|
1
|
|
|
""" |
|
2
|
|
|
Enarksh |
|
3
|
|
|
|
|
4
|
|
|
Copyright 2013-2016 Set Based IT Consultancy |
|
5
|
|
|
|
|
6
|
|
|
Licence MIT |
|
7
|
|
|
""" |
|
8
|
|
|
import os |
|
9
|
|
|
import signal |
|
10
|
|
|
import sys |
|
11
|
|
|
import select |
|
12
|
|
|
import pwd |
|
13
|
|
|
import zmq |
|
14
|
|
|
import enarksh |
|
15
|
|
|
from enarksh import SPAWNER_PULL_END_POINT, CONTROLLER_PULL_END_POINT, LOGGER_PULL_END_POINT |
|
16
|
|
|
from enarksh.Spawner.JobHandler import JobHandler |
|
17
|
|
|
|
|
18
|
|
|
|
|
19
|
|
|
class Spawner: |
|
|
|
|
|
|
20
|
|
|
_instance = None |
|
21
|
|
|
|
|
22
|
|
|
# ------------------------------------------------------------------------------------------------------------------ |
|
23
|
|
|
def __init__(self): |
|
24
|
|
|
Spawner._instance = self |
|
25
|
|
|
|
|
26
|
|
|
self._child_flag = False |
|
27
|
|
|
self._zmq_context = None |
|
28
|
|
|
self._zmq_pull_socket = None |
|
29
|
|
|
self._zmq_controller = None |
|
30
|
|
|
self._zmq_logger = None |
|
31
|
|
|
|
|
32
|
|
|
self._job_handlers = {} |
|
33
|
|
|
|
|
34
|
|
|
# ------------------------------------------------------------------------------------------------------------------ |
|
35
|
|
|
def _zmq_init(self): |
|
|
|
|
|
|
36
|
|
|
self._zmq_context = zmq.Context() |
|
37
|
|
|
|
|
38
|
|
|
# Create socket for asynchronous incoming messages. |
|
39
|
|
|
self._zmq_pull_socket = self._zmq_context.socket(zmq.PULL) |
|
|
|
|
|
|
40
|
|
|
self._zmq_pull_socket.bind(SPAWNER_PULL_END_POINT) |
|
41
|
|
|
|
|
42
|
|
|
# Create socket for sending asynchronous messages to the controller. |
|
43
|
|
|
self._zmq_controller = self._zmq_context.socket(zmq.PUSH) |
|
|
|
|
|
|
44
|
|
|
self._zmq_controller.connect(CONTROLLER_PULL_END_POINT) |
|
45
|
|
|
|
|
46
|
|
|
# Create socket for sending asynchronous messages to the logger. |
|
47
|
|
|
self._zmq_logger = self._zmq_context.socket(zmq.PUSH) |
|
|
|
|
|
|
48
|
|
|
self._zmq_logger.connect(LOGGER_PULL_END_POINT) |
|
49
|
|
|
|
|
50
|
|
|
# ------------------------------------------------------------------------------------------------------------------ |
|
51
|
|
|
@staticmethod |
|
52
|
|
|
def daemonize(): |
|
|
|
|
|
|
53
|
|
|
enarksh.daemonize(enarksh.HOME + '/var/lock/spawnerd.pid', |
|
54
|
|
|
'/dev/null', |
|
55
|
|
|
enarksh.HOME + '/var/log/spawnerd.log', |
|
56
|
|
|
enarksh.HOME + '/var/log/spawnerd.log') |
|
57
|
|
|
|
|
58
|
|
|
# ------------------------------------------------------------------------------------------------------------------ |
|
59
|
|
|
@staticmethod |
|
60
|
|
|
def sigchld_handler(signum, frame): |
|
|
|
|
|
|
61
|
|
|
""" |
|
62
|
|
|
Static method for SIGCHLD. Set a flag that a child has exited. |
|
63
|
|
|
""" |
|
64
|
|
|
Spawner._instance._child_flag = True |
|
|
|
|
|
|
65
|
|
|
|
|
66
|
|
|
# ------------------------------------------------------------------------------------------------------------------ |
|
67
|
|
|
@staticmethod |
|
68
|
|
|
def sighup_handler(signum, frame): |
|
|
|
|
|
|
69
|
