 |
posthog | posthog/temporal/ai_observability/evaluation_errors.py | .py | """Shared taxonomy for evaluation execution errors.
The split between user-actionable and PostHog-owned errors is load-bearing: most user errors
disable the evaluation without failing the Temporal activity, while PostHog-owned errors still
raise so Temporal/Grafana stays useful. `hog_input_error` is the one user error... | 253 | 9,384 |
posthog | posthog/temporal/ai_observability/team_discovery.py | .py | """
Team discovery activity for AI observability workflows.
Provides dynamic team discovery that combines a guaranteed allowlist
with a configurable random sample of teams that have AI events.
Config is read from the `llm-analytics-clustering-workflows` feature flag
payload so it can be changed from the PostHog UI wi... | 214 | 8,785 |
posthog | posthog/temporal/ai_observability/test_team_discovery.py | .py | import math
from contextlib import asynccontextmanager
import pytest
from unittest.mock import patch
from parameterized import parameterized
from posthog.temporal.ai_observability.team_discovery import (
DEFAULT_DISCOVERY_LOOKBACK_DAYS,
DEFAULT_GUARANTEED_TEAM_IDS,
DEFAULT_SAMPLE_PERCENTAGE,
TeamDisc... | 342 | 14,575 |
posthog | posthog/temporal/ai_observability/message_utils.py | .py | """Utilities for extracting and formatting LLM messages from event properties."""
import json
from typing import Any, Union
def extract_text_from_messages(messages: Union[str, list, dict, None]) -> str:
"""
Extract readable text from LLM message structures.
Handles common message formats from various LL... | 333 | 14,120 |
posthog | posthog/temporal/ai_observability/__init__.py | .py | from posthog.temporal.ai_observability.eval_reports.activities import (
check_count_triggered_eval_report_activity,
check_count_triggered_eval_reports_activity,
deliver_report_activity,
fetch_count_triggered_eval_report_candidates_activity,
fetch_due_eval_reports_activity,
prepare_report_context... | 194 | 7,283 |
posthog | posthog/temporal/ai_observability/shared_activities.py | .py | """Shared Temporal activities used by multiple llm_analytics workflows."""
import asyncio
from dataclasses import dataclass
from typing import Any, Literal, cast
from temporalio import activity
AnalysisLevel = Literal["trace", "generation", "evaluation"]
@dataclass
class FetchAllClusteringFiltersInput:
team_id... | 98 | 3,073 |
posthog | posthog/temporal/ai_observability/team_capture.py | .py | """Emitting AI observability events on behalf of a team, without a Postgres read per event.
Every emitted `$ai_evaluation` / `$ai_tag` event needs the team's `api_token` to call
`capture_internal`. Those emit closures run under `database_sync_to_async(thread_sensitive=False)`,
so each one lands on a fresh executor thr... | 101 | 3,781 |
posthog | posthog/temporal/ai_observability/model_resolution.py | .py | """Resolves which provider, model and key an eval or tagger run should use.
`model_configuration` is optional on both `Evaluation` and `Tagger`. A null value is not
"unconfigured" — it means "defer to the team's active key". `model_spec()` turns the (possibly
null) serialized config into a `ModelSpec` whose `resolve()... | 130 | 4,750 |
posthog | posthog/temporal/ai_observability/test_run_evaluation.py | .py | import json
import uuid
from datetime import UTC, datetime
from typing import Any
import pytest
from unittest.mock import MagicMock, patch
from asgiref.sync import sync_to_async
from parameterized import parameterized
from temporalio import activity
from temporalio.api.enums.v1 import EventType
from temporalio.except... | 2,325 | 95,338 |
posthog | posthog/temporal/ai_observability/evaluation_event_io.py | .py | from typing import Any
from posthog.dataclasses import frozen
@frozen
class EventIO:
input_raw: Any
output_raw: Any
def extract_event_io(event_type: str, properties: dict[str, Any]) -> EventIO:
"""Extract raw input and output values from event properties.
Returns an `EventIO` for use in Hog eval g... | 52 | 2,215 |
posthog | posthog/temporal/ai_observability/test_message_utils.py | .py | """Tests for message extraction utilities."""
import json
import pytest
from posthog.temporal.ai_observability.message_utils import extract_text_from_messages, format_tool_definitions
class TestExtractTextFromMessages:
def test_simple_string(self):
"""Test extraction from simple string"""
resul... | 695 | 28,972 |
posthog | posthog/temporal/ai_observability/run_aggregate_evaluation.py | .py | """Aggregate evaluation workflow: a settling phase in front of the shared trace-evaluation body.
