1 | #!/usr/bin/env python |
---|
2 | from __future__ import with_statement |
---|
3 | |
---|
4 | import glob |
---|
5 | import logging.handlers |
---|
6 | import optparse |
---|
7 | import os |
---|
8 | import socket |
---|
9 | import shutil |
---|
10 | import subprocess |
---|
11 | import sys |
---|
12 | import time |
---|
13 | from os import path |
---|
14 | |
---|
15 | OCF_SUCCESS=0 |
---|
16 | OCF_ERR_GENERIC=1 |
---|
17 | OCF_ERR_ARGS=2 |
---|
18 | OCF_ERR_UNIMPLEMENTED=3 |
---|
19 | OCF_ERR_PERM=4 |
---|
20 | OCF_ERR_INSTALLED=5 |
---|
21 | OCF_ERR_CONFIGURED=6 |
---|
22 | OCF_NOT_RUNNING=7 |
---|
23 | |
---|
24 | logger = logging.getLogger('cron') |
---|
25 | |
---|
26 | import os |
---|
27 | import subprocess |
---|
28 | |
---|
29 | HA_LOGD = os.environ.get('HA_LOGD') == 'xyes' |
---|
30 | |
---|
31 | class HaLogHandler(logging.Handler): |
---|
32 | """ |
---|
33 | A handler class which writes to ha_logger. |
---|
34 | """ |
---|
35 | def __init__(self, ha_tag): |
---|
36 | """ |
---|
37 | Initialize the handler. ha_tag is the name of this resource. |
---|
38 | """ |
---|
39 | logging.Handler.__init__(self) |
---|
40 | self.ha_tag = ha_tag |
---|
41 | |
---|
42 | def emit(self, record): |
---|
43 | """ |
---|
44 | Emit a record. |
---|
45 | """ |
---|
46 | print 'Passed', record |
---|
47 | try: |
---|
48 | levelname = record.levelname |
---|
49 | msg = self.format(record) |
---|
50 | subprocess.call(['/usr/sbin/ha_logger', '-t', self.ha_tag, msg]) |
---|
51 | except (KeyboardInterrupt, SystemExit): |
---|
52 | raise |
---|
53 | except: |
---|
54 | self.handleError(record) |
---|
55 | |
---|
56 | class lock(object): |
---|
57 | def __init__(self, name): |
---|
58 | self.name = name |
---|
59 | |
---|
60 | def __enter__(self): |
---|
61 | while True: |
---|
62 | try: |
---|
63 | self.lock = os.open(self.name, os.O_RDWR | os.O_CREAT | os.O_EXCL) |
---|
64 | except OSError: |
---|
65 | logger.error('Could not acquire lock %s. Sleeping...' % self.name) |
---|
66 | time.sleep(0.5) |
---|
67 | else: |
---|
68 | break |
---|
69 | |
---|
70 | def __exit__(self, type, value, traceback): |
---|
71 | os.close(self.lock) |
---|
72 | _remove(self.name) |
---|
73 | |
---|
74 | def _touch(path): |
---|
75 | """Effectively touches a file. Returns true if successful, false |
---|
76 | otherwise""" |
---|
77 | try: |
---|
78 | open(path, 'a').close() |
---|
79 | except IOError: |
---|
80 | return False |
---|
81 | else: |
---|
82 | return True |
---|
83 | |
---|
84 | def _remove(dest): |
---|
85 | try: |
---|
86 | if path.isdir(dest): |
---|
87 | os.rmdir(dest) |
---|
88 | else: |
---|
89 | os.remove(dest) |
---|
90 | except OSError, e: |
---|
91 | logging.error('Could not remove %s: %s' % (dest, e)) |
---|
92 | return False |
---|
93 | else: |
---|
94 | return True |
---|
95 | |
---|
96 | def _mkdir(dir): |
---|
97 | try: |
---|
98 | os.mkdir(dir) |
---|
99 | except OSError, e: |
---|
100 | logging.error('Could not mkdir %s: %s' % (dir, e)) |
---|
101 | return False |
---|
102 | else: |
---|
103 | return True |
---|
104 | |
---|
