Add API for uploading logs via host plugin

azurelinuxagent/common/protocol/hostplugin.py

@@ -20,29 +20,39 @@
 import base64
 import datetime
 import json
+import uuid
 
 from azurelinuxagent.common import logger
 from azurelinuxagent.common.errorstate import ErrorState, ERROR_STATE_HOST_PLUGIN_FAILURE
+from azurelinuxagent.common.event import WALAEventOperation, report_event
 from azurelinuxagent.common.exception import HttpError, ProtocolError
 from azurelinuxagent.common.future import ustr
 from azurelinuxagent.common.protocol.healthservice import HealthService
 from azurelinuxagent.common.utils import restutil
 from azurelinuxagent.common.utils import textutil
 from azurelinuxagent.common.utils.textutil import remove_bom
-from azurelinuxagent.common.version import PY_VERSION_MAJOR
+from azurelinuxagent.common.version import AGENT_NAME, AGENT_VERSION, PY_VERSION_MAJOR
 
 HOST_PLUGIN_PORT = 32526
+
 URI_FORMAT_GET_API_VERSIONS = "http://{0}:{1}/versions"
 URI_FORMAT_GET_EXTENSION_ARTIFACT = "http://{0}:{1}/extensionArtifact"
 URI_FORMAT_PUT_VM_STATUS = "http://{0}:{1}/status"
 URI_FORMAT_PUT_LOG = "http://{0}:{1}/vmAgentLog"
 URI_FORMAT_HEALTH = "http://{0}:{1}/health"
+
 API_VERSION = "2015-09-01"
+
+HEADER_CLIENT_NAME = "x-ms-client-name"
+HEADER_CLIENT_VERSION = "x-ms-client-version"
+HEADER_CORRELATION_ID = "x-ms-client-correlationid"
 HEADER_CONTAINER_ID = "x-ms-containerid"
+HEADER_DEPLOYMENT_ID = "x-ms-vmagentlog-deploymentid"
 HEADER_VERSION = "x-ms-version"
 HEADER_HOST_CONFIG_NAME = "x-ms-host-config-name"
 HEADER_ARTIFACT_LOCATION = "x-ms-artifact-location"
 HEADER_ARTIFACT_MANIFEST_LOCATION = "x-ms-artifact-manifest-location"
+
 MAXIMUM_PAGEBLOB_PAGE_SIZE = 4 * 1024 * 1024  # Max page size: 4MB
 
 
@@ -60,7 +70,7 @@ def __init__(self, endpoint, container_id, role_config_name):
         self.api_versions = None
         self.endpoint = endpoint
         self.container_id = container_id
-        self.deployment_id = None
+        self.deployment_id = self._extract_deployment_id(role_config_name)
         self.role_config_name = role_config_name
         self.manifest_uri = None
         self.health_service = HealthService(endpoint)
@@ -69,6 +79,11 @@ def __init__(self, endpoint, container_id, role_config_name):
         self.fetch_last_timestamp = None
         self.status_last_timestamp = None
 
+    @staticmethod
+    def _extract_deployment_id(role_config_name):
+        # Role config name consists of: <deployment id>.<incarnation>(...)
+        return role_config_name.split(".")[0] if role_config_name is not None else None
+
     @staticmethod
     def is_default_channel():
         return HostPluginProtocol._is_default_channel
@@ -82,6 +97,7 @@ def update_container_id(self, new_container_id):
 
     def update_role_config_name(self, new_role_config_name):
         self.role_config_name = new_role_config_name
+        self.deployment_id = self._extract_deployment_id(new_role_config_name)
 
     def update_manifest_uri(self, new_manifest_uri):
         self.manifest_uri = new_manifest_uri
@@ -91,7 +107,6 @@ def ensure_initialized(self):
             self.api_versions = self.get_api_versions()
             self.is_available = API_VERSION in self.api_versions
             self.is_initialized = self.is_available
-            from azurelinuxagent.common.event import WALAEventOperation, report_event
             report_event(WALAEventOperation.InitializeHostPlugin,
                          is_success=self.is_available)
         return self.is_available
@@ -202,7 +217,29 @@ def should_report(is_healthy, error_state, last_timestamp, period):
         return datetime.datetime.utcnow() >= (last_timestamp + period)
 
