Skip to content

: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.

fix(data-warehouse): measure the most recomputed views first across teams (#130137)

Thiago Salvatore63e466c4cd33
diff --git a/products/warehouse_suggestions/backend/logic/candidates/base.py b/products/warehouse_suggestions/backend/logic/candidates/base.py
index cbd4580f90f3..274b674f87cd 100644
--- a/products/warehouse_suggestions/backend/logic/candidates/base.py
+++ b/products/warehouse_suggestions/backend/logic/candidates/base.py
@@ -150,12 +150,18 @@ class Rejection:
reason: str
+@frozen
+class MeasurementRequest:
+ subject: Subject
+ consumer_runs_per_month: float
+
+
@frozen
class CandidateResult:
drafts: tuple[SuggestionDraft, ...]
rejections: tuple[Rejection, ...]
skipped_reason: str | None = None
- measurement_requests: tuple[Subject, ...] = ()
+ measurement_requests: tuple[MeasurementRequest, ...] = ()
class Candidate(ABC):
diff --git a/products/warehouse_suggestions/backend/logic/candidates/materialize.py b/products/warehouse_suggestions/backend/logic/candidates/materialize.py
index 88cee32d3586..b744dbc36d3a 100644
--- a/products/warehouse_suggestions/backend/logic/candidates/materialize.py
+++ b/products/warehouse_suggestions/backend/logic/candidates/materialize.py
@@ -41,6 +41,7 @@ from .base import (
Candidate,
CandidateContext,
CandidateResult,
+ MeasurementRequest,
QueryCost,
RealizedSavings,
Rejection,
@@ -565,9 +566,12 @@ def _read_seconds(reads: SubjectReads | None) -> float | None:
return reads.alone_duration_ms_median / MILLISECONDS_PER_SECOND
-def _most_recomputed(unmeasured: list[tuple[Subject, Recomputations]], limit: int) -> tuple[Subject, ...]:
+def _most_recomputed(unmeasured: list[tuple[Subject, Recomputations]], limit: int) -> tuple[MeasurementRequest, ...]:
ranked = sorted(unmeasured, key=lambda pair: pair[1].consumer_runs_per_month, reverse=True)
- return tuple(subject for subject, _ in ranked[:limit])
+ return tuple(
+ MeasurementRequest(subject=subject, consumer_runs_per_month=recomputations.consumer_runs_per_month)
+ for subject, recomputations in ranked[:limit]
+ )
def _read_cost_before_at(score_inputs: Mapping[str, float], reads_per_month: float) -> QueryCost | None:
diff --git a/products/warehouse_suggestions/backend/logic/job.py b/products/warehouse_suggestions/backend/logic/job.py
index 4df998cd425f..53b064bbb549 100644
--- a/products/warehouse_suggestions/backend/logic/job.py
+++ b/products/warehouse_suggestions/backend/logic/job.py
@@ -13,11 +13,11 @@ from posthog.models.team.extensions import get_or_create_team_extension
from ..facade.enums import WarehouseSuggestionKind, WarehouseSuggestionStatus
from ..models import WarehouseSuggestion, WarehouseSuggestionTeamConfig
from .analytics import report_candidates
-from .candidates.base import CandidateContext, CandidateResult
+from .candidates.base import CandidateContext, CandidateResult, MeasurementRequest
from .candidates.registry import evaluate_candidates
from .inventory import load_inventory
from .lifecycle import FLIP_KINDS, LifecycleResult, apply_run
-from .reads import ReadWindow, RollupDays, Subject, TeamReads, read_rollup_days, read_team_reads
+from .reads import ReadWindow, RollupDays, TeamReads, read_rollup_days, read_team_reads
from .rules import RULES, EligibilityRules, Rules
MAINTAINED_STATUSES = (WarehouseSuggestionStatus.PROPOSED, WarehouseSuggestionStatus.ACCEPTED)
@@ -35,7 +35,7 @@ class TeamRunResult:
status: TeamRunStatus
drafts: int = 0
lifecycle: LifecycleResult | None = None
- measurement_requests: tuple[Subject, ...] = ()
+ measurement_requests: tuple[MeasurementRequest, ...] = ()
def run_team(
@@ -70,7 +70,7 @@ def run_team(
status=status,
drafts=len(drafts),
lifecycle=lifecycle,