|
|
""" |
|
70
|
|
|
SIGHUP is send by the log rotate daemon. CLoses the current log file and create a new log file. |
|
71
|
|
|
""" |
|
72
|
|
|
sys.stdout.flush() |
|
73
|
|
|
sys.stderr.flush() |
|
74
|
|
|
|
|
75
|
|
|
with open(enarksh.HOME + '/var/log/spawnerd.log', 'wb', 0) as f: |
|
|
|
|
|
|
76
|
|
|
os.dup2(f.fileno(), sys.stdout.fileno()) |
|
77
|
|
|
|
|
78
|
|
|
with open(enarksh.HOME + '/var/log/spawnerd.log', 'wb', 0) as f: |
|
|
|
|
|
|
79
|
|
|
os.dup2(f.fileno(), sys.stderr.fileno()) |
|
80
|
|
|
|
|
81
|
|
|
# ------------------------------------------------------------------------------------------------------------------ |
|
82
|
|
|
def _install_signal_handlers(self): |
|
83
|
|
|
""" |
|
84
|
|
|
Install signal handlers for SIGCHLD and SIGHUP. |
|
85
|
|
|
""" |
|
86
|
|
|
# Install signal handler for child has exited. |
|
87
|
|
|
signal.signal(signal.SIGCHLD, self.sigchld_handler) |
|
88
|
|
|
|
|
89
|
|
|
# Install signal handler log rotate. |
|
90
|
|
|
signal.signal(signal.SIGHUP, self.sighup_handler) |
|
91
|
|
|
|
|
92
|
|
|
# Unfortunately, restart system calls doesnt work. |
|
93
|
|
|
# signal.siginterrupt(signal.SIGCHLD, False) |
|
94
|
|
|
# signal.siginterrupt(signal.SIGHUP, False) |
|
95
|
|
|
|
|
96
|
|
|
# ------------------------------------------------------------------------------------------------------------------ |
|
97
|
|
|
@staticmethod |
|
98
|
|
|
def _set_unprivileged_user(): |
|
99
|
|
|
""" |
|
100
|
|
|
Set the real and effective user and group to an unprivileged user. |
|
101
|
|
|
""" |
|
102
|
|
|
_, _, uid, gid, _, _, _ = pwd.getpwnam('enarksh') |
|
103
|
|
|
|
|
104
|
|
|
os.setresgid(gid, gid, 0) |
|
105
|
|
|
os.setresuid(uid, uid, 0) |
|
106
|
|
|
|
|
107
|
|
|
# ------------------------------------------------------------------------------------------------------------------ |
|
108
|
|
|
def _startup(self): |
|
109
|
|
|
""" |
|
110
|
|
|
Performs the necessary actions for starting the spawner daemon. |
|
111
|
|
|
""" |
|
112
|
|
|
# Set the effective user and group to an unprivileged user and group. |
|
113
|
|
|
self._set_unprivileged_user() |
|
114
|
|
|
|
|
115
|
|
|
# Become a daemon. |
|
116
|
|
|
# self.__daemonize() |
|
117
|
|
|
|
|
118
|
|
|
# Install signal handlers. |
|
119
|
|
|
self._install_signal_handlers() |
|
120
|
|
|
|
|
121
|
|
|
self._zmq_init() |
|
122
|
|
|
|
|
123
|
|
|
# Read all user names under which the controller is allowed to start jobs. |
|
124
|
|
|
JobHandler.read_allowed_users() |
|
125
|
|
|
|
|
126
|
|
|
# ------------------------------------------------------------------------------------------------------------------ |
|
127
|
|
|
def _handle_child_exits(self): |
|
128
|
|
|
""" |
|
129
|
|
|
Handles an exit of a child and ends a jon handler. |
|
130
|
|
|
""" |
|
131
|
|
|
try: |
|
132
|
|
|
pid = -1 |
|
133
|
|
|
while pid != 0: |
|
134
|
|
|
pid, status = os.waitpid(-1, os.WNOHANG + os.WUNTRACED + os.WCONTINUED) |
|
135
|
|
|
if pid != 0: |
|
136
|
|
|
job_handler = self._job_handlers[pid] |