Successor to `run-trace-evaluation`. The scheduler starts this workflow once per
(evaluation, trace) pair on the first condition-matching generation. Under the fixed_window
strategy the workflow just sleeps for the configu... | 590 | 27,852 |
posthog | posthog/temporal/ai_observability/evaluation_sentiment.py | .py | import json
import asyncio
from typing import Any
from temporalio import activity
from temporalio.exceptions import ApplicationError
from posthog.temporal.ai_observability.evaluation_event_io import extract_event_io
from posthog.temporal.ai_observability.evaluation_types import EvaluationActivityResult
from posthog.t... | 108 | 4,137 |
posthog | posthog/temporal/ai_observability/run_evaluation.py | .py | import json
from datetime import timedelta
from typing import Any, NotRequired, Required, TypedDict
from django.conf import settings
import temporalio
from temporalio.common import RetryPolicy, WorkflowIDReusePolicy
from temporalio.exceptions import ApplicationError
from posthog.temporal.ai_observability.evaluation_... | 401 | 16,566 |
posthog | posthog/temporal/ai_observability/test_evaluation_payload.py | .py | import pytest
from posthog.temporal.ai_observability.evaluation_payload import (
PAYLOAD_BYTES_EXPR,
payload_budget_bytes,
should_skip_for_payload,
)
class TestPayloadBudget:
def test_budget_tracks_the_targets_judge_budget(self):
assert payload_budget_bytes(500_000) > payload_budget_bytes(150... | 38 | 1,677 |
posthog | posthog/temporal/ai_observability/evaluation_workflow_activities.py | .py | import json
from dataclasses import dataclass
from datetime import UTC, datetime
from typing import Any
from django.conf import settings
from django.db import transaction
import structlog
import temporalio
from structlog.contextvars import bind_contextvars
from posthog.api.capture import CaptureInternalError
from po... | 376 | 15,323 |
posthog | posthog/temporal/ai_observability/metrics.py | .py | import typing
import datetime as dt
from django.conf import settings
from temporalio import activity, workflow
from temporalio.worker import (
ActivityInboundInterceptor,
ExecuteActivityInput,
ExecuteWorkflowInput,
Interceptor,
WorkflowInboundInterceptor,
WorkflowInterceptorClassInput,
)
from... | 259 | 10,512 |
posthog | posthog/temporal/ai_observability/test_evaluation_hog.py | .py | import json
import pytest
from parameterized import parameterized
from temporalio.exceptions import ApplicationError
from posthog.temporal.ai_observability.evaluation_errors import is_terminal_user_error_result
from posthog.temporal.ai_observability.evaluation_hog import execute_hog_eval_bytecode, finalize_hog_eval_... | 128 | 5,220 |
posthog | posthog/temporal/ai_observability/test_model_resolution.py | .py | import uuid
import pytest
from temporalio.exceptions import ApplicationError
from posthog.models import Organization, Team
from posthog.temporal.ai_observability.model_resolution import DefaultModelSpec, ExplicitModelSpec, model_spec
from products.ai_observability.backend.llm import DEFAULT_MODEL_BY_PROVIDER
from p... | 140 | 5,861 |
posthog | posthog/temporal/ai_observability/test_run_tagger.py | .py | import json
import uuid
from datetime import datetime
from typing import Any
import pytest
from unittest.mock import MagicMock, patch
from temporalio.exceptions import ApplicationError
from posthog.api.capture import CaptureInternalError
from posthog.models import Organization, Team
from products.ai_observability.b... | 778 | 29,674 |
posthog | posthog/temporal/ai_observability/evaluation_payload.py | .py | """Payload-size budget shared by the trace and session evaluation fetches.
The event-count caps (`MAX_TRACE_EVAL_EVENTS`, `MAX_SESSION_EVAL_EVENTS`) bound how many rows an
evaluation pulls, which is a poor proxy for what actually costs memory: 2,500 near-limit events and
2,500 one-line events are the same number and w... | 67 | 3,313 |
posthog | posthog/temporal/ai_observability/test_run_aggregate_evaluation.py | .py | import uuid
from datetime import UTC, datetime, timedelta
from typing import Any
import pytest
from freezegun import freeze_time
from temporalio import activity
from temporalio.exceptions import ApplicationError
from temporalio.testing import WorkflowEnvironment
from temporalio.worker import UnsandboxedWorkflowRunner... | 712 | 32,294 |
posthog | posthog/temporal/ai_observability/test_run_trace_evaluation.py | .py | import json
import uuid
from datetime import UTC, datetime, timedelta
from typing import Any
import pytest
from freezegun import freeze_time
from unittest.mock import MagicMock, patch
from asgiref.sync import async_to_sync
from posthog.schema import LLMTrace, LLMTraceEvent
from posthog.hogql import ast
from postho... | 752 | 30,666 |
posthog | posthog/temporal/ai_observability/evaluation_llm_judge.py | .py | import json
from dataclasses import dataclass
from datetime import timedelta
from typing import Any
import structlog
import temporalio
import posthoganalytics
from pydantic import BaseModel, model_validator
from structlog.contextvars import bind_contextvars
from temporalio.common import RetryPolicy
from temporalio.exc... | 470 | 17,275 |
posthog | posthog/temporal/ai_observability/conftest.py | .py | import pytest
from posthog.temporal.ai_observability.team_capture import clear_team_api_token_cache
@pytest.fixture(autouse=True)
def _clear_team_api_token_cache():
# The token cache is per worker process, so it would otherwise leak across tests that
# reuse team ids from a rolled-back transaction.
clear... | 13 | 387 |
posthog | posthog/temporal/ai_observability/clustering_agent/__init__.py | .py | """Shared machinery for LangGraph cluster-labeling agents across analysis levels.
