feat(generator): добавлен артефакт стартовой истории

- Зачем:
  - чистый стенд должен восстанавливать стартовую историю без повторной генерации.
- Что:
  - добавлены export/import артефакта через Kafka и compact-топики.
  - добавлена manifest-aware сверка ClickHouse и защита от смешения state.
  - добавлен runbook использования стартовой истории.
- Проверка:
  - uv run --with-requirements generator/requirements.txt pytest generator/tests -q.
  - bash -n scripts/check_startup_history_manifest.sh scripts/export_startup_history_artifact.sh scripts/import_startup_history_artifact.sh.
  - git diff --check.
This commit is contained in:
2026-07-04 18:23:54 +03:00
parent ee4c9f5857
commit 3c38465d89
16 changed files with 1746 additions and 177 deletions
+142 -13
View File
@@ -369,6 +369,46 @@ class TestGeneratorServiceBackfill:
assert first["digest"] == second["digest"]
assert first["manifest_digest"] == second["manifest_digest"]
def test_backfill_writes_portable_artifact_when_requested(
self, base_config, tmp_path
):
"""Backfill пишет переносимый файл с событиями, state и manifest."""
from clickstream_generator.startup_history_artifact import (
load_startup_history_artifact,
validate_startup_history_artifact,
)
model_t0 = datetime(2026, 1, 1, 10, 0, tzinfo=timezone.utc)
model_t_end = model_t0 + timedelta(minutes=1)
artifact_path = tmp_path / "startup-history.json"
config = replace(
base_config,
run_mode="backfill",
model_t0=model_t0,
model_t_end=model_t_end,
tick_seconds=60,
lambda_base_per_min=600,
jitter_pct=0,
min_events_per_tick=1,
max_events_per_tick=1000,
max_session_events=5,
max_active_sessions=250,
population_max=251,
state_enabled=True,
startup_history_artifact=artifact_path,
)
self._run_backfill(config)
artifact = load_startup_history_artifact(artifact_path)
state, manifest, topics, _ = validate_startup_history_artifact(
artifact,
expected_config=config,
)
assert topics["browser_events"]
assert state.last_batch_id == manifest["state"]["last_batch_id"]
assert manifest["totals"]["events"] == len(topics["browser_events"])
def test_live_start_uses_startup_manifest_without_wall_delta(
self, base_config, event_dictionary
):
@@ -403,6 +443,7 @@ class TestGeneratorServiceBackfill:
"model_t_end": model_t_end.isoformat(),
"model_timezone": config.model_timezone,
"generation_settings": GeneratorService(config)._generation_settings(),
"state_version": state.version,
"state": {"last_batch_id": state.last_batch_id},
}
state_manager = MagicMock()
@@ -468,6 +509,7 @@ class TestGeneratorServiceBackfill:
"model_t_end": manifest_t_end.isoformat(),
"model_timezone": config.model_timezone,
"generation_settings": GeneratorService(config)._generation_settings(),
"state_version": state.version,
"state": {"last_batch_id": state.last_batch_id},
}
state_manager = MagicMock()
@@ -506,12 +548,61 @@ class TestGeneratorServiceBackfill:
patch.object(GeneratorService, "_main_loop", return_value=None):
service = GeneratorService(config)
service.start()
with pytest.raises(ValueError, match="GEN_MODEL_T_END.*GEN_STATE_RESET=true"):
service.start()
restore_from_startup_history.assert_not_called()
restore_live_state.assert_not_called()
assert service._tick == 0
assert service._model_time == config.model_t0
def test_live_start_fails_loudly_on_readable_incompatible_state(
self, base_config, event_dictionary
):
"""Читаемый state другого мира при продолжении даёт жёсткий отказ."""