|
137
|
|
|
|
|
138
|
|
|
# Send message to controller that a job has finished. |
|
139
|
|
|
message = {'type': 'node_stop', |
|
140
|
|
|
'sch_id': job_handler.get_sch_id(), |
|
141
|
|
|
'rnd_id': job_handler.get_rnd_id(), |
|
142
|
|
|
'exit_status': status} |
|
143
|
|
|
self._zmq_controller.send_json(message) |
|
144
|
|
|
|
|
145
|
|
|
# Inform the job handler the job has finished. |
|
146
|
|
|
job_handler.set_job_has_finished() |
|
147
|
|
|
|
|
148
|
|
|
except OSError: |
|
|
|
|
|
|
149
|
|
|
# Ignore OSError. No more children to wait for. |
|
150
|
|
|
pass |
|
151
|
|
|
|
|
152
|
|
|
# ------------------------------------------------------------------------------------------------------------------ |
|
153
|
|
|
def _event_handler_start_node(self, sch_id, rnd_id, user_name, args): |
|
154
|
|
|
""" |
|
155
|
|
|
Creates a new job handler and starts the job. |
|
156
|
|
|
|
|
157
|
|
|
:param sch_id: The ID of the schedule of the job. |
|
158
|
|
|
:param rnd_id: The ID of the job. |
|
159
|
|
|
:param user_name: The user under which the job must run. |
|
160
|
|
|
:param args: The arguments for the job. |
|
161
|
|
|
""" |
|
162
|
|
|
job_handler = JobHandler(sch_id, rnd_id, user_name, args) |
|
163
|
|
|
job_handler.start_job() |
|
164
|
|
|
|
|
165
|
|
|
self._job_handlers[job_handler.pid] = job_handler |
|
166
|
|
|
|
|
167
|
|
|
# ------------------------------------------------------------------------------------------------------------------ |
|
168
|
|
|
def _read_message(self): |
|
169
|
|
|
""" |
|
170
|
|
|
Reads a message from the controller. |
|
171
|
|
|
""" |
|
172
|
|
|
try: |
|
173
|
|
|
while True: |
|
174
|
|
|
message = self._zmq_pull_socket.recv_json(zmq.NOBLOCK) |
|
|
|
|
|
|
175
|
|
|
|
|
176
|
|
|
if message['type'] == 'start_node': |
|
177
|
|
|
self._event_handler_start_node(message['sch_id'], |
|
178
|
|
|
message['rnd_id'], |
|
179
|
|
|
message['user_name'], |
|
180
|
|
|
message['args']) |
|
181
|
|
|
|
|
182
|
|
|
else: |
|
183
|
|
|
raise IndexError("Unknown message type '{0!s}'.".format(message['type'])) |
|
184
|
|
|
|
|
185
|
|
|
except zmq.ZMQError as e: |
|
|
|
|
|
|
186
|
|
|
# Ignore ZMQError with EAGAIN. Otherwise, re-raise the error. |
|
187
|
|
|
if e.errno != zmq.EAGAIN: |
|
|
|
|
|
|
188
|
|
|
raise e |
|
189
|
|
|
|
|
190
|
|
|
# ------------------------------------------------------------------------------------------------------------------ |
|
191
|
|
|
def main(self): |
|
192
|
|
|
""" |
|
193
|
|
|
The main function of the job spawner. |
|
194
|
|
|
""" |
|
195
|
|
|
# Perform the necessary actions for starting up the spawner. |
|
196
|
|
|
self._startup() |
|
197
|
|
|
|
|
198
|
|
|
while True: |
|
199
|
|
|
# List with all file descriptors for reading. |
|
200
|
|
|
read = [] |
|
201
|
|
|
|
|
202
|
|
|
# Add the queue for incoming messages to the list of read file descriptors. |
|
203
|
|
|
zmq_fd = self._zmq_pull_socket.get(zmq.FD) |
|
|
|
|
|
|
204
|
|
|
read.append(zmq_fd) |
|
205
|
|
|
|
|
206
|
|
|
