3 # Copyright 2010 Peter Palfrader
5 # Permission is hereby granted, free of charge, to any person obtaining
6 # a copy of this software and associated documentation files (the
7 # "Software"), to deal in the Software without restriction, including
8 # without limitation the rights to use, copy, modify, merge, publish,
9 # distribute, sublicense, and/or sell copies of the Software, and to
10 # permit persons to whom the Software is furnished to do so, subject to
11 # the following conditions:
13 # The above copyright notice and this permission notice shall be
14 # included in all copies or substantial portions of the Software.
16 # THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
17 # EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
18 # MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
19 # NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
20 # LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
21 # OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
22 # WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
24 # Checks if the WAL backups for several postgres clusters from
25 # different hosts are current. Might not catch all error instances.
27 # If called with -e will expire WALs and BASE backups no longer required.
29 # Needs files layed out like so:
30 # beethoven:/srv/pgbackup/pg# ls -l ries/ | head
32 # -rw------- 1 debbackup debbackup 378099591 May 1 02:26 dak.BASE.20100501-ries-dak-8.4-backup-F_37000020.tar.gz
33 # -rw------- 1 debbackup debbackup 382267407 May 8 02:25 dak.BASE.20100508-ries-dak-8.4-backup-F_A2000020.tar.gz
34 # -rw------- 1 debbackup debbackup 384054069 May 15 02:24 dak.BASE.20100515-ries-dak-8.4-backup-F_FB000020.tar.gz
35 # -rw------- 1 debbackup debbackup 386407500 May 22 02:27 dak.BASE.20100522-ries-dak-8.4-backup-10_58000020.tar.gz
36 # -rw------- 1 debbackup debbackup 16777216 May 1 02:26 dak.WAL.000000010000000F00000037
37 # -rw------- 1 debbackup debbackup 264 May 1 02:27 dak.WAL.000000010000000F00000037.00000020.backup
38 # -rw------- 1 debbackup debbackup 16777216 May 1 03:25 dak.WAL.000000010000000F00000038
39 # -rw------- 1 debbackup debbackup 16777216 May 1 09:11 dak.WAL.000000010000000F00000039
40 # -rw------- 1 debbackup debbackup 16777216 May 1 09:45 dak.WAL.000000010000000F0000003A
43 # needs write privileges to at least the .backup files
57 elif 'DSA_CHECK_BACKUPPG_CONF' in os.environ:
58 configfile = os.environ['DSA_CHECK_BACKUPPG_CONF']
60 configfile = '/etc/nagios/dsa-check-backuppg.conf'
63 config = yaml.load(f.read())
72 #def note_warning(key, host, db, value):
74 # problems_seq.append("[%s, %s]: %s: %s"%(host, db, key, value))
76 # global problems_per_db
77 # if not host in problems_per_db: problems_per_db[host] = {}
78 # problems_per_db[host][db] = True
79 def note_info(key, value, pre=None):
82 notices_seq.append("%s: %s"%(key, value))
84 notices_seq.append("[%s] %s: %s"%(pre, key, value))
86 def note_warning(key, value, pre=None):
89 problems_seq.append("%s: %s"%(key, value))
91 problems_seq.append("[%s] %s: %s"%(pre, key, value))
92 def note_warning_db(host, db, key, value):
93 note_warning(key, value, "%s, %s"%(host, db))
94 global problems_per_db
95 if not host in problems_per_db: problems_per_db[host] = {}
96 problems_per_db[host][db] = True
109 def parse_pg_backup_info(fn):
113 (k,v) = l.strip().split(': ', 2)
119 parser = optparse.OptionParser()
120 parser.set_usage("%prog [-c=<CONFFILE>] (nagios mode)\n" +
121 "Usage: %prog [-c=<CONFFILE>] -e [-d] [-v] (expire mode)")
122 parser.add_option("-c", "--config", dest="conffile", metavar="CONFFILE",
123 help="Config file location.")
124 parser.add_option("-e", "--expire", dest="expire", action="store_true",
125 help="Expire old files.")