model_t0 = datetime(2026, 1, 1, 10, 0, tzinfo=timezone.utc)
config = replace(
base_config,
model_t0=model_t0,
tick_seconds=60,
model_time_speed=10,
state_reset=False,
)
source_generator = EventGenerator(event_dictionary, config)
source_stream = TickStreamGenerator(source_generator)
source_stream.generate_tick(event_budget=10, tick_started_at=model_t0)
state = source_stream.to_state(
tick=5,
rng_state=source_generator.rng.getstate(),
last_batch_id="live-state",
last_timestamp=model_t0,
model_timestamp=model_t0,
wall_timestamp=datetime(2026, 1, 1, 0, 0, tzinfo=timezone.utc),
model_time_speed=config.model_time_speed,
model_timezone=config.model_timezone,
model_t0=config.model_t0,
gen_seed=config.seed + 1,
)
state_manager = MagicMock()
state_manager.load.return_value = state
manifest_manager = MagicMock()
manifest_manager.load.return_value = None
with patch("clickstream_generator.service.start_http_server"), \
patch("clickstream_generator.service.ensure_topics"), \
patch("clickstream_generator.service.KafkaPublisher"), \
patch("clickstream_generator.service.KafkaBatchHistory"), \
patch(
"clickstream_generator.service.KafkaStateManager",
return_value=state_manager,
), \
patch(
"clickstream_generator.service.KafkaStartupHistoryManifest",
return_value=manifest_manager,
), \
patch.object(GeneratorService, "_main_loop", return_value=None):
service = GeneratorService(config)
with pytest.raises(ValueError, match="GEN_SEED.*GEN_STATE_RESET=true"):
service.start()
def test_live_start_rejects_orphan_startup_state_without_live_restore(
self, base_config, event_dictionary, caplog
@@ -564,13 +655,14 @@ class TestGeneratorServiceBackfill:
caplog.at_level(logging.WARNING, logger="generator"):
service = GeneratorService(config)
service.start()
with pytest.raises(
ValueError,
match="generator_startup_history_manifest.*GEN_STATE_RESET=true",
):
service.start()
restore_from_startup_history.assert_not_called()
restore_live_state.assert_not_called()
assert service._tick == 0
assert service._model_time == model_t0
assert "startup-history state without matching manifest" in caplog.text
def test_startup_history_state_checks_state_fields(
self, base_config, event_dictionary
@@ -601,6 +693,44 @@ class TestGeneratorServiceBackfill:
"model_t_end": model_t_end.isoformat(),
"model_timezone": config.model_timezone,
"generation_settings": GeneratorService(config)._generation_settings(),
"state_version": state.version,
"state": {"last_batch_id": state.last_batch_id},
}
service = GeneratorService(config)
assert not service._is_startup_history_state(state, manifest)
def test_startup_history_state_checks_state_version(
self, base_config, event_dictionary
):
"""Startup-history state сверяется с версией state из manifest."""
model_t0 = datetime(2026, 1, 1, 10, 0, tzinfo=timezone.utc)
model_t_end = model_t0 + timedelta(minutes=5)
config = replace(base_config, model_t0=model_t0, model_time_speed=10)
source_generator = EventGenerator(event_dictionary, config)
source_stream = TickStreamGenerator(source_generator)
source_stream.generate_tick(event_budget=10, tick_started_at=model_t0)
state = source_stream.to_state(
tick=5,
rng_state=source_generator.rng.getstate(),
last_batch_id="startup-history-state",
last_timestamp=model_t_end,
model_timestamp=model_t_end,
wall_timestamp=datetime(2026, 1, 1, 0, 0, tzinfo=timezone.utc),
model_time_speed=config.model_time_speed,
model_timezone=config.model_timezone,
model_t0=config.model_t0,
gen_seed=config.seed,
)
manifest = {
"run_mode": "backfill",
"gen_seed": config.seed,
"model_t0": model_t0.isoformat(),
"model_t_end": model_t_end.isoformat(),
"model_timezone": config.model_timezone,
"generation_settings": GeneratorService(config)._generation_settings(),
"state_version": "1.0",
"state": {"last_batch_id": state.last_batch_id},
}
@@ -930,8 +1060,8 @@ class TestGeneratorServiceStateV2:
assert saved_state.active_visits
service.state_manager.flush.assert_called_once()
def test_incompatible_seed_state_starts_fresh(self, base_config, event_dictionary, caplog):
"""State от другого GEN_SEED не смешивается с текущим запуском."""
def test_incompatible_seed_state_fails_loudly(self, base_config, event_dictionary, caplog):
"""State от другого GEN_SEED даёт жёсткий отказ при продолжении."""
source_config = replace(base_config, seed=7)
source_generator = EventGenerator(event_dictionary, source_config)
source_stream = TickStreamGenerator(source_generator)
@@ -970,11 +1100,10 @@ class TestGeneratorServiceStateV2:
caplog.at_level(logging.WARNING, logger="generator"):
service = GeneratorService(base_config)
service.start()
with pytest.raises(ValueError, match="GEN_SEED.*GEN_STATE_RESET=true"):
service.start()
assert service._tick == 0
assert service._model_time == base_config.model_t0
assert "state config mismatch: gen_seed" in caplog.text
assert "Readable generator state is incompatible" in caplog.text
def test_state_reset_skips_loading_saved_state(self, base_config):
"""GEN_STATE_RESET=true запускает сервис с чистого состояния."""
