filebeat: Fix flaky test case on macOS (#39860) (#39897)

* fix: fix a flaky test on macos

* fix: fix more such test cases

* fix: only update ignore_older

* fix: also fix flaky test_restart_state

* fix: fix CI

(cherry picked from commit f9fec1e)

Co-authored-by: VihasMakwana <[email protected]>

filebeat/tests/system/test_crawler.py

@@ -485,7 +485,7 @@ def test_tail_files(self):
         self.wait_until(
             lambda: self.log_contains(
                 "Start next scan"),
-            max_timeout=5)
+            max_timeout=10)
 
         with open(testfile, 'a') as f:
             # write additional lines
@@ -596,7 +596,7 @@ def test_encodings(self):
         # run filebeat
         filebeat = self.start_beat()
         self.wait_until(lambda: self.output_has(lines=len(encodings)),
-                        max_timeout=15)
+                        max_timeout=25)
 
         # write another line in all files
         for _, enc_py, text in encodings:

filebeat/tests/system/test_harvester.py

@@ -858,6 +858,6 @@ def test_debug_reader(self):
 
         # 13 on unix, 14 on windows.
         self.wait_until(lambda: self.log_contains(re.compile(
-            'Matching null byte found at offset (13|14)')), max_timeout=5)
+            'Matching null byte found at offset (13|14)')), max_timeout=10)
 
         filebeat.check_kill_and_wait()

filebeat/tests/system/test_registrar.py

@@ -252,9 +252,11 @@ def test_registry_file_update_permissions(self):
 
         self.assertEqual(self.file_permissions(os.path.join(registry_path, "log.json")), "0o600")
 
+        registry_home = "a/b/c/d/registry_x"
+        registry_path = os.path.join(registry_home, "filebeat")
         self.render_config_template(
             path=os.path.abspath(self.working_dir) + "/log/*",
-            registry_home="a/b/c/registry_x",
+            registry_home=registry_home,
             registry_file_permissions=0o640
         )
 
@@ -266,7 +268,7 @@ def test_registry_file_update_permissions(self):
         # the logging and actual writing the file. Seems to happen on Windows.
         self.wait_until(
             lambda: self.has_registry(registry_path),
-            max_timeout=1)
+            max_timeout=10)
 
         # Wait a moment to make sure registry is completely written
         time.sleep(1)
@@ -950,7 +952,7 @@ def test_restart_state(self):
             path=os.path.abspath(self.working_dir) + "/log/*",
             close_inactive="200ms",
             ignore_older="2000ms",
-            clean_inactive="3s",
+            clean_inactive="10s",
         )
 
         filebeat = self.start_beat()
@@ -976,7 +978,7 @@ def test_restart_state_reset(self):
         self.render_config_template(
             path=os.path.abspath(self.working_dir) + "/log/*",
             clean_inactive="10s",
-            ignore_older="5s"
+            ignore_older="9s"
         )
         os.mkdir(self.working_dir + "/log/")
 
@@ -1003,7 +1005,7 @@ def test_restart_state_reset(self):
         self.render_config_template(
             path=os.path.abspath(self.working_dir) + "/log/test2.log",
             clean_inactive="10s",
-            ignore_older="5s",
+            ignore_older="9s",
         )
 
         filebeat = self.start_beat(output="filebeat2.log")
@@ -1137,7 +1139,7 @@ def test_restart_state_reset_ttl_no_clean_inactive(self):
         self.render_config_template(
             path=os.path.abspath(self.working_dir) + "/log/test.log",
             clean_inactive="10s",
-            ignore_older="5s"
+            ignore_older="9s"
         )
         os.mkdir(self.working_dir + "/log/")
 

filebeat/tests/system/test_shutdown.py

@@ -27,7 +27,13 @@ def test_shutdown(self):
         )
         for i in range(1, 5):
             proc = self.start_beat(logging_args=["-e", "-v"])
-            time.sleep(.5)
+
+            # Flaky on MacOS, see https://github.com/elastic/beats/issues/39613#issuecomment-2158812325
+            # we need to wait a bit longer for filebeat to start
+            if platform.system() == "Darwin":
+                time.sleep(10)
+            else:
+                time.sleep(.5)
             proc.check_kill_and_wait()
 
     @unittest.skip("Skipped as flaky: https://github.com/elastic/beats/issues/14647")