AMBARI-10728. Changes for yarn service check by ditributed shell jar.(vbrodetskyi)

ambari-server/src/main/resources/common-services/YARN/2.1.0.2.0/package/scripts/params_linux.py

@@ -153,6 +153,9 @@ exclude_file_path = default("/configurations/yarn-site/yarn.resourcemanager.node
 ats_host = set(default("/clusterHostInfo/app_timeline_server_hosts", []))
 has_ats = not len(ats_host) == 0
 
+nm_hosts = default("/clusterHostInfo/nm_hosts", [])
+number_of_nm = len(nm_hosts)
+
 # default kinit commands
 rm_kinit_cmd = ""
 yarn_timelineservice_kinit_cmd = ""

ambari-server/src/main/resources/common-services/YARN/2.1.0.2.0/package/scripts/service_check.py

@@ -22,9 +22,13 @@ Ambari Agent
 from resource_management.libraries.functions.version import compare_versions
 from resource_management import *
 import sys
+import json
+import re
+import subprocess
 from ambari_commons import OSConst
 from ambari_commons.os_family_impl import OsFamilyImpl
 
+CURL_CONNECTION_TIMEOUT = '5'
 
 class ServiceCheck(Script):
   def service_check(self, env):
@@ -85,7 +89,7 @@ class ServiceCheckDefault(ServiceCheck):
       path_to_distributed_shell_jar = "/usr/lib/hadoop-yarn/hadoop-yarn-applications-distributedshell*.jar"
 
     yarn_distrubuted_shell_check_cmd = format("yarn org.apache.hadoop.yarn.applications.distributedshell.Client "
-            "-shell_command ls -jar {path_to_distributed_shell_jar}")
+                                              "-shell_command ls -num_containers {number_of_nm} -jar {path_to_distributed_shell_jar}")
 
     if params.security_enabled:
       kinit_cmd = format("{kinit_path_local} -kt {smoke_user_keytab} {smokeuser_principal};")
@@ -93,13 +97,43 @@ class ServiceCheckDefault(ServiceCheck):
     else:
       smoke_cmd = yarn_distrubuted_shell_check_cmd
 
-    Execute(smoke_cmd,
-            tries=3,
-            try_sleep=5,
-            path='/usr/sbin:/sbin:/usr/local/bin:/bin:/usr/bin',
-            user=params.smokeuser,
-            logoutput=True
-    )
+    return_code, out = shell.checked_call(smoke_cmd,
+                                          path='/usr/sbin:/sbin:/usr/local/bin:/bin:/usr/bin',
+                                          user=params.smokeuser,
+                                          )
+
+    m = re.search("appTrackingUrl=(.*),\s", out)
+    app_url = m.group(1)
+
+    splitted_app_url = str(app_url).split('/')
+
+    for item in splitted_app_url:
+      if "application" in item:
+        application_name = item
+
+    json_response_received = False
+    for rm_host in params.rm_hosts:
+      info_app_url = "http://" + rm_host + ":" + params.rm_port + "/ws/v1/cluster/apps/" + application_name
+
+      get_app_info_cmd = "curl --negotiate -u : -sL --connect-timeout " + CURL_CONNECTION_TIMEOUT + " " + info_app_url
+
+      return_code, stdout = shell.checked_call(get_app_info_cmd,
+                                            path='/usr/sbin:/sbin:/usr/local/bin:/bin:/usr/bin',
+                                            user=params.smokeuser,
+                                            )
+
+      try:
+        json_response = json.loads(stdout)
+
+        json_response_received = True
+
+        if json_response['app']['state'] != "FINISHED" or json_response['app']['finalStatus'] != "SUCCEEDED":
+          raise Exception("Application " + app_url + " state/status is not valid. Should be FINISHED/SUCCEEDED.")
+      except Exception as e:
+        pass
+
+    if not json_response_received:
+      raise Exception("Could not get json response from YARN API")
 
 
 if __name__ == "__main__":

ambari-server/src/main/resources/upgrade/catalog/UpgradeCatalog_2.1_to_2.2.4.json