The trace/generation labeling agent (``trace_clustering/labeling_agent``) and the
evaluation labeling agent (``evaluation_clustering/labeling_agent``) are
structurally parallel — same ReAct loop, same ChatOpenAI guardrails, same
default-... | 79 | 3,042 |
posthog | posthog/temporal/ai_observability/eval_reports/types.py | .py | """Dataclass inputs/outputs for evaluation reports activities."""
import dataclasses
from typing import Any
@dataclasses.dataclass
class ScheduleAllEvalReportsWorkflowInputs:
buffer_minutes: int = 15
@dataclasses.dataclass
class CheckCountTriggeredReportsWorkflowInputs:
pass
@dataclasses.dataclass
class ... | 137 | 3,053 |
posthog | posthog/temporal/ai_observability/eval_reports/output_types.py | .py | """Trusted query definitions for evaluation report outcomes."""
from collections.abc import Mapping
from dataclasses import dataclass
@dataclass(frozen=True)
class EvaluationReportOutcomeDefinition:
outcomes: tuple[str, ...]
outcome_predicates: Mapping[str, str]
event_predicate: str
result_expression... | 58 | 2,714 |
posthog | posthog/temporal/ai_observability/eval_reports/workflow.py | .py | """Temporal workflows for evaluation reports."""
import json
import asyncio
from datetime import timedelta
from django.conf import settings
import temporalio.workflow
from structlog import get_logger
from temporalio.common import WorkflowIDReusePolicy
from temporalio.exceptions import WorkflowAlreadyStartedError
fr... | 438 | 19,161 |
posthog | posthog/temporal/ai_observability/eval_reports/emit_signal.py | .py | """Emit a signal per evaluation report run.
This is the only path from AI observability evaluations into Signals. Emitting one
signal per individual `$ai_evaluation` result flooded inboxes and said little; a
report already aggregates the underlying results into an analytical narrative
(trend, patterns, citations), so ... | 242 | 10,965 |
posthog | posthog/temporal/ai_observability/eval_reports/constants.py | .py | """Configuration constants for evaluation reports workflow."""
from datetime import timedelta
from temporalio.common import RetryPolicy
# Agent configuration
EVAL_REPORT_AGENT_MODEL = "gpt-5.2"
EVAL_REPORT_AGENT_RECURSION_LIMIT = 100
EVAL_REPORT_AGENT_TIMEOUT = 600.0 # 10 minutes
# Workflow names — all eval-report... | 96 | 4,213 |
posthog | posthog/temporal/ai_observability/eval_reports/schedule.py | .py | """Temporal schedule registration for evaluation reports."""
from dataclasses import asdict
from datetime import timedelta
from django.conf import settings
from temporalio.client import Client, Schedule, ScheduleActionStartWorkflow, ScheduleIntervalSpec, ScheduleSpec
from posthog.temporal.ai_observability.eval_repo... | 57 | 2,255 |
posthog | posthog/temporal/ai_observability/eval_reports/metrics.py | .py | """Prometheus metrics for the LLMA evaluation reports temporal workflows.
Follows the same pattern as trace_clustering/metrics.py.
Metrics are emitted via Temporal's built-in metric meter (activity/workflow context)
and scraped by the Prometheus endpoint on the worker pod.
"""
import typing
import datetime as dt
fro... | 210 | 7,830 |
posthog | posthog/temporal/ai_observability/eval_reports/delivery.py | .py | """Delivery logic for evaluation reports (email and Slack).
Content shape: `EvalReportContent` has a `title`, 1-6 titled `sections`, a list
of structured `citations`, and a `metrics` block. The renderers below build the
email HTML body and Slack Block Kit payloads from that shape.
"""
import re
from dataclasses impor... | 523 | 21,894 |
posthog | posthog/temporal/ai_observability/eval_reports/activities.py | .py | """Activities for evaluation reports workflow."""
import time
import datetime as dt
from collections import defaultdict
from typing import TYPE_CHECKING, NamedTuple
from zoneinfo import ZoneInfo
from django.db.models import Q
import temporalio.activity
from dateutil.rrule import rrulestr
from structlog import get_lo... | 855 | 36,336 |
posthog | posthog/temporal/ai_observability/eval_reports/targets.py | .py | """Evaluation-target descriptors shared by report queries, tools, and prompts.
