test(perf): record the row count and size of each table the Admin UI journeys query (#47683)

Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com>

tests/e2e/perf/setup/ui_query_fixture.py

@@ -3,7 +3,9 @@ from __future__ import annotations
 import hashlib
 import os
 import time
+from collections.abc import Iterable
 from datetime import datetime, timedelta
+from itertools import chain
 from typing import Final
 
 import psycopg
@@ -18,6 +20,33 @@ class UiFixture(BaseModel):
 
     anchor: AwareDatetime
     api_key: str
+    """`perf_ui_db_*` JUnit properties naming how much data the journeys ran against."""
+    volume: tuple[tuple[str, int], ...] = ()
+
+
+class TableVolume(BaseModel):
+    model_config = ConfigDict(frozen=True)
+
+    table: str
+    rows: int
+    bytes: int
+
+
+QUERIED_TABLES: Final = (
+    "LiteLLM_SpendLogs",
+    "LiteLLM_DailyUserSpend",
+    "LiteLLM_DailyTeamSpend",
+    "LiteLLM_DailyTagSpend",
+    "LiteLLM_VerificationToken",
+    "LiteLLM_TeamTable",
+    "LiteLLM_UserTable",
+    "LiteLLM_ProxyModelTable",
+)
+_VOLUME_SQL: Final = """
+    SELECT c.relname, GREATEST(c.reltuples, 0)::bigint, pg_total_relation_size(c.oid)
+    FROM pg_class c JOIN pg_namespace n ON n.oid = c.relnamespace
+    WHERE n.nspname = 'public' AND c.relname = ANY(%(tables)s)
+"""
 
 
 SEEDED_TEAMS: Final = 100
@@ -59,6 +88,30 @@ _SEED_SQL: Final = (
 )
 
 
+def volume_properties(volumes: Iterable[TableVolume]) -> tuple[tuple[str, int], ...]:
+    """One `perf_ui_db_rows_<table>` and one `perf_ui_db_bytes_<table>` property per table the journeys read,
+    so the dashboard can say what the latencies were measured against."""
+    return tuple(
+        chain.from_iterable(
+            ((f"perf_ui_db_rows_{v.table}", v.rows), (f"perf_ui_db_bytes_{v.table}", v.bytes)) for v in volumes
+        )
+    )
+
+
+def table_volumes(connection: psycopg.Connection, spend_log_rows: int) -> tuple[TableVolume, ...]:
+    """Planner row estimates and on-disk size (indexes included) of the queried tables. The restore's
+    `LiteLLM_SpendLogs` count comes from the seed checkpoint, exact where the estimate can be stale."""
+    adapter: Final = TypeAdapter(tuple[str, int, int])
+    found: Final = tuple(
+        adapter.validate_python(row)
+        for row in connection.execute(_VOLUME_SQL, {"tables": list(QUERIED_TABLES)}).fetchall()
+    )
+    return tuple(
+        TableVolume(table=table, rows=spend_log_rows if table == "LiteLLM_SpendLogs" else rows, bytes=size)
+        for table, rows, size in sorted(found)
+    )
+
+
 def prepare_fixture() -> UiFixture:
     validate_target(os.environ["DATABASE_URL"])
     _finish_upgrade()
@@ -71,6 +124,7 @@ def prepare_fixture() -> UiFixture:
             anchor=TypeAdapter(datetime).validate_python(row[0]),
             api_key=hashlib.md5(b"perf-key:0", usedforsecurity=False).hexdigest()
             + hashlib.md5(b"perf-key-tail:0", usedforsecurity=False).hexdigest(),
+            volume=volume_properties(table_volumes(connection, TypeAdapter(int).validate_python(row[1]))),
         )
         return fixture
 

tests/e2e/perf/test_ui_journeys.py

@@ -11,14 +11,14 @@ from types import FrameType
 from typing import Final
 
 import pytest
-from playwright.sync_api import Page
 from e2e_config import MASTER_KEY, UI_BASE_URL, UI_PASSWORD, UI_USERNAME
 from e2e_http import send_text_external, without_retries
+from playwright.sync_api import Page
 from setup.ui_journeys import (
-    assert_within_ceiling,
     Dashboard,
     Journey,
     Sample,
+    assert_within_ceiling,
     dashboard_page,
     database_failures,
     error_code,
@@ -111,6 +111,7 @@ def test_ui_journey_latency(journey_name: str, ui_fixture: UiFixture, request: p
             item: Final = request.node  # pyright: ignore[reportUnknownMemberType, reportUnknownVariableType]  # pytest: untyped
             assert isinstance(item, pytest.Item)
             item.user_properties.extend(ui_junit_properties(samples))
+            item.user_properties.extend(ui_fixture.volume)
         after: Final = settled_counter(lambda: _metrics(transport), evidence)
         item.user_properties.extend(
             (

tests/e2e_harness/perf/setup/test_ui_journeys.py

@@ -6,9 +6,9 @@ import pytest
 from setup.ui_journeys import (
     UI_JOURNEY_LIMITS,
     Sample,
-    error_code,
     assert_within_ceiling,
     database_failures,
+    error_code,
     journeys,
     summarize,
     ui_junit_properties,
@@ -153,3 +153,21 @@ def test_failed_data_request_status_becomes_the_error_code() -> None:
     metrics: Final = dict(ui_junit_properties((gateway_error, slow, hung)))
     assert metrics["perf_ui_error_http_500"] == 1 and metrics["perf_ui_error_wall_clock"] == 1
     assert "perf_ui_error_ui" not in metrics
+
+
+def test_volume_properties_name_rows_and_bytes_per_queried_table() -> None:
+    from setup.ui_query_fixture import TableVolume, volume_properties
+
+    properties: Final = volume_properties(
+        (
+            TableVolume(table="LiteLLM_SpendLogs", rows=850_000_000, bytes=1_000_000_000_000),
+            TableVolume(table="LiteLLM_VerificationToken", rows=5_000, bytes=4_194_304),
+        )
+    )
+
+    assert properties == (
+        ("perf_ui_db_rows_LiteLLM_SpendLogs", 850_000_000),
+        ("perf_ui_db_bytes_LiteLLM_SpendLogs", 1_000_000_000_000),
+        ("perf_ui_db_rows_LiteLLM_VerificationToken", 5_000),
+        ("perf_ui_db_bytes_LiteLLM_VerificationToken", 4_194_304),
+    )