- measurement_requests=tuple(subject for result in results.values() for subject in result.measurement_requests),
+ measurement_requests=tuple(request for result in results.values() for request in result.measurement_requests),
)
diff --git a/products/warehouse_suggestions/backend/temporal/activities.py b/products/warehouse_suggestions/backend/temporal/activities.py
index a02900f5ec96..41eec25f6e30 100644
--- a/products/warehouse_suggestions/backend/temporal/activities.py
+++ b/products/warehouse_suggestions/backend/temporal/activities.py
@@ -20,7 +20,7 @@ from ..logic.job import TeamRunStatus, run_team, team_ids_with_suggestions_to_ma
from ..logic.reads import ReadWindow, read_rollup_days
from ..logic.rules import RULES
from ..logic.view_costs import MeasurementOutcome, measure_view_cost
-from .contracts import BatchOutcome, ViewCostRequest, WarehouseSuggestionsInputs
+from .contracts import BatchOutcome, ViewCostRequest, WarehouseSuggestionsInputs, most_recomputed
LOGGER = get_logger(__name__)
FULL_ROLLOUT = 1.0
@@ -63,14 +63,19 @@ def run_batch(team_ids: list[int], run_id: str) -> BatchOutcome:
continue
outcomes[result.status] += 1
view_costs_to_measure.extend(
- ViewCostRequest(team_id=team_id, saved_query_id=str(subject.id)) for subject in result.measurement_requests
+ ViewCostRequest(
+ team_id=team_id,
+ saved_query_id=str(request.subject.id),
+ consumer_runs_per_month=request.consumer_runs_per_month,
+ )
+ for request in result.measurement_requests
)
return BatchOutcome(
processed=outcomes[TeamRunStatus.PROCESSED],
not_eligible=outcomes[TeamRunStatus.NOT_ELIGIBLE],
disabled=outcomes[TeamRunStatus.DISABLED],
failed=outcomes["failed"],
- view_costs_to_measure=view_costs_to_measure[: RULES.measurement.max_per_run],
+ view_costs_to_measure=most_recomputed(view_costs_to_measure, RULES.measurement.max_per_run),
)
diff --git a/products/warehouse_suggestions/backend/temporal/contracts.py b/products/warehouse_suggestions/backend/temporal/contracts.py
index 4c4d36924e0a..bb9049a883ec 100644
--- a/products/warehouse_suggestions/backend/temporal/contracts.py
+++ b/products/warehouse_suggestions/backend/temporal/contracts.py
@@ -15,6 +15,11 @@ class WarehouseSuggestionsInputs:
class ViewCostRequest:
team_id: int
saved_query_id: str
+ consumer_runs_per_month: float = 0.0
+
+
+def most_recomputed(requests: list[ViewCostRequest], limit: int) -> list[ViewCostRequest]:
+ return sorted(requests, key=lambda request: request.consumer_runs_per_month, reverse=True)[:limit]
@dataclass(frozen=True, kw_only=True)
diff --git a/products/warehouse_suggestions/backend/temporal/workflows.py b/products/warehouse_suggestions/backend/temporal/workflows.py
index 8a8ce6f14b72..0fb0dfe342da 100644
--- a/products/warehouse_suggestions/backend/temporal/workflows.py
+++ b/products/warehouse_suggestions/backend/temporal/workflows.py
@@ -9,7 +9,7 @@ from posthog.temporal.common.base import PostHogWorkflow
from ..logic.rules import RULES
from .activities import generate_warehouse_suggestions, get_warehouse_suggestion_team_batches, measure_view_costs
-from .contracts import WAREHOUSE_SUGGESTIONS_WORKFLOW_NAME, BatchOutcome, WarehouseSuggestionsInputs
+from .contracts import WAREHOUSE_SUGGESTIONS_WORKFLOW_NAME, BatchOutcome, WarehouseSuggestionsInputs, most_recomputed
MEASURE_VIEW_COSTS_PATCH = "measure-view-costs"
MEASUREMENT_MARGIN = dt.timedelta(minutes=10)
@@ -62,7 +62,7 @@ class WarehouseSuggestionsWorkflow(PostHogWorkflow):
}
)
total = dataclasses.replace(
- total, view_costs_to_measure=total.view_costs_to_measure[: RULES.measurement.max_per_run]
+ total, view_costs_to_measure=most_recomputed(total.view_costs_to_measure, RULES.measurement.max_per_run)
)
if total.view_costs_to_measure and workflow.patched(MEASURE_VIEW_COSTS_PATCH):
await workflow.execute_activity(
