| 123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223 |
- # -*- coding:utf-8 -*-
- """
- bin/cdc-monitor.py 单测:复制槽判定 / YARN 存活 / 消息渲染 / main 串联。
- 不连真 PG(pgdb.connect + query 走替身)、不跑真 yarn(subprocess 走替身)、
- 不发真消息(Alerter 走替身)。
- 脚本路径含连字符,用 importlib.util 动态加载为模块。
- """
- import importlib.util
- import os
- import re
- import sys
- from datetime import datetime
- from unittest.mock import MagicMock
- import pytest
- PROJECT_ROOT = os.path.dirname(os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))))
- SCRIPT_PATH = os.path.join(PROJECT_ROOT, 'bin', 'cdc-monitor.py')
- GB = 1024 ** 3
- LAG_CRIT = 5 * GB
- def _load_script():
- spec = importlib.util.spec_from_file_location('cdc_monitor', SCRIPT_PATH)
- mod = importlib.util.module_from_spec(spec)
- sys.modules['cdc_monitor'] = mod
- spec.loader.exec_module(mod)
- return mod
- MON = _load_script()
- def _stub_pg(monkeypatch, rows, conn=None):
- """替身 PG:connect 返回 conn,query 返回 rows。返回该 conn 供断言 close。"""
- conn = conn or MagicMock()
- monkeypatch.setattr(MON.pgdb, 'connect', lambda ds_ref: conn)
- monkeypatch.setattr(MON.pgdb, 'query', lambda c, sql: rows)
- return conn
- # ---------- 复制槽 ----------
- def test_cdc_slot_inactive_alerts(monkeypatch):
- _stub_pg(monkeypatch, [('cdc_large', False, 0)])
- assert MON.check_slots('postgresql/x', LAG_CRIT) == [
- ('槽 cdc_large', '空闲,采集作业可能已停')]
- def test_non_cdc_slot_inactive_is_ignored(monkeypatch):
- """PG→PG 那条槽 apply 卡住时会反复重连,active 抖动不算告警。"""
- _stub_pg(monkeypatch, [('finance_pg2pg', False, 0)])
- assert MON.check_slots('postgresql/x', LAG_CRIT) == []
- def test_slot_lag_over_threshold_alerts(monkeypatch):
- _stub_pg(monkeypatch, [('cdc_small', True, 6 * GB)])
- alerts = MON.check_slots('postgresql/x', LAG_CRIT)
- assert alerts == [('槽 cdc_small', '滞后 6.0 GB')]
- def test_slot_lag_at_threshold_not_alerted(monkeypatch):
- """阈值是 >,正好等于不报。"""
- _stub_pg(monkeypatch, [('cdc_small', True, LAG_CRIT)])
- assert MON.check_slots('postgresql/x', LAG_CRIT) == []
- def test_lag_applies_to_non_cdc_slot_too(monkeypatch):
- """active 只对 cdc_ 判,滞后对所有槽都判。"""
- _stub_pg(monkeypatch, [('finance_pg2pg', True, 9 * GB)])
- assert MON.check_slots('postgresql/x', LAG_CRIT) == [
- ('槽 finance_pg2pg', '滞后 9.0 GB')]
- def test_null_lag_does_not_crash(monkeypatch):
- """confirmed_flush_lsn 为 NULL 时 lag 是 None,当 0 处理。"""
- _stub_pg(monkeypatch, [('cdc_large', True, None)])
- assert MON.check_slots('postgresql/x', LAG_CRIT) == []
- def test_inactive_and_lagging_yields_two_alerts(monkeypatch):
- _stub_pg(monkeypatch, [('cdc_large', False, 7 * GB)])
- assert MON.check_slots('postgresql/x', LAG_CRIT) == [
- ('槽 cdc_large', '空闲,采集作业可能已停'),
- ('槽 cdc_large', '滞后 7.0 GB'),
- ]
- def test_connection_closed_after_query(monkeypatch):
- conn = _stub_pg(monkeypatch, [])
- MON.check_slots('postgresql/x', LAG_CRIT)
- conn.close.assert_called_once()
- def test_db_failure_becomes_alert(monkeypatch):
- def _boom(ds_ref):
- raise RuntimeError('数据源 ini 不存在')
- monkeypatch.setattr(MON.pgdb, 'connect', _boom)
- alerts = MON.check_slots('postgresql/x', LAG_CRIT)
- assert len(alerts) == 1
- assert alerts[0][0] == '主库'
- assert '数据源 ini 不存在' in alerts[0][1]
- # ---------- YARN ----------
- def test_all_jobs_running_no_alert(monkeypatch):
- monkeypatch.setattr(MON.subprocess, 'check_output',
- lambda *a, **kw: b'app-1 st-cdc-large RUNNING\napp-2 st-cdc-small RUNNING')
- assert MON.check_yarn(['st-cdc-large', 'st-cdc-small']) == []
- def test_missing_job_alerts(monkeypatch):
- monkeypatch.setattr(MON.subprocess, 'check_output',
- lambda *a, **kw: b'app-1 st-cdc-large RUNNING')