One entry per target, so adding a target is adding a row here rather than another
`if target == ...` branch in every consumer.
"""
from dataclasses import dataclass
GENERATION_TARGET = "generation"
TRACE_TARGET = "trace"
SESSION_TARGET = ... | 80 | 2,670 |
posthog | posthog/temporal/ai_observability/eval_reports/report_agent/state.py | .py | """State definitions for the evaluation report agent."""
from typing import Annotated, TypedDict
from langgraph.graph.message import add_messages
from langgraph.managed import RemainingSteps
from posthog.temporal.ai_observability.eval_reports.report_agent.schema import EvalReportContent
class EvalReportAgentState(... | 45 | 1,542 |
posthog | posthog/temporal/ai_observability/eval_reports/report_agent/__init__.py | .py | from posthog.temporal.ai_observability.eval_reports.report_agent.graph import run_eval_report_agent
__all__ = ["run_eval_report_agent"]
| 4 | 137 |
posthog | posthog/temporal/ai_observability/eval_reports/report_agent/schema.py | .py | """Dataclasses for evaluation report content, metrics, and citations.
The v2 schema splits the report into:
- `evaluation_target`: the evaluated unit (`generation`, `trace`, or `session`)
- `metrics`: structured numeric data computed mechanically from ClickHouse,
the agent cannot fabricate these
- `title`: the agent... | 361 | 14,538 |
posthog | posthog/temporal/ai_observability/eval_reports/report_agent/tools.py | .py | """Tool definitions for the evaluation report agent.
Uses LangGraph's InjectedState pattern so tools can access graph state directly.
All query tools hit ClickHouse live via HogQL since the dataset is too large/variable
to pre-load. Output tools mutate `state["report"]` (an EvalReportContent instance)
via list append ... | 1,642 | 62,954 |
posthog | posthog/temporal/ai_observability/eval_reports/report_agent/prompts.py | .py | """System prompt construction for the evaluation report agent."""
from posthog.temporal.ai_observability.eval_reports.output_types import get_outcome_definition
from posthog.temporal.ai_observability.eval_reports.report_agent.schema import MAX_REPORT_SECTIONS
from posthog.temporal.ai_observability.eval_reports.targets... | 251 | 15,451 |
posthog | posthog/temporal/ai_observability/eval_reports/report_agent/graph.py | .py | """LangGraph agent for evaluation report generation using create_react_agent."""
import uuid
from typing import Any
import structlog
from langchain_core.messages import HumanMessage
from langchain_core.runnables import RunnableConfig
from langgraph.prebuilt import create_react_agent
from posthog.llm.gateway_client i... | 397 | 15,142 |
posthog | posthog/temporal/ai_observability/eval_reports/report_agent/test/test_graph.py | .py | """Tests for the v2 graph helpers: _fallback_content and _validate_agent_output."""
from unittest.mock import MagicMock, patch
from django.test import SimpleTestCase
from parameterized import parameterized
from posthog.exceptions import ClickHouseAtCapacity
from posthog.temporal.ai_observability.eval_reports.report... | 535 | 22,478 |
posthog | posthog/temporal/ai_observability/eval_reports/report_agent/test/test_tools_ai_events_routing.py | .py | """Tests for the eval-report agent's heavy-prop tools.
`sample_generation_details`, `get_generation_detail`, and
`get_generation_text_repr` are the three tools that read heavy columns. They
route through `_execute_hogql_via_ai_events` which wraps the resolver. The
other query sites in `tools.py` deliberately stay on `... | 355 | 15,329 |
posthog | posthog/temporal/ai_observability/eval_reports/report_agent/test/test_schema.py | .py | """Tests for the v2 eval report schema."""
from django.test import SimpleTestCase
from parameterized import parameterized
from posthog.temporal.ai_observability.eval_reports.output_types import SUPPORTED_EVAL_REPORT_OUTPUT_TYPES
from posthog.temporal.ai_observability.eval_reports.report_agent.schema import (
MAX... | 285 | 11,883 |
posthog | posthog/temporal/ai_observability/eval_reports/report_agent/test/test_tools.py | .py | """Tests for the v2 eval report agent output tools (set_title, add_section, add_citation)."""