126 parser.add_option("-d", "--dry-run", dest="dry_run", action="store_true",
127 help="Do not really remove files.")
128 parser.add_option("-v", "--verbose", dest="verbose", action="store_true",
129 help="List files we are expiring.")
130 (options, args) = parser.parse_args()
137 config = load_conf(options.conffile)
139 os.chdir(config['rootdir'])
140 for dir in os.listdir('.'):
141 if dir.startswith('.') or dir.endswith('.old'):
142 note_info('IGNORED', dir)
145 if not os.path.isdir(dir):
147 mtime = os.path.getmtime(dir)
148 ctime = os.path.getctime(dir)
150 if e.errno == errno.ENOENT:
154 if min(mtime, ctime) + 3600*4 > time.time():
155 note_info('IGNORED', dir)
157 note_warning('NOT-A-DIR', dir)
160 if not dir in config['backups']:
161 note_warning('NOT-CONFIGURED', dir)
164 files = os.listdir(dir)
166 note_warning('EMPTY-DIR', dir)
171 unhandled_backups = copy.copy(config['backups'][dir])
175 # Go over all the files in a directory and check for various things
176 # - for a given cluster's backups we want the latest WAL file to be no
177 # older than a certain age,
178 # - we want all consecutive WAL files, i.e. no holes
179 # - we want a full backup at one point, and it shouldn't be too old
180 # - If our retention period is say 2 weeks, then we look for the
181 # tar file that's older than that, and everything before that can
183 while len(files) > 0:
185 ffn = os.path.join(dir, fn)
187 r = re.match('([a-z0-9-]+)\.(WAL|BASE)\..*', fn)
189 note_warning('CANNOT-PARSE', ffn)
192 (db, type) = r.groups(1)
193 if not isinstance(config['backups'][dir], list) or not db in config['backups'][dir]:
194 if not db in ignored_dbs:
195 note_warning_db(dir, db, 'NOT-CONFIGURED', '%s/%s'%(dir, db))
196 ignored_dbs[db] = True
197 if db in ignored_dbs:
199 if not db in backup_state:
200 backup_state[db] = {}
201 # can_expire_for_base_hit: We hit a BASE backup that is old enough
202 # so that once we hit all the required WAL files for this base
203 # backup to work we can start expiring everything older than that
205 backup_state[db]['can_expire_for_base_hit'] = False
206 # can_expire_next: Can expire all files that we handle from now on
207 backup_state[db]['can_expire_next'] = False
208 backup_state[db]['expires'] = []
210 # Apparently we already have seen a base backup and all its wal files
211 # which we want to keep, so everything what we see now is older than
212 # that and we can get rid of it
213 if backup_state[db]['can_expire_next']:
214 backup_state[db]['expires'].append(ffn)
217 # should have been taken care of before
218 # while handling a WAL.backup file
219 note_warning_db(dir, db, 'STRAY-BASE', ffn)
222 # handle .backup files - they live near the WAL "file namespace" and reference
223 # the corresponding full backup
224 r = re.match('[a-z0-9-]+\.WAL\.([0-9A-F]{8})([0-9A-F]{8})([0-9A-F]{8})\.[0-9A-F]{8}\.backup', fn)
226 info = parse_pg_backup_info(ffn)
227 basefn = '%s.BASE.%s-%s.tar.gz'%(db, info['label'], info['start wal location'].split(' ',2)[0].replace('/', '_'))
228 baseffn = os.path.join(dir, basefn)
229 if not basefn in files:
230 basefn = '%s.BASE.%s.tar.gz'%(db, info['label'])
231 baseffn = os.path.join(dir, basefn)
232 if not basefn in files:
233 note_warning_db(dir, db, 'MISSING-BASE', basefn)
236 if backup_state[db]['can_expire_next']:
237 backup_state[db]['expires'].append(baseffn)