@@ -509,4 +509,4 @@
       }
     }
   ]
-}
+}

ambari-server/src/test/python/stacks/2.0.6/YARN/test_yarn_service_check.py

@@ -18,6 +18,7 @@ See the License for the specific language governing permissions and
 limitations under the License.
 '''
 import os
+import re
 from mock.mock import MagicMock, call, patch
 from stacks.utils.RMFTestCase import *
 
@@ -27,40 +28,40 @@ class TestServiceCheck(RMFTestCase):
   COMMON_SERVICES_PACKAGE_DIR = "YARN/2.1.0.2.0/package"
   STACK_VERSION = "2.0.6"
 
-  def test_service_check_default(self):
+  @patch("re.search")
+  def test_service_check_default(self, re_search_mock):
+    m = MagicMock()
+    re_search_mock.return_value = m
+    m.group.return_value = "http://c6402.ambari.apache.org:8088/proxy/application_1429699682952_0010/"
 
     self.executeScript(self.COMMON_SERVICES_PACKAGE_DIR + "/scripts/service_check.py",
                           classname="ServiceCheck",
                           command="service_check",
                           config_file="default.json",
                           hdp_stack_version = self.STACK_VERSION,
-                          target = RMFTestCase.TARGET_COMMON_SERVICES
-    )
-    self.assertResourceCalled('Execute', 'yarn org.apache.hadoop.yarn.applications.distributedshell.Client '
-                                         '-shell_command ls -jar /usr/lib/hadoop-yarn/hadoop-yarn-applications-distributedshell*.jar',
-                          logoutput = True,
-                          path = ['/usr/sbin:/sbin:/usr/local/bin:/bin:/usr/bin'],
-                          tries = 3,
-                          user = 'ambari-qa',
-                          try_sleep = 5,
+                          target = RMFTestCase.TARGET_COMMON_SERVICES,
+                          checked_call_mocks = [(0, "some test text, appTrackingUrl=http:"
+                                "//c6402.ambari.apache.org:8088/proxy/application_1429885383763_0001/, some test text"),
+                                (0, "{ \"app\": {\"state\": \"FINISHED\",\"finalStatus\": \"SUCCEEDED\"}}")]
     )
     self.assertNoMoreResources()
 
-  def test_service_check_secured(self):
+
+  @patch("re.search")
+  def test_service_check_secured(self, re_search_mock):
+    m = MagicMock()
+    re_search_mock.return_value = m
+    m.group.return_value = "http://c6402.ambari.apache.org:8088/proxy/application_1429699682952_0010/"
+
     self.executeScript(self.COMMON_SERVICES_PACKAGE_DIR + "/scripts/service_check.py",
                           classname="ServiceCheck",
                           command="service_check",
                           config_file="secured.json",
                           hdp_stack_version = self.STACK_VERSION,
-                          target = RMFTestCase.TARGET_COMMON_SERVICES
-    )
-    self.assertResourceCalled('Execute', '/usr/bin/kinit -kt /etc/security/keytabs/smokeuser.headless.keytab '
-                                         'ambari-qa@EXAMPLE.COM; yarn org.apache.hadoop.yarn.applications.distributedshell.Client '
-                                         '-shell_command ls -jar /usr/lib/hadoop-yarn/hadoop-yarn-applications-distributedshell*.jar',
-                          logoutput = True,
-                          path = ['/usr/sbin:/sbin:/usr/local/bin:/bin:/usr/bin'],
-                          tries = 3,
-                          user = 'ambari-qa',
-                          try_sleep = 5,
+                          target = RMFTestCase.TARGET_COMMON_SERVICES,
+                          checked_call_mocks = [(0, "some test text, appTrackingUrl=http:"
+                               "//c6402.ambari.apache.org:8088/proxy/application_1429885383763_0001/, some test text"),
+                               (0, "{ \"app\": {\"state\": \"FINISHED\",\"finalStatus\": \"SUCCEEDED\"}}")]
     )
+
     self.assertNoMoreResources()