     def put_vm_log(self, content):
-        raise NotImplementedError("Unimplemented")
+        """
+        Try to upload VM logs, a compressed zip file, via the host plugin /vmAgentLog channel.
+        :param content: the binary content of the zip file to upload
+        """
+        if not self.ensure_initialized():
+            raise ProtocolError("HostGAPlugin: HostGAPlugin is not available")
+
+        if content is None or self.container_id is None or self.deployment_id is None:
+            raise ProtocolError("HostGAPlugin: Invalid arguments passed to upload VM logs. "
+                                "Content: {0}, container id: {1}, deployment id: {2}".format(content,
+                                                                                             self.container_id,
+                                                                                             self.deployment_id))
+
+        url = URI_FORMAT_PUT_LOG.format(self.endpoint, HOST_PLUGIN_PORT)
+        response = restutil.http_put(url,
+                                     data=content,
+                                     headers=self._build_log_headers())
+
+        if restutil.request_failed(response):
+            error_response = restutil.read_response_error(response)
+            raise HttpError("HostGAPlugin: Upload VM logs failed: {0}".format(error_response))
+        else:
+            logger.info("HostGAPlugin: Upload VM logs succeeded")
 
     def put_vm_status(self, status_blob, sas_url, config_blob_type=None):
         """
@@ -327,7 +364,17 @@ def _build_status_headers(self):
             HEADER_CONTAINER_ID: self.container_id,
             HEADER_HOST_CONFIG_NAME: self.role_config_name
         }
-
+
+    def _build_log_headers(self):
+        return {
+            HEADER_VERSION: API_VERSION,
+            HEADER_CONTAINER_ID: self.container_id,
+            HEADER_DEPLOYMENT_ID: self.deployment_id,
+            HEADER_CLIENT_NAME: AGENT_NAME,
+            HEADER_CLIENT_VERSION: AGENT_VERSION,
+            HEADER_CORRELATION_ID: str(uuid.uuid4())
+        }
+
     def _base64_encode(self, data):
         s = base64.b64encode(bytes(data))
         if PY_VERSION_MAJOR > 2:

azurelinuxagent/common/protocol/wire.py

@@ -29,7 +29,8 @@
 import azurelinuxagent.common.logger as logger
 import azurelinuxagent.common.utils.textutil as textutil
 from azurelinuxagent.common.datacontract import validate_param
-from azurelinuxagent.common.event import add_event, add_periodic, WALAEventOperation, EVENTS_DIRECTORY
+from azurelinuxagent.common.event import add_event, add_periodic, EVENTS_DIRECTORY
+from azurelinuxagent.common.event import report_event, WALAEventOperation
 from azurelinuxagent.common.exception import ProtocolNotFoundError, \
     ResourceGoneError, ExtensionDownloadError, InvalidContainerError, ProtocolError, HttpError
 from azurelinuxagent.common.future import httpclient, bytebuffer, ustr
@@ -192,6 +193,9 @@ def report_event(self, events):
         validate_param(EVENTS_DIRECTORY, events, TelemetryEventList)
         self.client.report_event(events)
 
+    def upload_logs(self, logs):
+        self.client.upload_logs(logs)
+
 
 def _build_role_properties(container_id, role_instance_id, thumbprint):
     xml = (u"<?xml version=\"1.0\" encoding=\"utf-8\"?>"
@@ -1116,9 +1120,6 @@ def report_event(self, event_list):
                 self.send_event(provider_id, buf[provider_id])
 
     def report_status_event(self, message, is_success):
-        from azurelinuxagent.common.event import report_event, \
-            WALAEventOperation
-
         report_event(op=WALAEventOperation.ReportStatus,
                      is_success=is_success,
                      message=message,
@@ -1195,14 +1196,73 @@ def get_artifacts_profile(self):
                     msg = "Content: [{0}]".format(profile)
                     logger.verbose(msg)
 