# Add the job handlers to the list of read file descriptors. |
|
207
|
|
|
remove = [] |
|
208
|
|
|
for pid, job_handler in self._job_handlers.items(): |
|
209
|
|
|
fd_stdout = job_handler.get_stdout() |
|
210
|
|
|
if fd_stdout >= 0: |
|
211
|
|
|
read.append(fd_stdout) |
|
212
|
|
|
fd_stderr = job_handler.get_stderr() |
|
213
|
|
|
if fd_stderr >= 0: |
|
214
|
|
|
read.append(fd_stderr) |
|
215
|
|
|
|
|
216
|
|
|
if fd_stdout == -1 and fd_stderr == -1 and job_handler.get_pid() == -1: |
|
217
|
|
|
# The job handler has read all data from stdout and stderr from job and the child process has |
|
218
|
|
|
# exited. |
|
219
|
|
|
remove.append(pid) |
|
220
|
|
|
|
|
221
|
|
|
# Remove jobs that are finished. |
|
222
|
|
|
for pid in remove: |
|
223
|
|
|
job_handler = self._job_handlers[pid] |
|
224
|
|
|
|
|
225
|
|
|
# Tell the job handler we are done with the job. |
|
226
|
|
|
job_handler.end_job() |
|
227
|
|
|
|
|
228
|
|
|
# Send messages to logger daemon that the stdout and stderr of the job can be loaded into |
|
229
|
|
|
# the database. |
|
230
|
|
|
self._zmq_logger.send_json(job_handler.get_logger_message('out')) |
|
231
|
|
|
self._zmq_logger.send_json(job_handler.get_logger_message('err')) |
|
232
|
|
|
|
|
233
|
|
|
# Remove the job from the dictionary with jobs. |
|
234
|
|
|
del self._job_handlers[pid] |
|
235
|
|
|
|
|
236
|
|
|
# Unblock interrupts. |
|
237
|
|
|
signal.pthread_sigmask(signal.SIG_UNBLOCK, {signal.SIGCHLD, signal.SIGHUP}) |
|
238
|
|
|
|
|
239
|
|
|
try: |
|
240
|
|
|
# Wait for a fd becomes available for read or wait for an interrupt. |
|
241
|
|
|
read, _, _ = select.select(read, [], []) |
|
242
|
|
|
|
|
243
|
|
|
except InterruptedError: |
|
|
|
|
|
|
244
|
|
|
# Ignore Interrupted system call errors (EINTR) in the select call. |
|
245
|
|
|
pass |
|
246
|
|
|
|
|
247
|
|
|
# Block all interrupts to prevent interrupted system call errors (EINTR). |
|
248
|
|
|
signal.pthread_sigmask(signal.SIG_BLOCK, {signal.SIGCHLD, signal.SIGHUP}) |
|
249
|
|
|
|
|
250
|
|
|
for fd in read: |
|
251
|
|
|
if fd == zmq_fd: |
|
252
|
|
|
# fd of the message queue is ready to receive data. |
|
253
|
|
|
self._read_message() |
|
254
|
|
|
else: |
|
255
|
|
|
# fd of one or more job handlers are ready to receive data. |
|
256
|
|
|
for _, job_handler in self._job_handlers.items(): |
|
257
|
|
|
if fd == job_handler.get_stdout(): |
|
258
|
|
|
job_handler.read(fd) |
|
259
|
|
|
if fd == job_handler.get_stderr(): |
|
260
|
|
|
job_handler.read(fd) |
|
261
|
|
|
|
|
262
|
|
|
if self._child_flag: |
|
263
|
|
|
# Process one or more exited child processes. |
|
264
|
|
|
self._handle_child_exits() |
|
265
|
|
|
|
|
266
|
|
|
|
|
267
|
|
|
# ---------------------------------------------------------------------------------------------------------------------- |
|
268
|
|
|
|
The coding style of this project requires that you add a docstring to this code element. Below, you find an example for methods:
If you would like to know more about docstrings, we recommend to read PEP-257: Docstring Conventions.