import json
import datetime as dt
from typing import NotRequired, TypedDict
from posthog.test.base import BaseTest
from unittest.mock import (
MagicMock,
call as mock_call,
patch,
)
from django.test import Simpl... | 1,152 | 50,880 |
posthog | posthog/temporal/ai_observability/eval_reports/test/test_delivery.py | .py | from unittest.mock import MagicMock, patch
from django.test import SimpleTestCase
from parameterized import parameterized
from posthog.temporal.ai_observability.eval_reports.delivery import (
CitationMap,
_build_citation_map,
_format_period_for_display,
_inline_email_styles,
_linkify_citations,
... | 607 | 27,018 |
posthog | posthog/temporal/ai_observability/eval_reports/test/test_activities.py | .py | import datetime as dt
from contextlib import asynccontextmanager
import pytest
from posthog.test.base import BaseTest, ClickhouseTestMixin, _create_event
from unittest.mock import MagicMock, Mock, patch
from django.test import SimpleTestCase
from django.utils import timezone
from parameterized import parameterized
... | 1,013 | 45,616 |
posthog | posthog/temporal/ai_observability/eval_reports/test/test_emit_signal.py | .py | import random
import pytest
from unittest.mock import AsyncMock, patch
import pytest_asyncio
from asgiref.sync import sync_to_async
from posthog.models import Organization, Team
from posthog.temporal.ai_observability.eval_reports.emit_signal import (
EmitEvalReportSignalInputs,
EvalReportSignalSummary,
_... | 269 | 10,990 |
posthog | posthog/temporal/ai_observability/eval_reports/test/test_workflow.py | .py | import asyncio
from uuid import UUID
import pytest
from unittest.mock import patch
from posthog.temporal.ai_observability.eval_reports.activities import (
deliver_report_activity,
update_next_delivery_date_activity,
)
from posthog.temporal.ai_observability.eval_reports.types import (
CheckCountTriggeredEv... | 449 | 17,810 |
posthog | posthog/temporal/ai_observability/trace_summarization/workflow.py | .py | """
Batch trace summarization workflow for LLMA.
This workflow runs on a schedule (e.g., hourly) to:
1. Query recent traces from a time window
2. Generate text representations and summaries for each trace
3. Store summaries as $ai_trace_summary events in ClickHouse
4. Queue embeddings for summaries via Kafka for async... | 287 | 11,685 |
posthog | posthog/temporal/ai_observability/trace_summarization/state.py | .py | """Redis intermediate storage for passing text_repr between activities.
Stores gzip-compressed text representations in Redis so that only small
references (~100 bytes) flow through Temporal workflow history while the
large payloads (up to 2 MB) go through Redis.
"""
import gzip
import structlog
from redis import asy... | 46 | 1,379 |
posthog | posthog/temporal/ai_observability/trace_summarization/utils.py | .py | """Utility functions for trace summarization."""
from datetime import datetime
def format_datetime_for_clickhouse(iso_string: str) -> str:
"""Convert ISO format datetime string to ClickHouse-compatible format."""
dt = datetime.fromisoformat(iso_string)
return dt.strftime("%Y-%m-%d %H:%M:%S")
| 10 | 308 |
posthog | posthog/temporal/ai_observability/trace_summarization/models.py | .py | """Data models for batch trace summarization."""
from dataclasses import dataclass, field
from typing import Any, Literal
from pydantic import BaseModel
from posthog.temporal.ai_observability.trace_summarization.constants import (
DEFAULT_BATCH_SIZE,
DEFAULT_MAX_ITEMS_PER_WINDOW,
DEFAULT_MODE,
DEFAUL... | 180 | 5,485 |
posthog | posthog/temporal/ai_observability/trace_summarization/sampling.py | .py | """Activity for sampling traces/generations from a time window.
Uses lightweight HogQL queries to sample trace IDs and timestamps.
The full trace data is fetched later by the summarization activity using
TraceQueryRunner per-item, so sampling only needs IDs and timestamps.
"""
from typing import Any
import structlog... | 280 | 12,153 |
posthog | posthog/temporal/ai_observability/trace_summarization/summarize_and_save.py | .py | """Activity for generating LLM summary and saving the result."""
import time
from dataclasses import dataclass
from uuid import uuid4
import structlog
import temporalio
from posthog.models.event.util import create_event
from posthog.models.team import Team
from posthog.redis import get_async_client
from posthog.sync... | 294 | 10,952 |
posthog | posthog/temporal/ai_observability/trace_summarization/__init__.py | .py | """Trace summarization workflows and activities for batch processing."""
from posthog.temporal.ai_observability.trace_summarization.constants import (
COORDINATOR_WORKFLOW_NAME,
DEFAULT_BATCH_SIZE,
DEFAULT_MAX_ITEMS_PER_WINDOW,
DEFAULT_MODE,
DEFAULT_WINDOW_MINUTES,
EVENT_NAME_TRACE_SUMMARY,
... | 46 | 1,812 |
posthog | posthog/temporal/ai_observability/trace_summarization/constants.py | .py | """Constants for batch trace summarization workflows."""
from datetime import timedelta
from temporalio.common import RetryPolicy
from products.ai_observability.backend.summarization.models import OpenAIModel, SummarizationMode
# Window processing configuration
DEFAULT_MAX_ITEMS_PER_WINDOW = (
15 # Max items t... | 156 | 7,701 |
posthog | posthog/temporal/ai_observability/trace_summarization/queries.py | .py | """ClickHouse queries for trace summarization.