-                    from azurelinuxagent.common.event import report_event, WALAEventOperation
                     report_event(op=WALAEventOperation.ArtifactsProfileBlob,
                                  is_success=False,
                                  message=msg,
                                  log_event=False)
 
         return artifacts_profile
 
+    def upload_logs(self, content):
+        try:
+            host_plugin = self.get_host_plugin()
+            host_plugin.put_vm_log(content)
+
+            msg = "Upload VM logs request succeeded using the host plugin channel for " \
+                  "container id {0} and role config file {1}".format(host_plugin.container_id,
+                                                                     host_plugin.role_config_name)
+            add_event(name=AGENT_NAME,
+                      version=CURRENT_VERSION,
+                      op=WALAEventOperation.HostPlugin,
+                      is_success=True,
+                      message=msg,
+                      log_event=True)
+        except (ResourceGoneError, InvalidContainerError) as e:
+            host_plugin = self.get_host_plugin()
+            old_container_id = host_plugin.container_id
+            old_role_config_name = host_plugin.role_config_name
+
+            msg = "Upload VM logs request failed with the current host plugin configuration. " \
+                  "ContainerId: {0}, role config file: {1}. Fetching new goal state and retrying the call." \
+                  "Error: {2}".format(old_container_id, old_role_config_name, ustr(e))
+            logger.info(msg)
+
+            self.update_host_plugin_from_goal_state()
+
+            new_container_id = host_plugin.container_id
+            new_role_config_name = host_plugin.role_config_name
+            msg = "Host plugin reconfigured with new parameters. " \
+                  "ContainerId: {0}, role config file: {1}.".format(new_container_id, new_role_config_name)
+            logger.info(msg)
+
+            try:
+                host = self.get_host_plugin()
+                host.put_vm_log(content)
+
+                msg = "Upload VM logs request succeeded using the host plugin channel after goal state refresh. " \
+                      "ContainerId changed from {0} to {1}, " \
+                      "role config file changed from {2} to {3}.".format(old_container_id, new_container_id,
+                                                                         old_role_config_name, new_role_config_name)
+                add_event(name=AGENT_NAME,
+                          version=CURRENT_VERSION,
+                          op=WALAEventOperation.HostPlugin,
+                          is_success=True,
+                          message=msg,
+                          log_event=True)
+
+            except (ResourceGoneError, InvalidContainerError) as e:
+                msg = "Upload VM logs request failed using the host plugin channel after goal state refresh. " \
+                      "ContainerId changed from {0} to {1}, role config file changed from {2} to {3}. " \
+                      "Exception type: {4}.".format(old_container_id, new_container_id, old_role_config_name,
+                                                    new_role_config_name, type(e).__name__)
+                add_event(name=AGENT_NAME,
+                          version=CURRENT_VERSION,
+                          op=WALAEventOperation.HostPlugin,
+                          is_success=False,
+                          message=msg,
+                          log_event=True)
+                raise
+
 
 class VersionInfo(object):
     def __init__(self, xml_text):

tests/common/test_event.py

@@ -22,19 +22,21 @@
 import re
 import shutil
 import threading
-import xml.dom
 from datetime import datetime, timedelta
+
+import xml.dom
+
 import azurelinuxagent.common.utils.textutil as textutil
 from azurelinuxagent.common import event, logger
 from azurelinuxagent.common.AgentGlobals import AgentGlobals
 from azurelinuxagent.common.event import add_event, add_periodic, add_log_event, elapsed_milliseconds, report_metric, \
     WALAEventOperation, parse_xml_event, parse_json_event, AGENT_EVENT_FILE_EXTENSION, EVENTS_DIRECTORY
 from azurelinuxagent.common.future import ustr
+from azurelinuxagent.common.osutil import get_osutil
+from azurelinuxagent.common.version import CURRENT_AGENT, CURRENT_VERSION, AGENT_EXECUTION_MODE
 from tests.protocol import mockwiredata
 from tests.protocol.mocks import mock_wire_protocol, HttpRequestPredicates, MockHttpResponse
-from azurelinuxagent.common.version import CURRENT_AGENT, CURRENT_VERSION, AGENT_EXECUTION_MODE
-from azurelinuxagent.common.osutil import get_osutil
-from tests.tools import AgentTestCase, data_dir, load_data, Mock, patch, skip_if_predicate_true
+from tests.tools import AgentTestCase, data_dir, load_data, patch, skip_if_predicate_true
 from tests.utils.event_logger_tools import EventLoggerTools
 
