Blame view

src/core/pyros_django/agent/AgentSST.py 13.4 KB
05316241   Alexis Koralewski   Adding AgentSST, ...
1
2
3
4
#!/usr/bin/env python3

from pathlib import Path
import subprocess
a04e004d   Alexis Koralewski   Fixing AgentCmd c...
5
6
import sys, os
from datetime import datetime, timezone, timedelta
05316241   Alexis Koralewski   Adding AgentSST, ...
7
8
9
10
11
12
13
##import utils.Logger as L
#import threading #, multiprocessing, os
import time

#from django.db import transaction
#from common.models import Command

dd27c2bc   Alexis Koralewski   Updating agent co...
14
from Agent import Agent, build_agent, log, extract_parameters
05316241   Alexis Koralewski   Adding AgentSST, ...
15
import socket
a04e004d   Alexis Koralewski   Fixing AgentCmd c...
16
from common.models import AgentCmd, AgentSurvey
05316241   Alexis Koralewski   Adding AgentSST, ...
17

dd27c2bc   Alexis Koralewski   Updating agent co...
18
from src.core.pyros_django.obsconfig.obsconfig_class import OBSConfig
05316241   Alexis Koralewski   Adding AgentSST, ...
19
20

class AgentSST(Agent):
04b0b442   Alexis Koralewski   Changing computer...
21
    computer = "XCY1"
05316241   Alexis Koralewski   Adding AgentSST, ...
22
23
24
25
    _previous_dir = ""
    PROJECT_ROOT_PATH = ""
    VENV_PYTHON = ""
    subprocess_dict = {}
7fd15ce5   Alexis Koralewski   Adding test mode ...
26
    agent_in_mode_test = {}
05316241   Alexis Koralewski   Adding AgentSST, ...
27

dd27c2bc   Alexis Koralewski   Updating agent co...
28
    AGENT_SPECIFIC_COMMANDS = [
7668f19b   Alexis Koralewski   Adding commands' ...
29
30
31
        ("do_kill_agent",10),
        ("do_restart_agent",20),
        ("do_start_agent",20),
dd27c2bc   Alexis Koralewski   Updating agent co...
32
    ]
7fd15ce5   Alexis Koralewski   Adding test mode ...
33
34
    TEST_COMMANDS_LIST = [
    ]
05316241   Alexis Koralewski   Adding AgentSST, ...
35

dd27c2bc   Alexis Koralewski   Updating agent co...
36
    def __init__(self, name:str=None,sim_computer=None):
05316241   Alexis Koralewski   Adding AgentSST, ...
37
        
dd27c2bc   Alexis Koralewski   Updating agent co...
38
        super().__init__()
05316241   Alexis Koralewski   Adding AgentSST, ...
39
40
41
        self.PROJECT_ROOT_PATH = os.environ["PROJECT_ROOT_PATH"]
        if name is None:
            name = self.__class__.__name__
ea1ca112   Alexis Koralewski   Fixing agentSST s...
42
        # self.computer = socket.gethostname()        
05316241   Alexis Koralewski   Adding AgentSST, ...
43
44
45
        WITH_DOCKER = False
        if os.environ.get("WITH_DOCKER"):
            WITH_DOCKER = True
99502363   Alexis Koralewski   add message when ...
46
            # if WITH_DOCKER socket.gethostname() bizarre 
05316241   Alexis Koralewski   Adding AgentSST, ...
47
48
49
50
51
52
53
54
55
56
57
58
59
60
        if WITH_DOCKER:
            VENV_ROOT = ""
            VENV = ""
            VENV_BIN = ""
        else:
            VENV_ROOT = "venv"
            VENV = "venv_py3_pyros"
            VENV_BIN = (
                self.PROJECT_ROOT_PATH
                + os.sep + VENV_ROOT
                + os.sep + VENV
                + os.sep + "bin"
                + os.sep
            )
dd27c2bc   Alexis Koralewski   Updating agent co...
61
        self.VENV_PYTHON = VENV_BIN + "python3"
05316241   Alexis Koralewski   Adding AgentSST, ...
62
        log.info(f"PC hostname is {self.computer}")
9bd7ac9e   Alexis Koralewski   Adding timeout co...
63
        #name_from_config = self.get_config().get_agent_sst_of_current_computer()
7fd15ce5   Alexis Koralewski   Adding test mode ...
64
65
66
    
    def init(self):
        super().init()
05316241   Alexis Koralewski   Adding AgentSST, ...
67
        self.start_agents()
7fd15ce5   Alexis Koralewski   Adding test mode ...
68
        self.TEST_MODE = False
ea1ca112   Alexis Koralewski   Fixing agentSST s...
69
70
        time.sleep(10)
        self.set_delay(3)
a3d0b4b0   Alexis Koralewski   AgentSST : adapti...
71