105 | def _strip(name): |
---|
106 | """Strip off the file extension, and leading /'s, if they exist""" |
---|
107 | return path.splitext(path.basename(name))[0] |
---|
108 | |
---|
109 | def _suffix(name, suffix): |
---|
110 | return '%s.%s' % (name, suffix) |
---|
111 | |
---|
112 | def _crondir(server): |
---|
113 | return path.join(CRONSPOOL_DIR, _suffix(server, 'cronspool')) |
---|
114 | |
---|
115 | def _server_exists(server): |
---|
116 | return path.exists(path.join(SERVER_DIR, server)) |
---|
117 | |
---|
118 | def _serverfile(server): |
---|
119 | return path.join(SERVER_DIR, server) |
---|
120 | |
---|
121 | def _servers(): |
---|
122 | """Get a list of the servers.""" |
---|
123 | return [_strip(f) for f in glob.glob(path.join(SERVER_DIR, '*'))] |
---|
124 | |
---|
125 | def _is_master(server): |
---|
126 | crondir = path.join(CRONSPOOL_DIR, _suffix(server, 'cronspool')) |
---|
127 | return path.islink(crondir) |
---|
128 | |
---|
129 | def start_cron(args, options): |
---|
130 | if not _touch(_serverfile(HOSTNAME)): |
---|
131 | return OCF_ERR_CONFIGURED |
---|
132 | |
---|
133 | logger.info('Starting %s' % HOSTNAME) |
---|
134 | if _is_master(HOSTNAME): |
---|
135 | logger.error('%s is already the master!' % HOSTNAME) |
---|
136 | for server in _servers(): |
---|
137 | crondir = _crondir(server) |
---|
138 | if server == HOSTNAME: |
---|
139 | _remove(crondir) |
---|
140 | os.symlink('../cronspool', crondir) |
---|
141 | logger.info('Created master symlink %s' % crondir) |
---|
142 | else: |
---|
143 | if path.islink(crondir): |
---|
144 | _remove(crondir) |
---|
145 | logger.info('Removed old master symlink: %s' % crondir) |
---|
146 | if not path.exists(crondir): |
---|
147 | _mkdir(crondir) |
---|
148 | logger.info('Created slave dummy directory %s' % crondir) |
---|
149 | |
---|
150 | if CRON_RESTART_COMMAND: |
---|
151 | ret = subprocess.call(CRON_RESTART_COMMAND) |
---|
152 | if ret: |
---|
153 | logger.error('Cron restart exited with return code %d' % ret) |
---|
154 | return OCF_ERR_GENERIC |
---|
155 | else: |
---|
156 | logger.info('Restarted crond') |
---|
157 | return OCF_SUCCESS |
---|
158 | |
---|
159 | def stop_cron(args, options): |
---|
160 | if not _is_master(HOSTNAME): |
---|
161 | logger.error('I am not the master!') |
---|
162 | else: |
---|
163 | crondir = _crondir(HOSTNAME) |
---|
164 | logger.info('Removing symlink %s' % crondir) |
---|
165 | _remove(crondir) |
---|
166 | _mkdir(crondir) |
---|
167 | return OCF_SUCCESS |
---|
168 | |
---|
169 | def monitor_cron(args, options): |
---|
170 | if _is_master(HOSTNAME): |
---|
171 | return OCF_SUCCESS |
---|
172 | else: |
---|
173 | return OCF_NOT_RUNNING |
---|
174 | |
---|
175 | def validate_all_cron(args, options): |
---|
176 | if not _touch(_serverfile(HOSTNAME)): |
---|
177 | logger.error('Could not touch %s' % _serverfile(HOSTNAME)) |
---|
178 | return OCF_GENERIC_ERR |
---|
179 | if not path.exists(CRONSPOOL_DIR): |
---|
180 | return OCF_GENERIC_ERR |
---|
181 | |
---|
182 | def setup(args, options): |
---|
183 | for d in [CRONSPOOL_DIR, SERVER_DIR]: |