Routes through `TraceQueryRunner` so reads land on the dedicated `ai_events`
table (falling back to the shared `events` table for data beyond the retention
window). The runner re-merges the heavy columns into `event.properties`, so
downstream formatters keep working.
"""
... | 38 | 1,521 |
posthog | posthog/temporal/ai_observability/trace_summarization/schedule.py | .py | """Schedule configuration for batch trace summarization."""
from datetime import timedelta
from django.conf import settings
from temporalio.client import (
Client,
Schedule,
ScheduleActionStartWorkflow,
ScheduleIntervalSpec,
ScheduleOverlapPolicy,
SchedulePolicy,
ScheduleSpec,
)
from pos... | 109 | 4,241 |
posthog | posthog/temporal/ai_observability/trace_summarization/coordinator.py | .py | """
Coordinator workflow for batch trace summarization.
This workflow discovers teams dynamically via the team discovery activity
and spawns child workflows to process traces for each team.
Uses continue_as_new after each batch to keep the workflow history bounded
(Temporal has a 50K event limit per execution).
Per-... | 315 | 14,479 |
posthog | posthog/temporal/ai_observability/trace_summarization/metrics.py | .py | """Prometheus metrics for the LLMA trace summarization temporal workflows.
Follows the same pattern as posthog.temporal.ai_observability.metrics (evals).
Metrics are emitted via Temporal's built-in metric meter (activity/workflow context)
and scraped by the Prometheus endpoint on the worker pod.
"""
import typing
imp... | 163 | 5,947 |
posthog | posthog/temporal/ai_observability/trace_summarization/fetch_and_format.py | .py | """Activity for fetching trace/generation data and storing formatted text in Redis."""
import time
import structlog
import temporalio
from posthog.hogql import ast
from posthog.hogql.parser import parse_select
from posthog.clickhouse.query_tagging import Feature, Product, tags_context
from posthog.hogql_queries.ai.... | 320 | 11,400 |
posthog | posthog/temporal/ai_observability/trace_summarization/tests/test_fetch_and_format_heavy_columns.py | .py | """Unit tests for the `_fetch_and_format_generation` query.
Verifies the by-uuid generation lookup routes through `query_ai_events` and reads
the heavy `input` / `output` from `posthog.ai_events` native columns.
"""
from typing import cast
import pytest
from unittest.mock import MagicMock, patch
from posthog.hogql ... | 126 | 5,071 |
posthog | posthog/temporal/ai_observability/trace_summarization/tests/test_metrics.py | .py | from unittest.mock import MagicMock, patch
from parameterized import parameterized
from posthog.temporal.ai_observability.trace_summarization.metrics import (
SUMMARIZATION_ACTIVITY_TYPES,
SUMMARIZATION_WORKFLOW_TYPES,
SummarizationMetricsInterceptor,
increment_embedding_result,
increment_errors,
... | 117 | 4,571 |
posthog | posthog/temporal/ai_observability/trace_summarization/tests/test_coordinator.py | .py | """Tests for batch trace summarization coordinator workflow."""
import pytest
from posthog.temporal.ai_observability.shared_activities import JobConfig, resolve_level_jobs_for_team
from posthog.temporal.ai_observability.trace_summarization.constants import (
DEFAULT_BATCH_SIZE,
DEFAULT_MAX_ITEMS_PER_WINDOW,
... | 168 | 6,237 |
posthog | posthog/temporal/ai_observability/trace_summarization/tests/test_summarize_and_save.py | .py | """Tests for unified summarize_and_save activity."""
import uuid
from contextlib import asynccontextmanager
from datetime import UTC, datetime
import pytest
from unittest.mock import patch
from posthog.temporal.ai_observability.trace_summarization.models import SummarizeAndSaveInput, TextReprExpiredError
from postho... | 203 | 9,050 |
posthog | posthog/temporal/ai_observability/trace_summarization/tests/test_workflow.py | .py | """Tests for batch trace summarization workflow and sampling."""
from contextlib import asynccontextmanager
import pytest
from unittest.mock import patch
from posthog.hogql.context import HogQLContext
from posthog.hogql.placeholders import replace_placeholders
from posthog.hogql.printer.utils import prepare_and_prin... | 390 | 15,751 |
posthog | posthog/temporal/ai_observability/trace_summarization/tests/test_queries.py | .py | """Tests for `fetch_trace` in the trace-summarization Temporal pipeline.
Heavy columns must come back populated on the dedicated `ai_events` path (they live
only there), and the function must hand back a fully-formed LLMTrace, not raise.