05316241   Alexis Koralewski   Adding AgentSST, ...
72
73
74
75
76
77
78
79
    def start_agents(self,agent_name=None):
        """
        Start all agents or one agent of obs_config

        Args:
            agent_name (_type_, optional): Specific agent name to start. Defaults to None.
        """
        obs_config = self.get_config()
7fd15ce5   Alexis Koralewski   Adding test mode ...
80
81
        test_mode = " -t"

05316241   Alexis Koralewski   Adding AgentSST, ...
82
83
84
85
86
87
88
89
90
91
        if agent_name:
            agent = agent_name
            # Start a specific agent of obs_config (restart)
            agent_informations = obs_config.get_agent_information(obs_config.unit_name,agent)
            protocol = agent_informations.get("protocol")
            if protocol:
                protocol_folder_abs_path = os.path.join(self.PROJECT_ROOT_PATH, os.path.dirname(protocol))
                
                protocol_script_name = protocol.split("/")[-1]
                if os.path.exists(protocol_folder_abs_path + os.sep + protocol_script_name):
dd27c2bc   Alexis Koralewski   Updating agent co...
92
                    cmd = self.VENV_PYTHON +" "+ protocol_folder_abs_path + os.sep + protocol_script_name
7fd15ce5   Alexis Koralewski   Adding test mode ...
93
94
95
96
                    if not agent in self.agent_in_mode_test:
                        self.agent_in_mode_test = self.TEST_MODE
                    if self.agent_in_mode_test[agent]:
                        cmd += test_mode
9bd7ac9e   Alexis Koralewski   Adding timeout co...
97
                    process = subprocess.Popen(f"{cmd}",shell=True)
8d9c8345   Alexis Koralewski   Renaming commands...
98
                    process.poll()
9bd7ac9e   Alexis Koralewski   Adding timeout co...
99
100
                    
                    self.subprocess_dict[agent]["process"] = process
d0760264   Alexis Koralewski   Fixing cmd test f...
101
102
103
                    nb_try_restart = self.subprocess_dict[agent].get("nb_try_restart",0)
                    nb_try_restart += 1
                    self.subprocess_dict[agent]["nb_try_restart"] = nb_try_restart
05316241   Alexis Koralewski   Adding AgentSST, ...
104
105
106
                    log.info(f"Start agent {agent} with cmd {cmd}")

        else:
027eaa78   Alexis Koralewski   Fixing AgentSST i...
107
108
109
110
111
112
113
114
115
            agents = obs_config.get_agents_per_computer(obs_config.unit_name).get(self.computer)
            if agents is None:
                available_hostnames = obs_config.get_agents_per_computer(obs_config.unit_name).keys()
                log.info("Computer not found in obs config")
                log.info(f"Available hostnames {available_hostnames}. Current hostname is {self.computer}")
                exit(1)
            #self.change_dir(self.PROJECT_ROOT_PATH)
            else:
                log.info(f"Agents associated to this computer : {agents}")
05316241   Alexis Koralewski   Adding AgentSST, ...
116
117
118
119
120
121
122
123
124
            # Start every agent of obs_config (initial start)
            for agent in agents:
                agent_informations = obs_config.get_agent_information(obs_config.unit_name,agent)
                protocol = agent_informations.get("protocol")
                if protocol:
                    protocol_folder_abs_path = os.path.join(self.PROJECT_ROOT_PATH, os.path.dirname(protocol))
                    
                    protocol_script_name = protocol.split("/")[-1]
                    if os.path.exists(protocol_folder_abs_path + os.sep + protocol_script_name):
05316241   Alexis Koralewski   Adding AgentSST, ...
125
                        