 
@@ -48,8 +50,8 @@ def setUp(self):
         osutil = get_osutil()
 
         self.expected_common_parameters = {
-            # common parameters computed at event creation; the timestamp (stored as the opcode name) is not included here and
-            # is checked separately from these parameters
+            # common parameters computed at event creation; the timestamp (stored as the opcode name) is not included
+            # here and is checked separately from these parameters
             'GAVersion': CURRENT_AGENT,
             'ContainerId': AgentGlobals.get_container_id(),
             'EventTid': threading.current_thread().ident,
@@ -66,7 +68,7 @@ def setUp(self):
             'VMName': 'MachineRole_IN_0',
             'TenantName': 'db00a7755a5e4e8a8fe4b19bc3b330c3',
             'RoleName': 'MachineRole',
-            'RoleInstanceName': 'MachineRole_IN_0',
+            'RoleInstanceName': 'b61f93d0-e1ed-40b2-b067-22c243233448.MachineRole_IN_0',
             # common parameters
             'Location': EventLoggerTools.mock_imds_data['location'],
             'SubscriptionId': EventLoggerTools.mock_imds_data['subscriptionId'],

tests/data/wire/goal_state.xml

@@ -12,15 +12,15 @@
      <ContainerId>c6d5526c-5ac2-4200-b6e2-56f2b70c5ab2</ContainerId>
      <RoleInstanceList>
        <RoleInstance>
-         <InstanceId>MachineRole_IN_0</InstanceId>
+         <InstanceId>b61f93d0-e1ed-40b2-b067-22c243233448.MachineRole_IN_0</InstanceId>
          <State>Started</State>
          <Configuration>
          <HostingEnvironmentConfig>http://168.63.129.16:80/hostingenvuri/</HostingEnvironmentConfig>
          <SharedConfig>http://168.63.129.16:80/sharedconfiguri/</SharedConfig>
          <Certificates>http://168.63.129.16:80/certificatesuri/</Certificates>
          <ExtensionsConfig>http://168.63.129.16:80/extensionsconfiguri/</ExtensionsConfig>
          <FullConfig>http://168.63.129.16:80/fullconfiguri/</FullConfig>
-         <ConfigName>DummyRoleConfigName.xml</ConfigName>
+         <ConfigName>b61f93d0-e1ed-40b2-b067-22c243233448.1.b61f93d0-e1ed-40b2-b067-22c243233448.2.MachineRole_IN_0.xml</ConfigName>
          </Configuration>
        </RoleInstance>
      </RoleInstanceList>

tests/data/wire/goal_state_no_ext.xml

@@ -12,13 +12,14 @@
      <ContainerId>c6d5526c-5ac2-4200-b6e2-56f2b70c5ab2</ContainerId>
      <RoleInstanceList>
        <RoleInstance>
-         <InstanceId>MachineRole_IN_0</InstanceId>
+         <InstanceId>b61f93d0-e1ed-40b2-b067-22c243233448.MachineRole_IN_0</InstanceId>
          <State>Started</State>
          <Configuration>
          <HostingEnvironmentConfig>http://168.63.129.16:80/hostingenvuri/</HostingEnvironmentConfig>
          <SharedConfig>http://168.63.129.16:80/sharedconfiguri/</SharedConfig>
          <Certificates>http://168.63.129.16:80/certificatesuri/</Certificates>
          <FullConfig>http://168.63.129.16:80/fullconfiguri/</FullConfig>
+         <ConfigName>b61f93d0-e1ed-40b2-b067-22c243233448.1.b61f93d0-e1ed-40b2-b067-22c243233448.2.MachineRole_IN_0.xml</ConfigName>
          </Configuration>
        </RoleInstance>
      </RoleInstanceList>