---|
184 | if not path.exists(d): |
---|
185 | os.makedirs(d) |
---|
186 | logger.info('Created %s' % d) |
---|
187 | else: |
---|
188 | logger.info('Already exists: %s' % d) |
---|
189 | |
---|
190 | def add_servers(servers, options): |
---|
191 | for server in servers: |
---|
192 | _touch(_serverfile(server)) |
---|
193 | |
---|
194 | def remove_servers(servers, options): |
---|
195 | for server in servers: |
---|
196 | os.unlink(_serverfile(server)) |
---|
197 | |
---|
198 | def meta_data_cron(args, options): |
---|
199 | print """<?xml version="1.0"?> |
---|
200 | <!DOCTYPE resource-agent SYSTEM "ra-api-1.dtd"> |
---|
201 | <resource-agent name="hacron" version="0.1"> |
---|
202 | <version>1.0</version> |
---|
203 | |
---|
204 | <longdesc lang="en"> |
---|
205 | This is the high-availability cron manager. It uses an extremely overpowered |
---|
206 | clustering solution to make it so that people can have their crontabs. Yay. |
---|
207 | </longdesc> |
---|
208 | <shortdesc lang="en">HA Cron</shortdesc> |
---|
209 | |
---|
210 | <parameters> |
---|
211 | <parameter name="cron_root" required="1"> |
---|
212 | <longdesc lang="en"> |
---|
213 | Base directory for storage of crontabs and server information. |
---|
214 | </longdesc> |
---|
215 | <shortdesc lang="en">Cron base directory</shortdesc> |
---|
216 | <content type="string" /> |
---|
217 | </parameter> |
---|
218 | |
---|
219 | <parameter name="cron_restart_cmd"> |
---|
220 | <longdesc lang="en"> |
---|
221 | Command to restart cron. |
---|
222 | </longdesc> |
---|
223 | <shortdesc lang="en">Restart cron cmd</shortdesc> |
---|
224 | <content type="string" /> |
---|
225 | </parameter> |
---|
226 | </parameters> |
---|
227 | |
---|
228 | <actions> |
---|
229 | <action name="start" timeout="90" /> |
---|
230 | <action name="stop" timeout="100" /> |
---|
231 | <action name="monitor" timeout="20" interval="10" depth="0" start-delay="0" /> |
---|
232 | <action name="reload" timeout="90" /> |
---|
233 | <action name="meta-data" timeout="5" /> |
---|
234 | <action name="validate-all" timeout="30" /> |
---|
235 | </actions> |
---|
236 | </resource-agent> |
---|
237 | """ |
---|
238 | return OCF_SUCCESS |
---|
239 | |
---|
240 | def usage(parser): |
---|
241 | parser.print_help() |
---|
242 | return 1 |
---|
243 | |
---|
244 | def _set_globals(args, options): |
---|
245 | global HOSTNAME, CRONROOT, CRONSPOOL_DIR, SERVER_DIR, CRON_RESTART_COMMAND, \ |
---|
246 | HA_RSCTMP, OCF_RESOURCE_INSTANCE |
---|
247 | if options.development: |
---|
248 | logging.basicConfig(level=logging.DEBUG) |
---|
249 | else: |
---|
250 | if HA_LOGD: |
---|
251 | handler = HaLogHandler('hacron') |
---|
252 | else: |
---|
253 | handler = logging.handlers.SysLogHandler('/dev/log') |
---|
254 | formatter = logging.Formatter("%(module)s: %(levelname)s %(message)s") |
---|
255 | handler.setLevel(logging.INFO) |
---|
256 | handler.setFormatter(formatter) |
---|
257 | logger.addHandler(handler) |
---|
258 | HOSTNAME = options.server or os.environ.get('HA_CURHOST') or socket.gethostname() |
