fix(worker): commit confirmed partition offsets and drain safely

This commit is contained in:
2026-09-19 21:30:33 +03:30
parent 33593609e3
commit 3b59d76f67
5 changed files with 483 additions and 125 deletions

View File

@@ -0,0 +1,193 @@
"""Durability regressions using broker acknowledgement and partition-aware doubles."""
import json
import time
import unittest
from concurrent.futures import Future
from types import SimpleNamespace
from unittest.mock import Mock, patch
from app.worker.consumer import ProcessingWorker
from app.worker.metrics import WorkerMetrics
from app.worker.normalizer import normalize_cdr_payload
def payload(**changes):
data = dict(raw_id="regression-1", device_id="310410998877665",
caller_num="0019875551234", svc="4G_DATA", rat=6,
up_bytes=100, down_bytes=200, lac="4021", ci="8812",
epoch_start=1726070000, epoch_end=1726070300, src_ip="10.0.0.1")
data.update(changes)
return data
class Message:
def __init__(self, data, offset=0, partition=0):
self.data = data if isinstance(data, bytes) else json.dumps(data).encode()
self.position = offset
self.partition_id = partition
def value(self): return self.data
def topic(self): return "cdr.raw"
def partition(self): return self.partition_id
def offset(self): return self.position
def timestamp(self): return (1, int((time.time() - 2) * 1000))
def error(self): return None
class Consumer:
def __init__(self, events, error=None, partition_error=False):
self.events = events
self.error = error
self.partition_error = partition_error
self.commits = []
def commit(self, offsets=None, message=None, asynchronous=False):
if self.error:
raise self.error
if offsets is not None:
positions = {(item.topic, item.partition): item.offset for item in offsets}
elif message is not None:
positions = {(message.topic(), message.partition()): message.offset() + 1}
else:
positions = "implicit-current-position"
self.events.append(("commit", positions))
self.commits.append(positions)
return [SimpleNamespace(error=RuntimeError("partition commit failed"))] if self.partition_error else []
def close(self): pass
class Producer:
def __init__(self, events, failures=()):
self.events = events
self.failures = set(failures)
self.messages = []
def produce(self, topic, key, value):
self.messages.append(json.loads(value))
result = Future()
if len(self.messages) in self.failures:
result.set_exception(ConnectionError("broker rejected DLQ record"))
else:
self.events.append(("dlq_ack", key))
result.set_result(None)
return result
def close(self): pass
def flush(self, timeout=5): return 0
class DeliverySafetyTests(unittest.TestCase):
def setUp(self):
self.events = []
self.consumer = Consumer(self.events)
self.producer = Producer(self.events)
self.ch = Mock()
self.ch.insert_batch.side_effect = lambda records: self.events.append(("insert", len(records))) or True
self.metrics = WorkerMetrics()
self.patches = [patch("app.worker.consumer.worker_metrics", self.metrics),
patch("app.worker.dlq.worker_metrics", self.metrics)]
for item in self.patches:
item.start()
self.addCleanup(item.stop)
self.worker = ProcessingWorker(consumer=self.consumer, dlq_producer=self.producer,
clickhouse=self.ch, batch_size=100, internal_server=Mock())
def test_invalid_record_cannot_commit_past_buffered_valid_record(self):
self.worker.process_single_message(Message(payload(), offset=0))
self.worker.process_single_message(Message(b"{malformed", offset=1))
self.assertEqual(self.events[0], ("insert", 1))
self.assertEqual(self.consumer.commits[-1], {("cdr.raw", 0): 2})
self.assertEqual(len(self.worker.buffer), 0)
def test_batch_commits_only_its_explicit_partition_offsets(self):
self.worker.process_single_message(Message(payload(), offset=4, partition=0))
self.worker.process_single_message(Message(payload(raw_id="second"), offset=9, partition=1))