"""
import pytest
from unittest.mock import MagicMock, patch
from posthog.schem... | 88 | 3,901 |
posthog | posthog/temporal/ai_observability/trace_summarization/tests/test_fetch_and_format.py | .py | """Tests for unified fetch_and_format activity."""
import uuid
from contextlib import asynccontextmanager
from datetime import UTC, datetime
import pytest
from unittest.mock import MagicMock, patch
from posthog.temporal.ai_observability.trace_summarization.fetch_and_format import (
_fetch_and_format_trace,
_... | 282 | 11,272 |
posthog | posthog/temporal/ai_observability/trace_summarization/tests/test_state.py | .py | """Tests for Redis intermediate storage helpers."""
import gzip
import pytest
from unittest.mock import AsyncMock
from posthog.temporal.ai_observability.trace_summarization.state import (
REDIS_KEY_PREFIX,
REDIS_TTL_SECONDS,
delete_text_repr,
generate_redis_key,
load_text_repr,
store_text_rep... | 92 | 2,814 |
posthog | posthog/temporal/ai_observability/trace_clustering/clustering.py | .py | """Clustering utilities: HDBSCAN and k-means implementations."""
import os
import sys
# Configure Numba threading layer BEFORE importing UMAP (which uses Numba internally).
# The default "workqueue" threading layer is not thread-safe and crashes when multiple
# Python threads call Numba functions concurrently - which... | 367 | 13,120 |
posthog | posthog/temporal/ai_observability/trace_clustering/workflow.py | .py | """Daily trace clustering workflow."""
import json
from datetime import timedelta
import temporalio.exceptions
from temporalio import workflow
from posthog.temporal.ai_observability.trace_clustering.activities import (
emit_cluster_events_activity,
generate_cluster_labels_activity,
perform_clustering_com... | 314 | 13,095 |
posthog | posthog/temporal/ai_observability/trace_clustering/models.py | .py | """Data models for trace clustering workflow."""
from dataclasses import dataclass, field
from typing import Any, TypedDict
from posthog.temporal.ai_observability.shared_activities import AnalysisLevel # noqa: F401
from posthog.temporal.ai_observability.trace_clustering.constants import (
DEFAULT_LOOKBACK_DAYS,
... | 352 | 13,333 |
posthog | posthog/temporal/ai_observability/trace_clustering/__init__.py | .py | """Trace clustering workflow exports."""
from posthog.temporal.ai_observability.trace_clustering.activities import (
emit_cluster_events_activity,
generate_cluster_labels_activity,
perform_clustering_compute_activity,
)
from posthog.temporal.ai_observability.trace_clustering.aggregates import compute_clust... | 22 | 842 |
posthog | posthog/temporal/ai_observability/trace_clustering/constants.py | .py | """Configuration constants for trace clustering workflow."""
from datetime import timedelta
from temporalio.common import RetryPolicy
# Clustering parameters
DEFAULT_LOOKBACK_DAYS = 7
DEFAULT_MAX_SAMPLES = 1500
DEFAULT_MIN_K = 2
DEFAULT_MAX_K = 10
# Minimum items required for clustering, checked per analysis level ... | 136 | 6,716 |
posthog | posthog/temporal/ai_observability/trace_clustering/data.py | .py | """Data access layer for trace clustering.
This module consolidates all HogQL queries used by the clustering workflow,
providing a single source of truth for data fetching operations.
All queries are team-scoped through HogQL's automatic team filtering.
"""
from dataclasses import dataclass
from datetime import datet... | 412 | 17,020 |
posthog | posthog/temporal/ai_observability/trace_clustering/schedule.py | .py | """Schedule configuration for daily trace clustering coordinator."""
from datetime import timedelta
from django.conf import settings
from temporalio.client import (
Client,
Schedule,
ScheduleActionStartWorkflow,
ScheduleIntervalSpec,
ScheduleOverlapPolicy,
SchedulePolicy,
ScheduleSpec,
)
... | 121 | 4,425 |
posthog | posthog/temporal/ai_observability/trace_clustering/coordinator.py | .py | """
Coordinator workflow for daily trace clustering.
This workflow discovers teams dynamically via the team discovery activity
and spawns child workflows to cluster traces for each team.
Uses continue_as_new after each batch to keep the workflow history bounded
(Temporal has a 50K event limit per execution).
Per-tea... | 305 | 14,146 |
posthog | posthog/temporal/ai_observability/trace_clustering/metrics.py | .py | """Prometheus metrics for the LLMA trace clustering temporal workflows.
Follows the same pattern as trace_summarization/metrics.py.
Metrics are emitted via Temporal's built-in metric meter (activity/workflow context)
and scraped by the Prometheus endpoint on the worker pod.
"""
import typing
import datetime as dt
fr... | 210 | 7,888 |
posthog | posthog/temporal/ai_observability/trace_clustering/activities.py | .py | """Temporal activities for trace clustering workflow.
