feat: 全面修改 ansible 执行方式为 receptor (#12975)

* feat: 修复 receptor kill job  的问题

* feat: 全面修改 ansible 执行方式为 receptor

---------

Co-authored-by: Aaron3S <chenyang@fit2cloud.com>

apps/ops/ansible/callback.py

@@ -48,9 +48,6 @@ class DefaultCallback:
         event = data.get('event', None)
         if not event:
             return
-        # pid = data.get('pid', None)
-        # if pid:
-        #     self.write_pid(pid)
         event_data = data.get('event_data', {})
         host = event_data.get('remote_addr', '')
         task = event_data.get('task', '')
@@ -158,11 +155,4 @@ class DefaultCallback:
     def status_handler(self, data, **kwargs):
         status = data.get('status', '')
         self.status = self.STATUS_MAPPER.get(status, 'unknown')
-
-        rc = kwargs.get('runner_config', None)
-        # self.private_data_dir = rc.private_data_dir if rc else '/tmp/'
-
-    # def write_pid(self, pid):
-    #     pid_filepath = os.path.join(self.private_data_dir, 'local.pid')
-    #     with open(pid_filepath, 'w') as f:
-    #         f.write(str(pid))
+        self.private_data_dir = data.get("private_data_dir", None)

apps/ops/ansible/receptor/receptor_runner.py

@@ -1,16 +1,27 @@
 import concurrent.futures
+import os
 import queue
 import socket
-
+from django.conf import settings
 import ansible_runner
+from django.utils.functional import LazyObject
 from receptorctl import ReceptorControl
 
-receptor_ctl = ReceptorControl('control.sock')
+
+class WarpedReceptorctl(LazyObject):
+    def _setup(self):
+        self._wrapped = self.get_receptorctl()
+
+    @staticmethod
+    def get_receptorctl():
+        return ReceptorControl(settings.ANSIBLE_RECEPTOR_SOCK_PATH)
 
 
-def init_receptor_ctl(sock_path):
-    global receptor_ctl
-    receptor_ctl = ReceptorControl(sock_path)
+receptor_ctl = WarpedReceptorctl()
+
+
+def cancel(unit_id):
+    return receptor_ctl.simple_command("work cancel {}".format(unit_id))
 
 
 def nodes():
@@ -27,6 +38,14 @@ class AnsibleReceptorRunner:
         self.runner_params = kwargs
         self.unit_id = None
 
+    def write_unit_id(self):
+        if not self.unit_id:
+            return
+        private_dir = self.runner_params.get("private_data_dir", "")
+        with open(os.path.join(private_dir, "local.unitid"), "w") as f:
+            f.write(self.unit_id)
+            f.flush()
+
     def run(self):
         input, output = socket.socketpair()
 
@@ -38,6 +57,7 @@ class AnsibleReceptorRunner:
             output.close()
 
             self.unit_id = result['unitid']
+            self.write_unit_id()
 
         transmitter_future.result()
 
@@ -69,20 +89,29 @@ class AnsibleReceptorRunner:
 
     def processor(self, _result_file, stdout_queue):
         try:
-            original_handler = self.runner_params.pop("event_handler", None)
+            original_event_handler = self.runner_params.pop("event_handler", None)
+            original_status_handler = self.runner_params.pop("status_handler", None)
 
             def event_handler(data, **kwargs):
                 stdout = data.get('stdout', '')
                 if stdout:
                     stdout_queue.put(stdout)
-                if original_handler:
-                    original_handler(data, **kwargs)
+                if original_event_handler:
+                    original_event_handler(data, **kwargs)
+
+            def status_handler(data, **kwargs):
+                private_data_dir = self.runner_params.get("private_data_dir", None)
+                if private_data_dir:
+                    data["private_data_dir"] = private_data_dir
+                if original_status_handler:
+                    original_status_handler(data, **kwargs)
 
             return ansible_runner.interface.run(
                 quite=True,
                 streamer='process',
                 _input=_result_file,
                 event_handler=event_handler,
+                status_handler=status_handler,
                 **self.runner_params,
             )
         finally:

apps/ops/ansible/runner.py

@@ -26,8 +26,6 @@ class AnsibleWrappedRunner(LazyObject):
     @staticmethod
     def get_runner():
         if settings.ANSIBLE_RECEPTOR_ENABLE and settings.ANSIBLE_RECEPTOR_SOCK_PATH:
-            logger.info("Ansible receptor enabled, run ansible task via receptor")
-            receptor_runner.init_receptor_ctl(settings.ANSIBLE_RECEPTOR_SOCK_PATH)
             return receptor_runner
         return ansible_runner
 

apps/ops/models/job.py

@@ -23,6 +23,7 @@ from assets.models import Asset
 from assets.automations.base.manager import SSHTunnelManager
 from common.db.encoder import ModelJSONFieldEncoder
 from ops.ansible import JMSInventory, AdHocRunner, PlaybookRunner, CommandInBlackListException, UploadFileRunner
+from ops.ansible.receptor import receptor_runner
 from ops.mixin import PeriodTaskModelMixin
 from ops.variables import *
 from ops.const import Types, RunasPolicies, JobStatus, JobModules
@@ -521,15 +522,16 @@ class JobExecution(JMSOrgBaseModel):
             ssh_tunnel.local_gateway_clean(runner)
 
     def stop(self):
-        from ops.signal_handlers import job_execution_stop_pub_sub
 
-        with open(os.path.join(self.private_dir, 'local.pid')) as f:
-            try:
-                pid = f.read()
-                job_execution_stop_pub_sub.publish(int(pid))
-            except Exception as e:
-                print(e)
-        self.set_error('Job stop by "kill -9 {}"'.format(pid))
+        unit_id_path = os.path.join(self.private_dir, "local.unitid")
+        if os.path.exists(unit_id_path):
+            with open(unit_id_path) as f:
+                try:
+                    unit_id = f.read()
+                    receptor_runner.cancel(unit_id)
+                except Exception as e:
+                    print(e)
+            self.set_error('Job stop by "receptor worker cancel, unit id is {}"'.format(unit_id))
 
     class Meta:
         verbose_name = _("Job Execution")