7fd15ce5   Alexis Koralewski   Adding test mode ...
126
127
128
129
130
                        cmd = self.VENV_PYTHON +" "+ protocol_folder_abs_path + os.sep + protocol_script_name
                        if not agent in self.agent_in_mode_test:
                            self.agent_in_mode_test[agent] = self.TEST_MODE
                        if self.agent_in_mode_test[agent]:
                            cmd += test_mode
a04e004d   Alexis Koralewski   Fixing AgentCmd c...
131
132
                        # process = subprocess.Popen(f"{cmd}", shell=True, stdout=subprocess.DEVNULL,stderr=subprocess.STDOUT)
                        process = subprocess.Popen(f"{cmd}", shell=True)
9bd7ac9e   Alexis Koralewski   Adding timeout co...
133
134
                        self.subprocess_dict[agent] = {}
                        self.subprocess_dict[agent]["process"] = process
3449489e   Alexis Koralewski   Fixing current_nb...
135
136
                        # Reset to zero nb_try when AgentSST start (launch all agents)
                        self.subprocess_dict[agent]["nb_try_restart"] = 0
05316241   Alexis Koralewski   Adding AgentSST, ...
137
138
                        log.info(f"Start agent {agent} with cmd {cmd}")

2918ab51   Alexis Koralewski   Adding do_start_a...
139
    def do_start_agent(self, agent_name:str):
05316241   Alexis Koralewski   Adding AgentSST, ...
140
141
142
143
144
145
146
        """
        Start a specific agent of obs_config (Restart)

        Args:
            agent_name (str): Name of agent to start
        """
        self.start_agents(agent_name)
3449489e   Alexis Koralewski   Fixing current_nb...
147
148
149
150
151
        nb_try_restart_agent = self.subprocess_dict[agent_name]["nb_try_restart"]
        agent_survey = AgentSurvey.objects.get(name=agent_name)
        agent_survey.current_nb_restart = nb_try_restart_agent
        
        agent_survey.save()
05316241   Alexis Koralewski   Adding AgentSST, ...
152

dd27c2bc   Alexis Koralewski   Updating agent co...
153

afbc5c95   Alexis Koralewski   Renaming AgentSST...
154
155
    def do_kill_agent(self, agent)->None:
        # agent = args[0]
dd27c2bc   Alexis Koralewski   Updating agent co...
156
        if agent in self.subprocess_dict.keys():
a3d0b4b0   Alexis Koralewski   AgentSST : adapti...
157
158
            cmd = self.send_cmd_to(agent,"do_exit")
            #cmd = self.send_cmd_to(agent,"do_exit","asap")
a04e004d   Alexis Koralewski   Fixing AgentCmd c...
159
            return cmd
dd27c2bc   Alexis Koralewski   Updating agent co...
160

afbc5c95   Alexis Koralewski   Renaming AgentSST...
161
162
    def do_restart_agent(self, agent)->None:
        # agent = args[0]
9bd7ac9e   Alexis Koralewski   Adding timeout co...
163
        nb_try_restart_agent = self.subprocess_dict[agent]["nb_try_restart"]
c957b9e1   Alexis Koralewski   fixing restart of...
164
        if nb_try_restart_agent < AgentSurvey.objects.get(name=agent).nb_restart_max:
9bd7ac9e   Alexis Koralewski   Adding timeout co...
165
166
167
168
169
            self.do_kill_agent(agent)
            self.do_start_agent(agent)
        else:
            #sendmail
            pass
a3d0b4b0   Alexis Koralewski   AgentSST : adapti...
170
171
172
        agent_survey = AgentSurvey.objects.get(name=agent)
        agent_survey.current_nb_restart = nb_try_restart_agent
        agent_survey.save()
9bd7ac9e   Alexis Koralewski   Adding timeout co...
173

a04e004d   Alexis Koralewski   Fixing AgentCmd c...
174
175
176
177
178
179
        # if agent in self.subprocess_dict.keys():
        #     cmd.set_result(f"Agent {agent} restarted")
        #     cmd.set_as_processed()
        # else:
        #     cmd.set_result(f"Agent {agent} failed to restart")
        #     log.debug(f"Agent {agent} failed to restart")
dd27c2bc   Alexis Koralewski   Updating agent co...
180

a04e004d   Alexis Koralewski   Fixing AgentCmd c...
181
    def force_kill_agent(self, *args)->None:
d4ebe565   Alexis Koralewski   Adding pyros stop...
182
183
184
        if args:
            agent = args[0]
            if self.subprocess_dict.get(agent) is not None:
9bd7ac9e   Alexis Koralewski   Adding timeout co...
185
                process = self.subprocess_dict.get(agent).get("process")
d4ebe565   Alexis Koralewski   Adding pyros stop...
186
187
188
189
190
191
                # process.terminate()
                # process.wait()
                # Kill is better when using Popen(shell=True) because it will remove the created child process
                process.kill()
            else:
                return None
a04e004d   Alexis Koralewski   Fixing AgentCmd c...
192
193
194
195

