[AI-4964] Fix process isolation to handle complex string formats and external tags

datadog_checks_base/changelog.d/22973.fixed

@@ -0,0 +1 @@
+Fix process_isolation parameter to handle log formats and external tags.

datadog_checks_base/datadog_checks/base/utils/replay/execute.py

@@ -95,6 +95,10 @@ def _kill_on_timeout():
                 elif message_type == 'datadog_agent':
                     method = message['method']
                     value = getattr(datadog_agent, method)(*message['args'], **message['kwargs'])
+                    args = message['args']
+                    if method == 'set_external_tags':
+                        args = [[tuple(item) for item in args[0]]]
+                    value = getattr(datadog_agent, method)(*args, **message['kwargs'])
                     if method not in KNOWN_DATADOG_AGENT_SETTER_METHODS:
                         try:
                             process.stdin.write(b'%s\n' % ensure_bytes(json.encode({'value': value})))
@@ -110,7 +114,6 @@ def _kill_on_timeout():
                         message_type,
                     )
                     break
-
     if timed_out:
         check.log.error('Check timed out after %s seconds', timeout)
         check.warning('Check timed out and possibly reported incomplete data.')

datadog_checks_base/datadog_checks/base/utils/replay/redirect.py

@@ -59,11 +59,16 @@ def method(*args, **kwargs):
 
 
 class ReplayLogger(logging.Logger):
-    def log(self, level, *args, **kwargs):
+    def log(self, level, msg, *args, **kwargs):
+        if args:
+            try:
+                msg = msg % args
+            except (TypeError, ValueError, OverflowError, KeyError):
+                return
         print(
             '{}:log:{}'.format(
                 MESSAGE_INDICATOR,
-                to_native_string(json.encode({'method': LOG_METHODS[level], 'args': [str(a) for a in args]})),
+                to_native_string(json.encode({'method': LOG_METHODS[level], 'args': [str(msg)]})),
             )
         )
 

datadog_checks_base/tests/base/utils/replay/test_check.py

@@ -29,6 +29,9 @@ def initialize(self):
     def check(self, _):
         self.gauge('metric', 0, tags=self.tags)
         self.service_check('sc', ServiceCheck.OK if self.redirecting else ServiceCheck.CRITICAL, tags=self.tags)
+        self.log.debug('Metric count: %d', 42)
+        self.log.debug('Test log: %s', 'test')
+        self.set_external_tags([('myhost', {'src': ['tag:val']})])
 
 
 @pytest.mark.parametrize(
@@ -56,12 +59,33 @@ def test_replay_all(caplog, dd_run_check, aggregator, datadog_agent, init_config
     aggregator.assert_service_check('replay.sc', ServiceCheck.OK, count=1, tags=expected_tags)
     aggregator.assert_all_metrics_covered()
 
-    expected_message = 'Initializing - replay - test:123'
-    for _, level, message in caplog.record_tuples:
-        if level == logging.DEBUG and message == expected_message:
-            break
-    else:
-        raise AssertionError('Expected DEBUG log with message: {}'.format(expected_message))
+    captured_logs = {(level, message) for _, level, message in caplog.record_tuples}
+    assert (logging.DEBUG, 'Initializing - replay - test:123') in captured_logs
+    assert (logging.DEBUG, 'Metric count: 42') in captured_logs
+
+    datadog_agent.assert_external_tags('myhost', {'src': ['tag:val']})
+    datadog_agent.assert_external_tags_count(1)
+
+
+class ReplayCheckBadLog(AgentCheck):
+    __NAMESPACE__ = 'replay'
+
+    def check(self, _):
+        self.log.debug('TypeError format: %d', 'not_a_number')
+        self.log.debug('OverflowError format: %c', 2**32)
+
+
+def test_replay_skips_invalid_log_format(caplog, dd_run_check, aggregator, datadog_agent):
+    datadog_agent._config['log_level'] = 'debug'
+
+    check = ReplayCheckBadLog('replay', {}, [{'process_isolation': True}])
+    check.check_id = 'test:123'
+
+    with caplog.at_level(logging.DEBUG):
+        dd_run_check(check)
+
+    assert 'TypeError format' not in caplog.text
+    assert 'OverflowError format' not in caplog.text
 
 
 class SlowReplayCheck(AgentCheck):