self.worker.flush_buffer()
self.assertEqual(self.consumer.commits, [{("cdr.raw", 0): 5, ("cdr.raw", 1): 10}])
def test_failed_dlq_ack_never_commits_or_counts_success(self):
self.producer.failures = {1}
with self.assertRaises(Exception):
self.worker.process_single_message(Message(b"{malformed"))
self.assertEqual(self.consumer.commits, [])
self.assertEqual(self.metrics.get_internal_stats()["pipeline"]["total_dlq"], 0)
def test_partial_batch_dlq_failure_keeps_all_offsets_uncommitted(self):
self.ch.insert_batch.side_effect = ConnectionError("ClickHouse unavailable")
self.producer.failures = {2}
self.worker.process_single_message(Message(payload(), offset=0))
self.worker.process_single_message(Message(payload(raw_id="second"), offset=1))
with self.assertRaises(Exception):
self.worker.flush_buffer()
self.assertEqual(self.consumer.commits, [])
self.assertEqual(self.metrics.get_internal_stats()["pipeline"]["total_dlq"], 1)
def test_commit_exception_propagates_without_routing_inserted_batch_to_dlq(self):
self.consumer.error = ConnectionError("commit unavailable")
self.worker.process_single_message(Message(payload()))
with self.assertRaises(Exception):
self.worker.flush_buffer()
self.ch.insert_batch.assert_called_once()
self.assertEqual(self.producer.messages, [])
def test_partition_commit_error_is_not_success(self):
self.consumer.partition_error = True
self.worker.process_single_message(Message(payload()))
with self.assertRaises(Exception):
self.worker.flush_buffer()
self.assertEqual(self.producer.messages, [])
def test_clickhouse_failure_dlq_retains_original_vendor_payload(self):
self.ch.insert_batch.side_effect = ConnectionError("ClickHouse unavailable")
original = payload(vendor_extension={"keep": True})
self.worker.process_single_message(Message(original))
self.worker.flush_buffer()
received = self.producer.messages[0]["original_payload"]
self.assertEqual(json.loads(received) if isinstance(received, str) else received, original)
def test_successful_insert_populates_kafka_latency(self):
self.worker.process_single_message(Message(payload()))
self.worker.flush_buffer()
self.assertIn("cdr_kafka_to_clickhouse_latency_seconds_count 1", self.metrics.get_prometheus_metrics())
class InputBoundaryTests(unittest.TestCase):
def test_invalid_values_are_rejected_without_raising(self):
cases = [("epoch_start", 10**100), ("epoch_end", 10**100), ("epoch_end", 2**32),
("epoch_start", -1), ("epoch_start", 1726070000.5),
("rat", 6.9), ("rat", True), ("up_bytes", -0.5),
("up_bytes", True), ("up_bytes", 2**64),
("up_bytes", 2**64 - 1), ("src_ip", "999.1.2.3"),
("src_ip", "::1"), ("raw_id", 123)]
for field, value in cases:
with self.subTest(field=field, value=value):
record, reason, failed_field, _ = normalize_cdr_payload(json.dumps(payload(**{field: value})).encode())
self.assertIsNone(record)
self.assertIsNotNone(reason)
self.assertIsNotNone(failed_field)
def test_invalid_utf8_and_nonfinite_json_are_rejected(self):
for value in (b"\xff", json.dumps(payload(up_bytes=float("nan"))).encode()):
with self.subTest(value=value):
record, reason, _, _ = normalize_cdr_payload(value)
self.assertIsNone(record)
self.assertIsNotNone(reason)
def test_clickhouse_rows_use_utc_and_native_ipv4(self):
from datetime import timedelta
from ipaddress import IPv4Address
record, reason, _, _ = normalize_cdr_payload(json.dumps(payload()).encode())
self.assertIsNone(reason)
row = record.to_row()
self.assertEqual(row[11].utcoffset(), timedelta(0))
self.assertIsInstance(row[13], IPv4Address)
def test_prometheus_labels_escape_quotes_backslashes_and_newlines(self):
metrics = WorkerMetrics()
metrics.record_processed('svc"\\\nname', "LTE_4G")
self.assertIn('service_type="svc\\"\\\\\\nname"', metrics.get_prometheus_metrics())
if __name__ == "__main__":
unittest.main()