@@ -0,0 +1,518 @@
"""
Тесты портативного артефакта стартовой истории.
"""
from dataclasses import replace
from datetime import datetime, timezone
import pytest
from generator import GeneratorState
def _state() -> GeneratorState:
import random
now = datetime(2026, 1, 1, 1, 0, tzinfo=timezone.utc)
return GeneratorState(
tick=1,
rng_state=random.Random(42).getstate(),
last_batch_id="startup-history-abc",
last_timestamp=now,
model_timestamp=now,
wall_timestamp=now,
model_time_speed=1,
model_timezone="UTC",
model_t0=datetime(2026, 1, 1, 0, 0, tzinfo=timezone.utc),
gen_seed=42,
population=[
{
"user_domain_id": "user-1",
"seed_click_id": "seed-1",
"active_click_id": None,
"last_finished_at": None,
}
],
)
def _batch() -> dict[str, list[dict]]:
return {
"browser_events": [
{
"event_id": "event-1",
"click_id": "click-1",
"user_domain_id": "user-1",
"event_timestamp": "2026-01-01 00:00:00.000000",
}
],
"location_events": [
{
"event_id": "event-1",
"page_url_path": "/home",
}
],
"device_events": [
{
"click_id": "click-1",
"user_domain_id": "user-1",
}
],
"geo_events": [
{
"click_id": "click-1",
"country": "RU",
}
],
}
def test_artifact_roundtrip_keeps_events_state_and_manifest(base_config):
"""Артефакт хранит события, state и manifest одним проверяемым набором."""
from clickstream_generator.startup_history_artifact import (
StartupHistoryArtifactBuilder,
build_manifest,
load_startup_history_artifact,
validate_startup_history_artifact,
write_startup_history_artifact,
)
state = _state()
builder = StartupHistoryArtifactBuilder()
builder.add_batch(_batch())
manifest = build_manifest(
config=replace(
base_config,
model_t0=state.model_t0,
model_t_end=state.model_timestamp,
model_time_speed=1,
model_timezone="UTC",
seed=42,
),
counters=builder.counters,
state=state,
)
artifact = builder.to_artifact(manifest=manifest, state=state)
(
loaded_state,
loaded_manifest,
loaded_topics,
loaded_raw_topics,
) = validate_startup_history_artifact(
artifact,
expected_config=replace(
base_config,
model_t0=state.model_t0,
model_t_end=state.model_timestamp,
model_time_speed=1,
model_timezone="UTC",
seed=42,
),
)
assert artifact["artifact_version"] == "1.0"
assert loaded_state.last_batch_id == state.last_batch_id
assert loaded_manifest["state"]["last_batch_id"] == state.last_batch_id
assert loaded_topics["browser_events"][0]["event_id"] == "event-1"
assert loaded_raw_topics["browser_events"][0]["value_json"].encode("utf-8") == (
b'{"event_id": "event-1", "click_id": "click-1", '
b'"user_domain_id": "user-1", '
b'"event_timestamp": "2026-01-01 00:00:00.000000"}'
)
path = base_config.data_dir.parent / "tmp-startup-history-artifact.json"
try:
write_startup_history_artifact(path, artifact)
loaded = load_startup_history_artifact(path)
assert loaded["manifest"]["totals"]["events"] == 1
finally:
path.unlink(missing_ok=True)
def test_artifact_validation_rejects_mismatched_manifest(base_config):
"""Валидация отвергает артефакт, где manifest не совпадает с событиями."""
from clickstream_generator.startup_history_artifact import (
StartupHistoryArtifactBuilder,
build_manifest,
validate_startup_history_artifact,
)
state = _state()
config = replace(
base_config,
model_t0=state.model_t0,
model_t_end=state.model_timestamp,
model_time_speed=1,
model_timezone="UTC",
seed=42,
)
builder = StartupHistoryArtifactBuilder()
builder.add_batch(_batch())
manifest = build_manifest(config=config, counters=builder.counters, state=state)
artifact = builder.to_artifact(manifest=manifest, state=state)
artifact["manifest"]["gen_seed"] = 43
with pytest.raises(ValueError, match="GEN_SEED"):
validate_startup_history_artifact(artifact, expected_config=config)
def test_manifest_clickhouse_stats_detect_mismatch():
"""Сверка ClickHouse с manifest ловит расхождение контрольных чисел."""
