: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
from datetime import timedelta
from typing import Any
from django.db.models import Exists, F, OuterRef, Q
from django.utils import timezone
from celery import shared_task
from structlog import get_logger
from posthog.celery_queues import CeleryQueue
from posthog.scoping_audit import skip_team_scope_audit
from products.data_modeling.backend.facade.api import HasDependentsError
from products.data_modeling.backend.facade.models import DataModelingJob, DataModelingJobStatus
from products.endpoints.backend.logic.ducklake_shadow import run_ducklake_shadow_comparison
from products.endpoints.backend.logic.materialization_dependencies import EndpointMaterializationDependencyService
from products.endpoints.backend.metrics import ENDPOINT_MATERIALIZATION_EVENT_TOTAL
from products.endpoints.backend.models import EndpointVersion
logger = get_logger(__name__)
STALE_THRESHOLD_DAYS = 30
@shared_task(
ignore_result=True,
queue=CeleryQueue.LONG_RUNNING.value,
name="products.endpoints.backend.tasks.shadow_compare_ducklake_execution",
)
@skip_team_scope_audit
def shadow_compare_ducklake_execution(
team_id: int,
endpoint_id: str,
version_id: str,
variables: dict[str, Any] | None,
execution_type: str,
clickhouse_cached: bool,
clickhouse_ms: float,
clickhouse_row_count: int | None,
limit: int | None,
offset: int | None,
) -> None:
run_ducklake_shadow_comparison(
team_id=team_id,
endpoint_id=endpoint_id,
version_id=version_id,
variables=variables,
execution_type=execution_type,
clickhouse_cached=clickhouse_cached,
clickhouse_ms=clickhouse_ms,
clickhouse_row_count=clickhouse_row_count,
limit=limit,
offset=offset,
)
@shared_task(ignore_result=True, name="products.endpoints.backend.tasks.deactivate_stale_materializations")
@skip_team_scope_audit
def deactivate_stale_materializations() -> None:
"""
Deactivate materializations for endpoint versions that haven't been executed in over 30 days.
This task finds endpoint versions where:
1. The version has an active materialization (saved_query.is_materialized = True)
2. The materialization has run in the past 24h (a finished job, or saved_query.last_run_at)
3. The materialization was enabled at least 30 days ago (saved_query.created_at)
4. The version was last executed over 30 days ago (via API key), or it is a superseded
version that was never executed
5. No other view, endpoint or metric reads the saved query
For matching versions, the materialization is reverted to save resources.
"""
now = timezone.now()
twenty_four_hours_ago = now - timedelta(hours=24)
stale_threshold = now - timedelta(days=STALE_THRESHOLD_DAYS)
# A still-running job does not count: reverting soft-deletes the saved query under a live workflow.
recent_job = DataModelingJob.objects.filter(
saved_query_id=OuterRef("saved_query_id"),
last_run_at__gte=twenty_four_hours_ago,
).exclude(status=DataModelingJobStatus.RUNNING)
ran_recently = Q(saved_query__last_run_at__gte=twenty_four_hours_ago) | Q(Exists(recent_job))
# Both stamps are written together on every API-key call, so a superseded version with no stamp
# of its own has not been called since the stamp existed. The endpoint stamp belongs to the
# version that replaced it.
superseded = ~Q(version=F("endpoint__current_version"))
version_stale = (
Q(last_executed_at__lt=stale_threshold)
| Q(last_executed_at__isnull=True, endpoint__last_executed_at__lt=stale_threshold)
| (Q(last_executed_at__isnull=True) & superseded)
)
stale_versions = EndpointVersion.objects.filter(
ran_recently,
version_stale,
saved_query__isnull=False,
saved_query__is_materialized=True,
saved_query__deleted=False,
saved_query__created_at__lte=stale_threshold,
endpoint__deleted=False,
).select_related("saved_query", "endpoint")
if not stale_versions.exists():
logger.info("deactivate_stale_materializations_no_candidates")
return
deactivated_count = 0
skipped_count = 0
dependency_services: dict[int, EndpointMaterializationDependencyService] = {}
for version in stale_versions:
team_id = version.endpoint.team_id
log_context = {
"endpoint_id": str(version.endpoint.id),
"endpoint_name": version.endpoint.name,
"version": version.version,
"team_id": team_id,
"saved_query_name": version.saved_query.name if version.saved_query else None,
}
dependency_service = dependency_services.setdefault(team_id, EndpointMaterializationDependencyService(team_id))
try:
dependency_service.check_deletable(version)
except HasDependentsError as error:
logger.info(
"deactivate_stale_materialization_skipped_dependents",
dependents=len(error.dependents),
**log_context,
)
skipped_count += 1
ENDPOINT_MATERIALIZATION_EVENT_TOTAL.labels(action="deactivate_stale", status="skipped_dependents").inc()
continue
except Exception:
logger.exception("deactivate_stale_materialization_dependents_check_failed", **log_context)
skipped_count += 1
ENDPOINT_MATERIALIZATION_EVENT_TOTAL.labels(action="deactivate_stale", status="skipped_dependents").inc()
continue
try:
_deactivate_version_materialization(version)
except Exception as e:
logger.exception(
"deactivate_stale_materialization_failed",
error=str(e),
**log_context,
)
ENDPOINT_MATERIALIZATION_EVENT_TOTAL.labels(action="deactivate_stale", status="error").inc()
continue
deactivated_count += 1
ENDPOINT_MATERIALIZATION_EVENT_TOTAL.labels(action="deactivate_stale", status="success").inc()
logger.info(
"deactivate_stale_materializations_completed",
deactivated_count=deactivated_count,
skipped_count=skipped_count,
)
def _deactivate_version_materialization(version: EndpointVersion) -> None:
"""
Deactivate materialization for an endpoint version.
This reverts the materialization (removes Temporal schedule, cleans up S3 tables)
and soft-deletes the saved_query.
"""
saved_query = version.saved_query
if not saved_query:
return
logger.info(
"deactivating_stale_materialization",
endpoint_id=str(version.endpoint.id),
endpoint_name=version.endpoint.name,
version=version.version,
team_id=version.endpoint.team_id,
last_executed_at=str(version.endpoint.last_executed_at) if version.endpoint.last_executed_at else None,
last_run_at=str(saved_query.last_run_at) if saved_query.last_run_at else None,
)
version.disable_materialization()