diff --git a/products/warehouse_suggestions/backend/tests/test_candidates.py b/products/warehouse_suggestions/backend/tests/test_candidates.py
index 6b660438725e..ef15e0f5ae29 100644
--- a/products/warehouse_suggestions/backend/tests/test_candidates.py
+++ b/products/warehouse_suggestions/backend/tests/test_candidates.py
@@ -773,7 +773,9 @@ class TestMaterializeRecomputedUpstream(SimpleTestCase):
):
result = MaterializeCandidate().evaluate(candidate_context)
- assert result.measurement_requests == ((view_subject(VIEW_ID),) if expect_measurement_request else ())
+ assert tuple(request.subject for request in result.measurement_requests) == (
+ (view_subject(VIEW_ID),) if expect_measurement_request else ()
+ )
if not expect_draft:
assert result.drafts == ()
return
diff --git a/products/warehouse_suggestions/backend/tests/test_temporal.py b/products/warehouse_suggestions/backend/tests/test_temporal.py
index 48d11e7e3316..43212943d356 100644
--- a/products/warehouse_suggestions/backend/tests/test_temporal.py
+++ b/products/warehouse_suggestions/backend/tests/test_temporal.py
@@ -1,4 +1,5 @@
import uuid
+import dataclasses
from posthog.test.base import BaseTest
from unittest.mock import patch
@@ -12,6 +13,7 @@ from posthog.models.team import Team
from products.warehouse_suggestions.backend.facade.contracts import CertifyPayload
from products.warehouse_suggestions.backend.facade.enums import WarehouseSuggestionKind, WarehouseSuggestionStatus
+from products.warehouse_suggestions.backend.logic.candidates.base import MeasurementRequest
from products.warehouse_suggestions.backend.logic.candidates.registry import CANDIDATES
from products.warehouse_suggestions.backend.logic.job import TeamRunResult, TeamRunStatus
from products.warehouse_suggestions.backend.logic.reads import RollupDays
@@ -32,6 +34,7 @@ from products.warehouse_suggestions.backend.temporal.contracts import (
from products.warehouse_suggestions.backend.tests.factories import context, team_reads, view_subject
ACTIVITIES = "products.warehouse_suggestions.backend.temporal.activities"
+WORKFLOWS_MODULE = "products.warehouse_suggestions.backend.temporal.workflows"
class TestTeamBatches(BaseTest):
@@ -87,12 +90,22 @@ class TestTeamBatches(BaseTest):
assert batches == [[self.team.pk]]
-async def test_a_worker_runs_a_nonempty_batch_and_measures_the_views_a_team_asked_for() -> None:
+async def test_a_worker_measures_the_most_recomputed_views_across_teams_first() -> None:
task_queue = str(uuid.uuid4())
- unmeasured_view = uuid.uuid4()
- measurement_requests = {1: (view_subject(unmeasured_view),), 2: ()}
+ rarely_recomputed_view = uuid.uuid4()
+ often_recomputed_view = uuid.uuid4()
+ measurement_requests = {
+ 1: (MeasurementRequest(subject=view_subject(rarely_recomputed_view), consumer_runs_per_month=10),),
+ 2: (MeasurementRequest(subject=view_subject(often_recomputed_view), consumer_runs_per_month=100),),
+ 3: (),
+ }
+ one_measurement_per_run = dataclasses.replace(
+ RULES, measurement=dataclasses.replace(RULES.measurement, max_per_run=1)
+ )
with (
- patch(f"{ACTIVITIES}.team_batches", return_value=[[1, 2]]),
+ patch(f"{ACTIVITIES}.team_batches", return_value=[[1, 2, 3]]),
+ patch(f"{ACTIVITIES}.RULES", one_measurement_per_run),
+ patch(f"{WORKFLOWS_MODULE}.RULES", one_measurement_per_run),
patch(f"{ACTIVITIES}.read_rollup_days", return_value=RollupDays(days_with_data=30, recent_days_with_data=7)),
patch(
f"{ACTIVITIES}.run_team",
@@ -120,5 +133,5 @@ async def test_a_worker_runs_a_nonempty_batch_and_measures_the_views_a_team_aske
result_type=BatchOutcome,
)
- assert outcome.processed == 2
- measure_view_cost.assert_called_once_with(1, unmeasured_view, RULES.measurement)
+ assert outcome.processed == 3
+ measure_view_cost.assert_called_once_with(2, often_recomputed_view, one_measurement_per_run.measurement)