---|
259 | CRONROOT = options.cronroot or os.environ.get('OCF_RESKEY_cron_root') |
---|
260 | if not CRONROOT: |
---|
261 | logging.error('No cron_root specified.') |
---|
262 | return OCF_ERR_CONFIGURED |
---|
263 | CRONSPOOL_DIR = path.join(CRONROOT, 'server-cronspools') |
---|
264 | SERVER_DIR = path.join(CRONROOT, 'servers') |
---|
265 | CRON_RESTART_COMMAND = options.cron_restart or os.environ.get('OCF_RESKEY_cron_restart_cmd') |
---|
266 | |
---|
267 | HA_RSCTMP = os.environ.get('HA_RSCTMP', '/tmp') |
---|
268 | OCF_RESOURCE_INSTANCE = os.environ.get('OCF_RESOURCE_INSTANCE', 'default') |
---|
269 | return OCF_SUCCESS |
---|
270 | |
---|
271 | def main(): |
---|
272 | cmds = ['start', 'reload', 'stop', 'monitor', 'validate-all', 'setup', |
---|
273 | 'remove-servers', 'meta-data'] |
---|
274 | usage_str = "usage: %%prog [%s]" % '|'.join(cmds) |
---|
275 | parser = optparse.OptionParser(usage=usage_str) |
---|
276 | parser.add_option("-s", "--server", |
---|
277 | action="store", dest="server", |
---|
278 | default=None, |
---|
279 | help="choose which server to run script as") |
---|
280 | parser.add_option("-c", "--cronroot", |
---|
281 | action="store", dest="cronroot", |
---|
282 | default=None, |
---|
283 | help="pick root of cron dir") |
---|
284 | parser.add_option("-d", "--development", |
---|
285 | action="store_true", dest="development", |
---|
286 | default=False, |
---|
287 | help="run in production") |
---|
288 | parser.add_option("-r", "--cron-restart", |
---|
289 | action="store", dest="cron_restart", |
---|
290 | default=None, |
---|
291 | help="run in production") |
---|
292 | (options, args) = parser.parse_args() |
---|
293 | if len(args) < 1: |
---|
294 | return usage(parser) |
---|
295 | command = args[0] |
---|
296 | args = args[1:] |
---|
297 | |
---|
298 | if command == 'meta-data': |
---|
299 | return meta_data_cron(args, options) |
---|
300 | globals_status = _set_globals(args, options) |
---|
301 | with lock('%s/hacron-%s.lock' % (HA_RSCTMP, OCF_RESOURCE_INSTANCE)): |
---|
302 | if globals_status: |
---|
303 | return globals_status |
---|
304 | if command == 'start': |
---|
305 | return start_cron(args, options) |
---|
306 | elif command == 'reload': |
---|
307 | return start_cron(args, options) |
---|
308 | elif command == 'stop': |
---|
309 | return stop_cron(args, options) |
---|
310 | elif command == 'monitor': |
---|
311 | return monitor_cron(args, options) |
---|
312 | elif command == 'validate-all': |
---|
313 | return validate_all_cron(args, options) |
---|
314 | elif command == 'setup': |
---|
315 | return setup(args, options) |
---|
316 | elif command == 'add-servers': |
---|
317 | return remove_servers(args, options) |
---|
318 | else: |
---|
319 | usage(parser) |
---|
320 | return OCF_ERR_UNIMPLEMENTED |
---|
321 | |
---|
322 | if __name__ == '__main__': |
---|
323 | try: |
---|
324 | ret = main() |
---|
325 | except Exception, e: |
---|
326 | logger.error('exception from main: %s' % e) |
---|
327 | ret = OCF_ERR_GENERIC |
---|
328 | raise |
---|
329 | sys.exit(ret) |
---|