tests/data/wire/goal_state_remote_access.xml

@@ -14,15 +14,15 @@
      </RemoteAccessInfo>
      <RoleInstanceList>
        <RoleInstance>
-         <InstanceId>MachineRole_IN_0</InstanceId>
+         <InstanceId>b61f93d0-e1ed-40b2-b067-22c243233448.MachineRole_IN_0</InstanceId>
          <State>Started</State>
          <Configuration>
          <HostingEnvironmentConfig>http://168.63.129.16:80/hostingenvuri/</HostingEnvironmentConfig>
          <SharedConfig>http://168.63.129.16:80/sharedconfiguri/</SharedConfig>
          <Certificates>http://168.63.129.16:80/certificatesuri/</Certificates>
          <ExtensionsConfig>http://168.63.129.16:80/extensionsconfiguri/</ExtensionsConfig>
          <FullConfig>http://168.63.129.16:80/fullconfiguri/</FullConfig>
-         <ConfigName>DummyRoleConfigName.xml</ConfigName>
+         <ConfigName>b61f93d0-e1ed-40b2-b067-22c243233448.1.b61f93d0-e1ed-40b2-b067-22c243233448.2.MachineRole_IN_0.xml</ConfigName>
          </Configuration>
        </RoleInstance>
      </RoleInstanceList>

tests/ga/test_monitor.py

@@ -256,7 +256,7 @@ def test_collect_and_send_events(self, mock_lib_dir, patch_send_event, *_):
                              '<Param Name="VMName" Value="MachineRole_IN_0" T="mt:wstr" />' \
                              '<Param Name="TenantName" Value="db00a7755a5e4e8a8fe4b19bc3b330c3" T="mt:wstr" />' \
                              '<Param Name="RoleName" Value="MachineRole" T="mt:wstr" />' \
-                             '<Param Name="RoleInstanceName" Value="MachineRole_IN_0" T="mt:wstr" />' \
+                             '<Param Name="RoleInstanceName" Value="b61f93d0-e1ed-40b2-b067-22c243233448.MachineRole_IN_0" T="mt:wstr" />' \
                              '<Param Name="Location" Value="uswest" T="mt:wstr" />' \
                              '<Param Name="SubscriptionId" Value="AAAAAAAA-BBBB-CCCC-DDDD-EEEEEEEEEEEE" T="mt:wstr" />' \
                              '<Param Name="ResourceGroupName" Value="test-rg" T="mt:wstr" />' \

tests/protocol/mocks.py

@@ -99,6 +99,8 @@ def http_request(method, url, data, **kwargs):
                 return protocol.mock_wire_data.mock_http_get(url, **kwargs)
             if method == 'POST':
                 return protocol.mock_wire_data.mock_http_post(url, data, **kwargs)
+            if method == 'PUT':
+                return protocol.mock_wire_data.mock_http_put(url, data, **kwargs)
 
         # the request was not handled; fail or call the original resutil.http_request
         if fail_on_unknown_request:
@@ -201,6 +203,11 @@ def is_host_plugin_in_vm_artifacts_profile_request(url, request_kwargs):
         artifact_location = HttpRequestPredicates._get_host_plugin_request_artifact_location(url, request_kwargs)
         return HttpRequestPredicates.is_in_vm_artifacts_profile_request(artifact_location)
 
+    @staticmethod
+    def is_host_plugin_put_logs_request(url):
+        return url.lower() == 'http://{0}:{1}/vmagentlog'.format(restutil.KNOWN_WIRESERVER_IP,
+                                                                 restutil.HOST_PLUGIN_PORT)
+
 
 class MockHttpResponse:
     def __init__(self, status, body=''):

tests/protocol/mockwiredata.py

@@ -99,6 +99,7 @@ def __init__(self, data_files=DATA_FILE):
             "/versions": 0,
             "/health": 0,
             "/HealthService": 0,
+            "/vmAgentLog": 0,
             "goalstate": 0,
             "hostingenvuri": 0,
             "sharedconfiguri": 0,
@@ -244,6 +245,21 @@ def mock_http_post(self, url, *args, **kwargs):
         resp.read = Mock(return_value=content.encode("utf-8"))
         return resp
 
+    def mock_http_put(self, url, *args, **kwargs):
+        content = None
+
+        resp = MagicMock()
+        resp.status = httpclient.OK
+
+        if url.endswith('/vmAgentLog'):
+            self.call_counts['/vmAgentLog'] += 1
+            content = ''
+        else:
+            raise Exception("Bad url {0}".format(url))
+
+        resp.read = Mock(return_value=content.encode("utf-8"))
+        return resp
+
     def mock_crypt_util(self, *args, **kw):
         #Partially patch instance method of class CryptUtil
         cryptutil = CryptUtil(*args, **kw)