from clickstream_generator.startup_history_artifact import (
compare_clickhouse_stats_to_manifest,
)
manifest = {
"totals": {
"events": 10,
"visits": 4,
"users": 3,
"min_event_timestamp": "2026-01-01 00:00:00.000000",
"max_event_timestamp": "2026-01-01 00:10:00.000000",
}
}
stats = {
"events": "9",
"visits": "4",
"users": "3",
"min_event_timestamp": "2026-01-01 00:00:00.000000",
"max_event_timestamp": "2026-01-01 00:10:00.000000",
}
assert compare_clickhouse_stats_to_manifest(manifest, stats) == ["events"]
def test_raw_topic_value_can_keep_original_json_bytes(base_config):
"""Raw value проверяется как JSON, но импортируется без пересборки строки."""
from clickstream_generator.startup_history_artifact import (
StartupHistoryArtifactBuilder,
build_manifest,
import_startup_history_artifact,
)
state = _state()
config = replace(
base_config,
model_t0=state.model_t0,
model_t_end=state.model_timestamp,
model_time_speed=1,
model_timezone="UTC",
seed=42,
)
builder = StartupHistoryArtifactBuilder()
builder.add_batch(_batch())
artifact = builder.to_artifact(
manifest=build_manifest(config=config, counters=builder.counters, state=state),
state=state,
)
raw_value = (
'{"event_timestamp":"2026-01-01 00:00:00.000000",'
'"user_domain_id":"user-1","click_id":"click-1","event_id":"event-1"}'
)
artifact["raw_topics"]["browser_events"][0]["value_json"] = raw_value
class Publisher:
def __init__(self):
self.records = None
def publish_records(self, topic, records):
if topic == "browser_events":
self.records = records
return len(records), 0
def flush(self):
return None
class CompactWriter:
def save(self, value):
self.saved = value
def flush(self):
return None
publisher = Publisher()
import_startup_history_artifact(
artifact,
publisher=publisher,
state_manager=CompactWriter(),
manifest_manager=CompactWriter(),
expected_config=config,
)
assert publisher.records[0]["value_json"] == raw_value
def test_artifact_v1_requires_raw_topics_for_import(base_config):
"""Artifact v1.0 без raw value не может обещать byte-for-byte импорт."""
from clickstream_generator.startup_history_artifact import (
StartupHistoryArtifactBuilder,
build_manifest,
import_startup_history_artifact,
)
state = _state()
config = replace(
base_config,
model_t0=state.model_t0,
model_t_end=state.model_timestamp,
model_time_speed=1,
model_timezone="UTC",
seed=42,
)
builder = StartupHistoryArtifactBuilder()
builder.add_batch(_batch())
artifact = builder.to_artifact(
manifest=build_manifest(config=config, counters=builder.counters, state=state),
state=state,
)
del artifact["raw_topics"]
class Publisher:
def publish_records(self, topic, records):
raise AssertionError("publish must not be called")
with pytest.raises(ValueError, match="raw_topics"):
import_startup_history_artifact(
artifact,
publisher=Publisher(),
state_manager=None,
manifest_manager=None,
expected_config=config,
)
def test_import_rejects_publisher_without_raw_publish_records(base_config):
"""Импорт не должен пересобирать dict, если publisher не умеет raw records."""
from clickstream_generator.startup_history_artifact import (
StartupHistoryArtifactBuilder,
build_manifest,
import_startup_history_artifact,
)
state = _state()
config = replace(
base_config,
model_t0=state.model_t0,
model_t_end=state.model_timestamp,
model_time_speed=1,
model_timezone="UTC",
seed=42,
)
builder = StartupHistoryArtifactBuilder()
builder.add_batch(_batch())
artifact = builder.to_artifact(
manifest=build_manifest(config=config, counters=builder.counters, state=state),
state=state,
)
artifact["raw_topics"]["browser_events"][0]["value_json"] = (
'{"event_timestamp":"2026-01-01 00:00:00.000000",'
'"user_domain_id":"user-1","click_id":"click-1","event_id":"event-1"}'
)
class DictPublisher:
def publish(self, topic, events):
raise AssertionError("dict publish must not be called")
def flush(self):
return None
with pytest.raises(TypeError, match="publish_records"):
import_startup_history_artifact(
artifact,
publisher=DictPublisher(),
state_manager=None,
manifest_manager=None,
expected_config=config,
)
def test_import_rejects_dirty_kafka_topics_before_publish(base_config):
"""Повторный импорт не должен дописывать дубли в непустые data-топики."""