    def do_things_before_exit(self,abort_cmd_sender):
        kill_agent_commands = {}
        for agent in self.subprocess_dict.keys():
8d9c8345   Alexis Koralewski   Renaming commands...
196
            cmd = self.do_kill_agent(agent)    
a04e004d   Alexis Koralewski   Fixing AgentCmd c...
197
            kill_agent_commands[agent] = cmd
3449489e   Alexis Koralewski   Fixing current_nb...
198
199
200
            agent_survey = AgentSurvey.objects.get(name=agent)
            # Reset counter before exiting
            agent_survey.current_nb_restart = 0
a3d0b4b0   Alexis Koralewski   AgentSST : adapti...
201
            agent_survey.save()                             
d4ebe565   Alexis Koralewski   Adding pyros stop...
202
203
        # wait 10 seconds in order to agents to exit themselves properly 
        time.sleep(10)
ea1ca112   Alexis Koralewski   Fixing agentSST s...
204
205
206
        for agent in self.subprocess_dict.keys():
            while self.subprocess_dict[agent].get("process").poll() is None:
                time.sleep(0.5)
a04e004d   Alexis Koralewski   Fixing AgentCmd c...
207

2ac0a02f   Alexis Koralewski   Fixing AgentSST (...
208
    def routine_process_after_body(self):
a04e004d   Alexis Koralewski   Fixing AgentCmd c...
209
        now_time = datetime.now(timezone.utc) 
027eaa78   Alexis Koralewski   Fixing AgentSST i...
210
        last_running_commands = AgentCmd.get_commands_sent_by_agent("AgentSST").filter(state="CMD_RUNNING",recipient__in=list(self.subprocess_dict.keys()))
a04e004d   Alexis Koralewski   Fixing AgentCmd c...
211
        for cmd in last_running_commands:
fd880010   Alexis Koralewski   Fixing error in A...
212
            last_running_cmd = cmd.full_name
a04e004d   Alexis Koralewski   Fixing AgentCmd c...
213
            if last_running_cmd == "KILL_AGENT" and cmd.is_expired():
fd880010   Alexis Koralewski   Fixing error in A...
214
                agent = cmd.args[0]
2918ab51   Alexis Koralewski   Adding do_start_a...
215
                self.force_kill_agent(agent)
a04e004d   Alexis Koralewski   Fixing AgentCmd c...
216

027eaa78   Alexis Koralewski   Fixing AgentSST i...
217
        # checking status of agent if they are timeout
a04e004d   Alexis Koralewski   Fixing AgentCmd c...
218
        for agent in self.subprocess_dict.keys():
fd880010   Alexis Koralewski   Fixing error in A...
219
220
            try:
                agent_survey = AgentSurvey.objects.get(name=agent)
027eaa78   Alexis Koralewski   Fixing AgentSST i...
221
222
223
224
225
226
227
228
229
230
231
            except AgentSurvey.DoesNotExist:
                # If there is no entry in AgentSurvey for this agent go to next iteration (it surely means that the agentSST launched this agent for the first time, and it didn't had enough time to create an entry in AgentSurvey)
                continue
            