- assert MON.check_yarn(['st-cdc-large', 'st-cdc-small']) == [
- ('作业 st-cdc-small', '不在 YARN RUNNING 列表')]
- def test_yarn_failure_becomes_single_alert(monkeypatch):
- def _boom(*a, **kw):
- raise OSError('yarn: command not found')
- monkeypatch.setattr(MON.subprocess, 'check_output', _boom)
- alerts = MON.check_yarn(['st-cdc-large', 'st-cdc-small'])
- assert len(alerts) == 1 # 查不到就是一条,不按作业数量刷屏
- assert alerts[0][0] == 'YARN'
- assert 'command not found' in alerts[0][1]
- # ---------- 渲染:走一圈全部五类告警 ----------
- FIXED_NOW = datetime(2026, 8, 26, 15, 21, 27)
- def test_render_single_alert():
- assert MON.render([('槽 cdc_large', '空闲,采集作业可能已停')], now=FIXED_NOW) == (
- '### <font color="warning">CDC 监控告警</font>\n'
- '> 检测时间:2026-08-26 15:21:27\n'
- '> 槽 cdc_large:<font color="warning">空闲,采集作业可能已停</font>')
- def test_render_defaults_to_current_time(monkeypatch):
- """不传 now 时取当前时间,格式 yyyy-MM-dd HH:mm:ss。"""
- out = MON.render([('主库', '连不上')])
- assert re.match(r'^> 检测时间:\d{4}-\d{2}-\d{2} \d{2}:\d{2}:\d{2}$', out.split('\n')[1])
- def test_render_all_alert_kinds():
- alerts = [
- ('槽 cdc_large', '空闲,采集作业可能已停'),
- ('槽 cdc_small', '滞后 6.0 GB'),
- ('作业 st-cdc-large', '不在 YARN RUNNING 列表'),
- ('主库', '复制槽查询失败:connection refused'),
- ('YARN', '查询失败:timeout'),
- ]
- out = MON.render(alerts, now=FIXED_NOW)
- lines = out.split('\n')
- assert lines[0] == '### <font color="warning">CDC 监控告警</font>'
- assert lines[1] == '> 检测时间:2026-08-26 15:21:27'
- assert len(lines) == 7
- for (key, desc), line in zip(alerts, lines[2:]):
- assert line == '> {0}:<font color="warning">{1}</font>'.format(key, desc)
- # ---------- main 串联 ----------
- def _run_main(monkeypatch, slot_alerts, yarn_alerts):
- """跑 main,返回替身 alerter 实例供断言。"""
- monkeypatch.setattr(sys, 'argv', ['cdc-monitor.py'])
- monkeypatch.setattr(MON, 'check_slots', lambda ds, lag: list(slot_alerts))
- monkeypatch.setattr(MON, 'check_yarn', lambda jobs: list(yarn_alerts))
- instance = MagicMock()
- monkeypatch.setattr(MON, 'Alerter', MagicMock(return_value=instance))
- MON.main()
- return instance
- def test_main_sends_nothing_when_healthy(monkeypatch):
- instance = _run_main(monkeypatch, [], [])
- instance.send_markdown.assert_not_called()
- def test_main_sends_once_with_all_alerts(monkeypatch):
- instance = _run_main(
- monkeypatch,
- [('槽 cdc_large', '空闲,采集作业可能已停')],
- [('作业 st-cdc-small', '不在 YARN RUNNING 列表')])
- instance.send_markdown.assert_called_once()
- content = instance.send_markdown.call_args[0][0]
- assert '槽 cdc_large' in content
- assert '作业 st-cdc-small' in content
- def test_main_builds_alerter_before_checks(monkeypatch):
- """健康时也要构造 Alerter:配置坏了要立刻暴露,不能等真出事才发现发不出去。"""
- monkeypatch.setattr(sys, 'argv', ['cdc-monitor.py'])
- monkeypatch.setattr(MON, 'check_slots', lambda ds, lag: [])
- monkeypatch.setattr(MON, 'check_yarn', lambda jobs: [])
- fake_cls = MagicMock()
- monkeypatch.setattr(MON, 'Alerter', fake_cls)
- MON.main()
- fake_cls.assert_called_once_with(channel='default')
- def test_main_passes_cli_overrides(monkeypatch):
- seen = {}
- monkeypatch.setattr(sys, 'argv', [
- 'cdc-monitor.py', '-ds', 'postgresql/other', '-channel', 'realtime',
- '-jobs', 'a, b ,', '-lag-gb', '2'])
- monkeypatch.setattr(MON, 'check_slots',
- lambda ds, lag: seen.update(ds=ds, lag=lag) or [])
- monkeypatch.setattr(MON, 'check_yarn', lambda jobs: seen.update(jobs=jobs) or [])
- fake_cls = MagicMock()
- monkeypatch.setattr(MON, 'Alerter', fake_cls)
- MON.main()
- assert seen['ds'] == 'postgresql/other'
- assert seen['lag'] == 2 * GB
- assert seen['jobs'] == ['a', 'b'] # 逗号分隔去空白、丢空项
- fake_cls.assert_called_once_with(channel='realtime')
|