from clickstream_generator.startup_history_artifact import (
StartupHistoryArtifactBuilder,
build_manifest,
import_startup_history_artifact,
)
state = _state()
config = replace(
base_config,
model_t0=state.model_t0,
model_t_end=state.model_timestamp,
model_time_speed=1,
model_timezone="UTC",
seed=42,
)
builder = StartupHistoryArtifactBuilder()
builder.add_batch(_batch())
artifact = builder.to_artifact(
manifest=build_manifest(config=config, counters=builder.counters, state=state),
state=state,
)
class Publisher:
def publish(self, topic, events):
raise AssertionError("publish must not be called")
class DirtyTopics:
def assert_data_topics_empty(self):
raise RuntimeError("Kafka data topics are not empty")
with pytest.raises(RuntimeError, match="not empty"):
import_startup_history_artifact(
artifact,
publisher=Publisher(),
state_manager=None,
manifest_manager=None,
expected_config=config,
topic_inspector=DirtyTopics(),
)
def test_import_rolls_back_kafka_topics_after_partial_publish(base_config):
"""При частичной публикации импорт откатывает Kafka-топики до повтора."""
from clickstream_generator.startup_history_artifact import (
StartupHistoryArtifactBuilder,
build_manifest,
import_startup_history_artifact,
)
state = _state()
config = replace(
base_config,
model_t0=state.model_t0,
model_t_end=state.model_timestamp,
model_time_speed=1,
model_timezone="UTC",
seed=42,
)
builder = StartupHistoryArtifactBuilder()
builder.add_batch(_batch())
artifact = builder.to_artifact(
manifest=build_manifest(config=config, counters=builder.counters, state=state),
state=state,
)
class PartialPublisher:
def __init__(self):
self.calls = []
def publish_records(self, topic, records):
self.calls.append(topic)
if topic == "location_events":
raise RuntimeError("location down")
return len(records), 0
class CompactWriter:
def save(self, value):
raise AssertionError("compact topics must not be written")
class Inspector:
def __init__(self):
self.snapshot = None
self.rollback = None
def assert_data_topics_empty(self):
return None
def snapshot_import_topics(self):
self.snapshot = {"topics": ["browser_events", "location_events"]}
return self.snapshot
def rollback_import_topics(self, snapshot):
self.rollback = snapshot
publisher = PartialPublisher()
inspector = Inspector()
with pytest.raises(RuntimeError, match="location down"):
import_startup_history_artifact(
artifact,
publisher=publisher,
state_manager=CompactWriter(),
manifest_manager=CompactWriter(),
expected_config=config,
topic_inspector=inspector,
)
assert publisher.calls == ["browser_events", "location_events"]
assert inspector.rollback == inspector.snapshot
def test_import_replays_events_and_compact_topics(base_config):
"""Импорт пишет события в Kafka и служебные compact-топики, не трогая ClickHouse."""
from clickstream_generator.startup_history_artifact import (
StartupHistoryArtifactBuilder,
build_manifest,
import_startup_history_artifact,
)
state = _state()
config = replace(
base_config,
model_t0=state.model_t0,
model_t_end=state.model_timestamp,
model_time_speed=1,
model_timezone="UTC",
seed=42,
)
builder = StartupHistoryArtifactBuilder()
builder.add_batch(_batch())
artifact = builder.to_artifact(
manifest=build_manifest(config=config, counters=builder.counters, state=state),
state=state,
)
class Publisher:
def __init__(self):
self.calls = []
self.flushed = False
def publish_records(self, topic, records):
self.calls.append((topic, records))
return len(records), 0
def flush(self):
self.flushed = True
class CompactWriter:
def __init__(self):
self.saved = None
self.flushed = False
def save(self, value):
self.saved = value
def flush(self):
self.flushed = True
publisher = Publisher()
state_manager = CompactWriter()
manifest_manager = CompactWriter()
result = import_startup_history_artifact(
artifact,
publisher=publisher,
state_manager=state_manager,
manifest_manager=manifest_manager,
expected_config=config,
)
assert [topic for topic, _ in publisher.calls] == [
"browser_events",
"location_events",
"device_events",
"geo_events",
]
assert publisher.calls[0][1][0]["value_json"].encode("utf-8") == (
b'{"event_id": "event-1", "click_id": "click-1", '
b'"user_domain_id": "user-1", '
b'"event_timestamp": "2026-01-01 00:00:00.000000"}'
)
assert result["events"] == 4
assert state_manager.saved.last_batch_id == state.last_batch_id
assert manifest_manager.saved["state"]["last_batch_id"] == state.last_batch_id
assert publisher.flushed
assert state_manager.flushed
assert manifest_manager.flushed