This module contains the 3 activities that make up the clustering pipeline:
1. perform_clustering_compute_activity - fetch embeddings, cluster, compute distances
2. generate_cluster_labels_activity - LLM labeling for clusters
3. emit_cluster_events_activity - emit ... | 411 | 16,265 |
posthog | posthog/temporal/ai_observability/trace_clustering/aggregates.py | .py | """Compute aggregate metrics (cost, latency, tokens, errors) for clusters.
This activity is best-effort: it returns whatever metrics it managed to compute.
"""
import time
import asyncio
from django.utils.dateparse import parse_datetime
import structlog
from temporalio import activity
from posthog.models.team impo... | 129 | 4,477 |
posthog | posthog/temporal/ai_observability/trace_clustering/event_emission.py | .py | """Event emission for clustering results.
This module contains functions for emitting clustering results to ClickHouse:
- Build cluster data structures with centroids and distances
- Emit $ai_trace_clusters event with all clustering metadata
- Handle noise/outlier cluster from HDBSCAN (cluster_id = -1)
"""
import uui... | 327 | 12,766 |
posthog | posthog/temporal/ai_observability/trace_clustering/labeling.py | .py | """LLM-based cluster labeling for trace clustering workflow.
This module provides cluster labeling using a LangGraph agent that can
iteratively explore cluster structure and generate high-quality labels.
"""
from datetime import datetime
import numpy as np
from posthog.models.team import Team
from posthog.temporal.... | 169 | 6,660 |
posthog | posthog/temporal/ai_observability/trace_clustering/tests/test_event_emission.py | .py | """Tests for event emission in trace clustering."""
import pytest
from unittest.mock import patch
import numpy as np
from parameterized import parameterized
from posthog.temporal.ai_observability.trace_clustering.constants import NOISE_CLUSTER_ID
from posthog.temporal.ai_observability.trace_clustering.event_emission... | 378 | 14,645 |
posthog | posthog/temporal/ai_observability/trace_clustering/tests/test_metrics.py | .py | from unittest.mock import MagicMock, patch
from parameterized import parameterized
from posthog.temporal.ai_observability.trace_clustering.metrics import (
CLUSTERING_ACTIVITY_TYPES,
CLUSTERING_WORKFLOW_TYPES,
ClusteringMetricsInterceptor,
increment_errors,
increment_workflow_finished,
increme... | 137 | 5,723 |
posthog | posthog/temporal/ai_observability/trace_clustering/tests/test_coordinator.py | .py | """Tests for trace clustering coordinator workflow."""
import pytest
from posthog.temporal.ai_observability.shared_activities import JobConfig, resolve_level_jobs_for_team
from posthog.temporal.ai_observability.trace_clustering import constants
from posthog.temporal.ai_observability.trace_clustering.coordinator impor... | 142 | 5,199 |
posthog | posthog/temporal/ai_observability/trace_clustering/tests/test_workflow.py | .py | """Tests for trace clustering workflow."""
import uuid
from datetime import UTC, datetime
import pytest
from unittest.mock import patch
import numpy as np
from temporalio import activity
from temporalio.testing import WorkflowEnvironment
from temporalio.worker import UnsandboxedWorkflowRunner, Worker
from posthog.t... | 464 | 18,369 |
posthog | posthog/temporal/ai_observability/trace_clustering/tests/test_clustering.py | .py | """Tests for clustering utilities: HDBSCAN, k-means, and dimensionality reduction."""
import pytest
import numpy as np
from parameterized import parameterized
# Check if UMAP threading backend is available (required for UMAP tests)
def _umap_threading_available() -> bool:
"""Check if UMAP can run without thread... | 244 | 9,148 |
posthog | posthog/temporal/ai_observability/trace_clustering/tests/test_labeling_agent.py | .py | """Tests for the cluster labeling agent."""
import json
from unittest.mock import ANY, MagicMock, patch
from langchain_core.messages import HumanMessage
from posthog.temporal.ai_observability.trace_clustering.labeling_agent.graph import (
_apply_fallbacks as _fill_missing_labels,
run_labeling_agent,
)
from ... | 473 | 16,414 |
posthog | posthog/temporal/ai_observability/trace_clustering/tests/test_data.py | .py | """Tests for data access layer in trace clustering."""
from datetime import UTC, datetime
import pytest
from unittest.mock import MagicMock, patch
from posthog.temporal.ai_observability.trace_clustering.data import (
fetch_item_embeddings_for_clustering,
fetch_item_summaries,
)
@pytest.fixture
def mock_tea... | 333 | 13,351 |
posthog | posthog/temporal/ai_observability/trace_clustering/labeling_agent/state.py | .py | """State definitions for the cluster labeling agent."""
from typing import Annotated, TypedDict
from langgraph.graph.message import add_messages
from langgraph.managed import RemainingSteps
from posthog.temporal.ai_observability.trace_clustering.models import ClusterLabel, TraceSummary
class TraceMetadata(TypedDic... | 97 | 2,482 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.