239 if not 'newest-base' in backup_state[db]:
240 backup_state[db]['newest-base'] = baseffn
241 backup_state[db]['oldest-base'] = baseffn
243 startre = re.search('\(file ([0-9A-F]{24})\)', info['start wal location'])
245 note_warning_db(dir, db, 'CANNOT-PARSE-START_WAL_LOCATION', ffn)
247 start_file = startre.group(1)
248 walbase = '%s.WAL.%s'%(db, start_file)
249 backup_state[db]['base_needs_wal_until'] = walbase
251 start = time.mktime(time.strptime(info['start time'], '%Y-%m-%d %H:%M:%S %Z'))
252 if start + config['retention'] < time.time():
253 backup_state[db]['can_expire_for_base_hit'] = True
257 r = re.match('[a-z0-9-]+\.WAL\.([0-9A-F]{8})([0-9A-F]{8})([0-9A-F]{8})', fn)
259 if 'base_needs_wal_until' in backup_state[db]:
260 if backup_state[db]['base_needs_wal_until'] == fn:
261 del backup_state[db]['base_needs_wal_until']
262 if backup_state[db]['can_expire_for_base_hit']:
263 backup_state[db]['can_expire_next'] = True
265 (timeline, wal1, wal2) = map(lambda x: int(x,16), r.groups())
266 if not timeline == 1:
267 note_warning_db(dir, db, 'CANNOT-HANDLE-TIMELINES_NOT_1', ffn)
270 thissegment = (wal1, wal2)
271 if not 'newest-wal' in backup_state[db]:
272 backup_state[db]['newest-wal'] = thissegment
273 backup_state[db]['newest-wal-file'] = ffn
275 if not wal_pre(backup_state[db]['oldest-wal']) == thissegment:
276 note_warning_db(dir, db, 'WAL-MISSING-AFTER', ffn)
277 ignored_dbs[db] = True
279 backup_state[db]['oldest-wal'] = thissegment
283 note_warning_db(dir, db, 'CANNOT-PARSE-WAL', ffn)
285 note_warning_db(dir, db, 'INVALID-TYPE', ffn)
288 for db in backup_state:
289 if 'base_needs_wal_until' in backup_state[db]:
290 note_warning_db(dir, db, 'MISSING_WAL_FOR_BASE', backup_state[db]['base_needs_wal_until'])
292 for db in backup_state:
293 if not 'newest-base' in backup_state[db]:
294 note_warning_db(dir, db, 'NO-BASE', 'no base backup found?')
296 age = time.time() - os.stat(backup_state[db]['newest-base']).st_mtime
297 if age > config['warn-age']['base']:
298 note_warning_db(dir, db, 'BASE-IS-OLD', 'latest base backup is too old')
300 if not 'newest-wal-file' in backup_state[db]:
301 note_warning_db(dir, db, 'NO-BASE', 'no WAL files found?')
303 age = time.time() - os.stat(backup_state[db]['newest-wal-file']).st_mtime
304 if age > config['warn-age']['wal']:
305 note_warning_db(dir, db, 'WAL-IS-OLD', 'latest wal file is too old')
307 for db in backup_state:
308 if len(backup_state[db]['expires']) > 0:
309 if dir in problems_per_db and db in problems_per_db[dir] and problems_per_db[dir][db]:
310 note_warning_db(dir, db, 'NOT-EXPIRING-DUE-TO-WARNINGS', 'have seen warnings, will not expire anything')
312 backup_state[db]['expires'].reverse()
313 for f in backup_state[db]['expires']:
314 global_expires.append(f)
316 #if not db in backup_state:
317 # note_warning('BASE-WITHOUT-WAL', ffn)
318 # ignored_dbs[db] = True
321 #age = time.time() - os.stat(ffn).st_mtime
322 #if age > config['warn-age']['wal']:
323 # note_warning('OLD-WAL', backup_state[db]['newest-wal-file'])
324 # ignored_dbs[db] = True
328 for p in problems_seq:
330 for p in notices_seq:
334 for f in global_expires:
335 if options.verbose: print "Expiring %s"%(f)
336 if not options.dry_run: os.unlink(f)
338 if len(problems_seq) > 0:
341 if not options.expire or options.verbose:
342 print "OK: no problems detected"
347 # vim:set shiftwidth=4: