1 | #!/bin/sh |
---|
2 | |
---|
3 | # Licensed to the Apache Software Foundation (ASF) under one or more |
---|
4 | # contributor license agreements. See the NOTICE file distributed with |
---|
5 | # this work for additional information regarding copyright ownership. |
---|
6 | # The ASF licenses this file to You under the Apache License, Version 2.0 |
---|
7 | # (the "License"); you may not use this file except in compliance with |
---|
8 | # the License. You may obtain a copy of the License at |
---|
9 | # |
---|
10 | # http://www.apache.org/licenses/LICENSE-2.0 |
---|
11 | # |
---|
12 | # Unless required by applicable law or agreed to in writing, software |
---|
13 | # distributed under the License is distributed on an "AS IS" BASIS, |
---|
14 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
---|
15 | # See the License for the specific language governing permissions and |
---|
16 | # limitations under the License. |
---|
17 | |
---|
18 | |
---|
19 | """:" |
---|
20 | work_dir=$(dirname $0) |
---|
21 | base_name=$(basename $0) |
---|
22 | cd $work_dir |
---|
23 | |
---|
24 | if [ $HOD_PYTHON_HOME ]; then |
---|
25 | exec $HOD_PYTHON_HOME -OO $base_name ${1+"$@"} |
---|
26 | elif [ -e /usr/bin/python ]; then |
---|
27 | exec /usr/bin/python -OO $base_name ${1+"$@"} |
---|
28 | elif [ -e /usr/local/bin/python ]; then |
---|
29 | exec /usr/local/bin/python -OO $base_name ${1+"$@"} |
---|
30 | else |
---|
31 | exec python -OO $base_name ${1+"$@"} |
---|
32 | fi |
---|
33 | ":""" |
---|
34 | |
---|
35 | """The executable to be used by the user""" |
---|
36 | import sys, os, re, getpass |
---|
37 | |
---|
38 | myName = os.path.basename(sys.argv[0]) |
---|
39 | myName = re.sub(".*/", "", myName) |
---|
40 | binDirectory = os.path.realpath(sys.argv[0]) |
---|
41 | rootDirectory = re.sub("/bin/.*", "", binDirectory) |
---|
42 | libDirectory = rootDirectory |
---|
43 | |
---|
44 | sys.path.append(libDirectory) |
---|
45 | |
---|
46 | from hodlib.RingMaster.ringMaster import main |
---|
47 | from hodlib.Common.setup import * |
---|
48 | from hodlib.Common.descGenerator import * |
---|
49 | from hodlib.Common.util import local_fqdn, filter_warnings, to_http_url, \ |
---|
50 | get_exception_string, get_exception_error_string |
---|
51 | from hodlib.Common.logger import getLogger, ensureLogDir |
---|
52 | from hodlib.Common.xmlrpc import hodXRClient |
---|
53 | import logging |
---|
54 | |
---|
55 | filter_warnings() |
---|
56 | |
---|
57 | reVersion = re.compile(".*(\d+_\d+).*") |
---|
58 | |
---|
59 | VERSION = '$HeadURL$' |
---|
60 | |
---|
61 | reMatch = reVersion.match(VERSION) |
---|
62 | if reMatch: |
---|
63 | VERSION = reMatch.group(1) |
---|
64 | VERSION = re.sub("_", ".", VERSION) |
---|
65 | else: |
---|
66 | VERSION = 'DEV' |
---|
67 | |
---|
68 | # Definition tuple is of the form: |
---|
69 | # (name, type, description, default value, required?, validate?) |
---|
70 | # |
---|
71 | defList = { 'ringmaster' : ( |
---|
72 | ('work-dirs', 'list', 'hod work directories', |
---|
73 | False, None, True, False), |
---|
74 | |
---|
75 | ('temp-dir', 'directory', 'Ringmaster temporary directory.', |
---|
76 | False, None, True, False), |
---|
77 | |
---|
78 | ('log-dir', 'directory', 'hod logging directory.', |
---|
79 | False, os.path.join(rootDirectory, 'logs'), False, True), |
---|
80 | |
---|
81 | ('syslog-address', 'address', 'Syslog address.', |
---|
82 | False, None, False, True), |
---|
83 | |
---|
84 | ('xrs-port-range', 'range', 'XML-RPC port range n-m.', |
---|
85 | False, None, True, True), |
---|
86 | |
---|
87 | ('http-port-range', 'range', 'HTTP port range n-m.', |
---|
88 | False, None, True, True), |
---|
89 | |
---|
90 | ('debug', 'pos_int', 'Debugging level, 0-4.', |
---|
91 | False, 3, True, True), |
---|
92 | |
---|
93 | ('register', 'bool', 'Register with service registry?', |
---|
94 | False, True, True, True), |
---|
95 | |
---|
96 | ('stream', 'bool', 'Output to stderr.', |
---|
97 | False, False, False, True), |
---|
98 | |
---|
99 | ('userid', 'user_account', |
---|
100 | 'User ID the hod shell is running under.', |
---|
101 | False, None, True, False), |
---|
102 | |
---|
103 | ('svcrgy-addr', 'address', 'Download HTTP address.', |
---|
104 | False, None, False, True), |
---|
105 | |
---|
106 | ('hadoop-tar-ball', 'uri', 'hadoop program tar ball.', |
---|
107 | False, None, False, False), |
---|
108 | |
---|
109 | ('max-connect','pos_int','max connections allowed for a single tarball server', |
---|
110 | False, 30, False, True), |
---|
111 | |
---|
112 | ('jt-poll-interval', 'pos_int', 'How often to poll the Job tracker for idleness', |
---|
113 | False, 120, False, True), |
---|
114 | |
---|
115 | ('idleness-limit', 'pos_int', 'Limit after which to deallocate the cluster', |
---|
116 | False, 3600, False, True), |
---|
117 | |
---|
118 | ('max-master-failures', 'pos_int', |
---|
119 | 'Defines how many times a master can fail before' \ |
---|
120 | ' failing cluster allocation', False, 5, True, True), |
---|
121 | |
---|
122 | ('workers_per_ring', 'pos_int', 'Defines number of workers per service per hodring', |
---|
123 | False, 1, False, True)), |
---|
124 | |
---|
125 | 'resource_manager' : ( |
---|
126 | ('id', 'string', 'Batch scheduler ID: torque|condor.', |
---|
127 | False, None, True, True), |
---|
128 | |
---|
129 | ('pbs-user', 'user_account', 'User ID jobs are submitted under.', |
---|
130 | False, None, False, True), |
---|
131 | |
---|
132 | ('pbs-server', 'hostname', 'Hostname of PBS server.', |
---|
133 | False, None, False, True), |
---|
134 | |
---|
135 | ('pbs-account', 'string', 'User Account jobs are submitted under.', |
---|
136 | False, None, False, False), |
---|
137 | |
---|
138 | ('queue', 'string', 'Queue of the batch scheduler to query.', |
---|
139 | False, None, False, False), |
---|
140 | |
---|
141 | ('batch-home', 'directory', 'Scheduler installation directory.', |
---|
142 | False, None, True, True), |
---|
143 | |
---|
144 | ('options', 'keyval', 'Options to pass to the scheduler.', |
---|
145 | False, None, False, True), |
---|
146 | |
---|
147 | ('env-vars', 'keyval', 'Environment variables to pass to the submitted jobs.', |
---|
148 | False, None, False, True)), |
---|
149 | |
---|
150 | 'gridservice-mapred' : ( |
---|
151 | ('external', 'bool', "Connect to an already running MapRed?", |
---|
152 | False, False, True, True), |
---|
153 | |
---|
154 | ('host', 'hostname', 'Mapred hostname.', |
---|
155 | False, 'localhost', False, True), |
---|
156 | |
---|
157 | ('info_port', 'pos_int', 'Mapred info port.', |
---|
158 | False, None, True, True), |
---|
159 | |
---|
160 | ('tracker_port', 'pos_int', 'Mapred job tracker port.', |
---|
161 | False, None, True, True), |
---|
162 | |
---|
163 | ('cmdline-params', 'keyval', 'Hadoop cmdline key/value list.', |
---|
164 | False, None, False, False), |
---|
165 | |
---|
166 | ('server-params', 'keyval', 'Hadoop xml key/value list', |
---|
167 | False, None, False, False), |
---|
168 | |
---|
169 | ('final-server-params', 'keyval', 'Hadoop final xml params', |
---|
170 | False, None, False, False), |
---|
171 | |
---|
172 | ('envs', 'keyval', 'environment to run this package in', |
---|
173 | False, None, False, False), |
---|
174 | |
---|
175 | ('pkgs', 'directory', "directory where the package is installed", |
---|
176 | False, None, False, False)), |
---|
177 | |
---|
178 | |
---|
179 | 'gridservice-hdfs' : ( |
---|
180 | ('external', 'bool', "Connect to an already running HDFS?", |
---|
181 | False, False, True, True), |
---|
182 | |
---|
183 | ('host', 'hostname', 'HDFS hostname.', |
---|
184 | False, 'localhost', True, True), |
---|
185 | |
---|
186 | ('fs_port', 'pos_int', 'HDFS port range.', |
---|
187 | False, None, True, True), |
---|
188 | |
---|
189 | ('info_port', 'pos_int', 'HDFS info port.', |
---|
190 | False, None, True, True), |
---|
191 | |
---|
192 | ('cmdline-params', 'keyval', 'Hadoop cmdline key/value list.', |
---|
193 | False, None, False, False), |
---|
194 | |
---|
195 | ('server-params', 'keyval', 'Hadoop xml key/value list', |
---|
196 | False, None, False, False), |
---|
197 | |
---|
198 | ('final-server-params', 'keyval', 'Hadoop final xml params', |
---|
199 | False, None, False, False), |
---|
200 | |
---|
201 | ('envs', 'keyval', 'Environment in which to run this package.', |
---|
202 | False, None, False, False), |
---|
203 | |
---|
204 | ('pkgs', 'directory', "directory where the package is installed", |
---|
205 | False, None, False, False)), |
---|
206 | |
---|
207 | |
---|
208 | 'hodring' : ( |
---|
209 | ('temp-dir', 'directory', 'hod work directories', |
---|
210 | False, None, True, False), |
---|
211 | |
---|
212 | ('log-dir', 'directory', 'hod logging directory.', |
---|
213 | False, os.path.join(rootDirectory, 'logs'), False, False), |
---|
214 | |
---|
215 | ('log-destination-uri', 'string', |
---|
216 | 'URI to store logs to, local://some_path or ' |
---|
217 | + 'hdfs://host:port/some_path', |
---|
218 | False, None, False, True), |
---|
219 | |
---|
220 | ('pkgs', 'directory', 'Path to Hadoop to use in case of uploading to HDFS', |
---|
221 | False, None, False, True), |
---|
222 | |
---|
223 | ('syslog-address', 'address', 'Syslog address.', |
---|
224 | False, None, False, True), |
---|
225 | |
---|
226 | ('java-home', 'directory', 'Java home directory.', |
---|
227 | False, None, True, False), |
---|
228 | |
---|
229 | ('debug', 'pos_int', 'Debugging level, 0-4.', |
---|
230 | False, 3, True, True), |
---|
231 | |
---|
232 | ('register', 'bool', 'Register with service registry?', |
---|
233 | False, True, True, True), |
---|
234 | |
---|
235 | ('stream', 'bool', 'Output to stderr.', |
---|
236 | False, False, False, True), |
---|
237 | |
---|
238 | ('userid', 'user_account', |
---|
239 | 'User ID the hod shell is running under.', |
---|
240 | False, None, True, False), |
---|
241 | |
---|
242 | ('xrs-port-range', 'range', 'XML-RPC port range n-m.', |
---|
243 | False, None, True, True), |
---|
244 | |
---|
245 | ('http-port-range', 'range', 'HTTP port range n-m.', |
---|
246 | False, None, True, True), |
---|
247 | |
---|
248 | ('command', 'string', 'Command for hodring to run.', |
---|
249 | False, None, False, True), |
---|
250 | |
---|
251 | ('service-id', 'string', 'Service ID.', |
---|
252 | False, None, False, True), |
---|
253 | |
---|
254 | ('download-addr', 'address', 'Download HTTP address.', |
---|
255 | False, None, False, True), |
---|
256 | |
---|
257 | ('svcrgy-addr', 'address', 'Download HTTP address.', |
---|
258 | False, None, False, True), |
---|
259 | |
---|
260 | ('ringmaster-xrs-addr', 'address', 'Ringmaster XML-RPC address.', |
---|
261 | False, None, False, True), |
---|
262 | |
---|
263 | ('tarball-retry-initial-time', 'pos_float','initial retry time for tarball download', |
---|
264 | False, 1, False, True), |
---|
265 | |
---|
266 | ('tarball-retry-interval', 'pos_float','interval to spread retries for tarball download', |
---|
267 | False, 3, False, True), |
---|
268 | |
---|
269 | ('cmd-retry-initial-time', 'pos_float','initial retry time for getting commands', |
---|
270 | False, 2, False, True), |
---|
271 | |
---|
272 | ('cmd-retry-interval', 'pos_float','interval to spread retries for getting commands', |
---|
273 | False, 2, False, True), |
---|
274 | |
---|
275 | ('mapred-system-dir-root', 'string', 'Root under which mapreduce system directory names are generated by HOD.', |
---|
276 | False, '/mapredsystem', False, False)) |
---|
277 | } |
---|
278 | |
---|
279 | |
---|
280 | defOrder = [ 'ringmaster', 'hodring', 'resource_manager', |
---|
281 | 'gridservice-mapred', 'gridservice-hdfs' ] |
---|
282 | |
---|
283 | if __name__ == '__main__': |
---|
284 | confDef = definition() |
---|
285 | confDef.add_defs(defList, defOrder) |
---|
286 | ringMasterOptions = options(confDef, "./%s [OPTIONS]" % myName, VERSION) |
---|
287 | log = logging.getLogger() |
---|
288 | |
---|
289 | try: |
---|
290 | |
---|
291 | # Set up logging before anything else. |
---|
292 | ensureLogDir(ringMasterOptions.normalizeValue('ringmaster', 'log-dir')) |
---|
293 | log = getLogger(ringMasterOptions['ringmaster'],'ringmaster') |
---|
294 | # End of setting up logging |
---|
295 | |
---|
296 | # Verify and process options |
---|
297 | statusMsgs = [] |
---|
298 | # Conditional validation |
---|
299 | if not ringMasterOptions['ringmaster'].has_key('hadoop-tar-ball') or \ |
---|
300 | not ringMasterOptions['ringmaster']['hadoop-tar-ball']: |
---|
301 | # If tarball is not used |
---|
302 | if not ringMasterOptions.normalizeValue('gridservice-hdfs', 'external'): |
---|
303 | # And if hdfs is not external, validate gridservice-hdfs.pkgs |
---|
304 | statusMsgs.extend(ringMasterOptions.validateValue( |
---|
305 | 'gridservice-hdfs', 'pkgs')) |
---|
306 | statusMsgs.extend(ringMasterOptions.validateValue( |
---|
307 | 'gridservice-mapred', 'pkgs')) |
---|
308 | |
---|
309 | if len(statusMsgs) != 0: |
---|
310 | # format status messages into a single string |
---|
311 | errStr = '' |
---|
312 | for msg in statusMsgs: |
---|
313 | errStr = "%s%s\n" % (errStr, msg) |
---|
314 | raise Exception("%s" % errStr) |
---|
315 | # End of conditional validation |
---|
316 | |
---|
317 | (status, statusMsgs) = ringMasterOptions.verify() |
---|
318 | if not status: |
---|
319 | # format status messages into a single string |
---|
320 | errStr = '' |
---|
321 | for msg in statusMsgs: |
---|
322 | errStr = "%s%s\n" % (errStr, msg) |
---|
323 | raise Exception("%s" % errStr) |
---|
324 | |
---|
325 | ringMasterOptions.replace_escape_seqs() |
---|
326 | ringMasterOptions['ringmaster']['base-dir'] = rootDirectory |
---|
327 | # End of option processing |
---|
328 | |
---|
329 | ret = main(ringMasterOptions,log) |
---|
330 | sys.exit(ret) |
---|
331 | except Exception, e: |
---|
332 | log.error("bin/ringmaster failed to start.%s. Stack trace follows:\n%s" % (get_exception_error_string(),get_exception_string())) |
---|
333 | |
---|
334 | # Report errors to the client if possible |
---|
335 | try: |
---|
336 | serviceAddr = to_http_url(ringMasterOptions.normalizeValue( \ |
---|
337 | 'ringmaster', 'svcrgy-addr')) |
---|
338 | serviceClient = hodXRClient(serviceAddr) |
---|
339 | if serviceClient is not None: |
---|
340 | serviceClient.setRMError([local_fqdn(), str(e), \ |
---|
341 | get_exception_string()]) |
---|
342 | log.info("Reported errors to service registry at %s" % serviceAddr) |
---|
343 | except Exception, e: |
---|
344 | log.error("Failed to report errors to service registry.") |
---|
345 | log.error("Reason : %s" % get_exception_string()) |
---|
346 | # End of reporting errors to the client |
---|
347 | |
---|
348 | # Ringmaster failing to start is a ringmaster error. Exit with the appropriate exit code. |
---|
349 | sys.exit(6) |
---|