            validity_duration = agent_survey.validity_duration
            last_update_from_agent = agent_survey.updated
            validity_duration_timedelta = timedelta(seconds=validity_duration)
            timeout_datetime = last_update_from_agent + validity_duration_timedelta
            timeout_datetime = timeout_datetime.replace(tzinfo=timezone.utc)
            # if agent latest state is timeout, restart it
            if timeout_datetime < now_time:
9bd7ac9e   Alexis Koralewski   Adding timeout co...
232
                if self.subprocess_dict[agent].get("process").poll() != None:
027eaa78   Alexis Koralewski   Fixing AgentSST i...
233
234
                    last_executed_start_agent_cmd =  AgentCmd.objects.filter(state="CMD_EXECUTED",full_name=f"do_start_agent {agent}",recipient=self.name).order_by("-s_deposit_time")
                    if last_executed_start_agent_cmd.exists():
7fd15ce5   Alexis Koralewski   Adding test mode ...
235
236
237
238
239
                        cmd_outdated_datetime_start = datetime.utcnow() - timedelta(seconds=30)
                        cmd_outdated_datetime_end = datetime.utcnow() - timedelta(seconds=25)
                        cmd_outdated_datetime_start = cmd_outdated_datetime_start.replace(tzinfo=timezone.utc)
                        cmd_outdated_datetime_end = cmd_outdated_datetime_end.replace(tzinfo=timezone.utc)
                        # cmd outdated if deposit time was between 25 and 30 seconds ago from now
027eaa78   Alexis Koralewski   Fixing AgentSST i...
240
                        # if last start cmd for this agent was executed and this agent isn't currently running, ask again a start.
3449489e   Alexis Koralewski   Fixing current_nb...
241
                        if cmd_outdated_datetime_start >= last_executed_start_agent_cmd.first().s_deposit_time and cmd_outdated_datetime_end >= last_executed_start_agent_cmd.first().s_deposit_time:
027eaa78   Alexis Koralewski   Fixing AgentSST i...
242
243
244
245
246
247
                            self.send_cmd_to("AgentSST","do_start_agent", agent)
                    else:
                        try:
                            # Check if do_start_agent cmd already asked by agentSST in previous iterations and not exectuted. If the query success (no exception raised), we don't send again a cmd
                            AgentCmd.get_pending_and_running_commands_for_agent(self.name).get(full_name=f"do_start_agent {agent}")
                        except:
a3d0b4b0   Alexis Koralewski   AgentSST : adapti...
248
                            self.send_cmd_to(self.name,"do_start_agent", agent)
027eaa78   Alexis Koralewski   Fixing AgentSST i...
249
250
251
                else:
                    last_executed_start_or_restart_agent_cmd =  AgentCmd.objects.filter(state="CMD_EXECUTED",full_name__in=(f"do_start_agent {agent}",f"do_restart_agent {agent}"),recipient=self.name).order_by("-s_deposit_time")
                    if last_executed_start_or_restart_agent_cmd.exists():
7fd15ce5   Alexis Koralewski   Adding test mode ...
252
253
254
255
256
                        cmd_outdated_datetime_start = datetime.utcnow() - timedelta(seconds=30)
                        cmd_outdated_datetime_end = datetime.utcnow() - timedelta(seconds=25)
                        cmd_outdated_datetime_start = cmd_outdated_datetime_start.replace(tzinfo=timezone.utc)
                        cmd_outdated_datetime_end = cmd_outdated_datetime_end.replace(tzinfo=timezone.utc)
                        # cmd outdated if deposit time was between 25 and 30 seconds ago from now
027eaa78   Alexis Koralewski   Fixing AgentSST i...
257
                        # if last start or restart cmd for this agent was executed and this agent isn't currently running, ask again a restart.
3449489e   Alexis Koralewski   Fixing current_nb...
258
                        if cmd_outdated_datetime_start >= last_executed_start_or_restart_agent_cmd.first().s_deposit_time and cmd_outdated_datetime_end >=  last_executed_start_or_restart_agent_cmd.first().s_deposit_time:
a3d0b4b0   Alexis Koralewski   AgentSST : adapti...
259
                            self.send_cmd_to(self.name,"do_restart_agent", agent)    
027eaa78   Alexis Koralewski   Fixing AgentSST i...
260
261
262
263
264
                    else:
                        try:
                            # Check if do_restart_agent cmd already asked by agentSST in previous iterations and not exectuted. If the query success (no exception raised), we don't send again a cmd
                            AgentCmd.get_pending_and_running_commands_for_agent(self.name).get(full_name=f"do_restart_agent {agent}")
                        except:
a3d0b4b0   Alexis Koralewski   AgentSST : adapti...
265
                            self.send_cmd_to(self.name,"do_restart_agent", agent)
027eaa78   Alexis Koralewski   Fixing AgentSST i...
266

a04e004d   Alexis Koralewski   Fixing AgentCmd c...
267
268
        log.info("Check status of process")
        for agent in self.subprocess_dict:
9bd7ac9e   Alexis Koralewski   Adding timeout co...
269
            proc = self.subprocess_dict.get(agent).get("process")
dd27c2bc   Alexis Koralewski   Updating agent co...
270
            log.info(f"{agent} poll result is {proc.poll()}")
a04e004d   Alexis Koralewski   Fixing AgentCmd c...
271
     
dd27c2bc   Alexis Koralewski   Updating agent co...
272

05316241   Alexis Koralewski   Adding AgentSST, ...
273
274
275
276
277

if __name__ == "__main__":

    agent = build_agent(AgentSST)
    agent.run()