2 """The database houses information on slivers. This information
3 reaches the sliver manager in two different ways: one, through the
4 GetSlivers() call made periodically; two, by users delivering tickets.
5 The sync() method of the Database class turns this data into reality.
7 The database format is a dictionary that maps account names to records
8 (also known as dictionaries). Inside each record, data supplied or
9 computed locally is stored under keys that begin with an underscore,
10 and data from PLC is stored under keys that don't.
12 In order to maintain service when the node reboots during a network
13 partition, the database is constantly being dumped to disk.
26 # We enforce minimum allocations to keep the clueless from hosing their slivers.
27 # Disallow disk loans because there's currently no way to punish slivers over quota.
28 MINIMUM_ALLOCATION = {'cpu_pct': 0,
36 LOANABLE_RESOURCES = MINIMUM_ALLOCATION.keys()
38 DB_FILE = '/var/lib/nodemanager/database.pickle'
41 # database object and associated lock
42 db_lock = threading.RLock()
45 # these are used in tandem to request a database dump from the dumper daemon
46 db_cond = threading.Condition(db_lock)
47 dump_requested = False
49 # decorator that acquires and releases the database lock before and after the decorated operation
50 # XXX - replace with "with" statements once we switch to 2.5
52 def sync_fn(*args, **kw_args):
54 try: return fn(*args, **kw_args)
55 finally: db_lock.release()
56 sync_fn.__doc__ = fn.__doc__
57 sync_fn.__name__ = fn.__name__
63 self._min_timestamp = 0
65 def _compute_effective_rspecs(self):
66 """Calculate the effects of loans and store the result in field _rspec.
67 At the moment, we allow slivers to loan only those resources that they have received directly from PLC.
68 In order to do the accounting, we store three different rspecs:
69 * field 'rspec', which is the resources given by PLC;
70 * field '_rspec', which is the actual amount of resources the sliver has after all loans;
71 * and variable resid_rspec, which is the amount of resources the sliver
72 has after giving out loans but not receiving any."""
74 for name, rec in self.iteritems():
76 rec['_rspec'] = rec['rspec'].copy()
78 for rec in slivers.itervalues():
79 eff_rspec = rec['_rspec']
80 resid_rspec = rec['rspec'].copy()
81 for target, resource_name, amount in rec.get('_loans', []):
82 if target in slivers and amount <= resid_rspec[resource_name] - MINIMUM_ALLOCATION[resource_name]:
83 eff_rspec[resource_name] -= amount
84 resid_rspec[resource_name] -= amount
85 slivers[target]['_rspec'][resource_name] += amount
87 def deliver_record(self, rec):
88 """A record is simply a dictionary with 'name' and 'timestamp'
89 keys. We keep some persistent private data in the records under keys
90 that start with '_'; thus record updates should not displace such
92 if rec['timestamp'] < self._min_timestamp: return
94 old_rec = self.get(name)
95 if old_rec == None: self[name] = rec
96 elif rec['timestamp'] > old_rec['timestamp']:
97 for key in old_rec.keys():
98 if not key.startswith('_'): del old_rec[key]
101 def set_min_timestamp(self, ts):
102 """The ._min_timestamp member is the timestamp on the last comprehensive update.
103 We use it to determine if a record is stale.
104 This method should be called whenever new GetSlivers() data comes in."""
105 self._min_timestamp = ts
106 for name, rec in self.items():
107 if rec['timestamp'] < ts: del self[name]
110 """Synchronize reality with the database contents. This
111 method does a lot of things, and it's currently called after every
112 single batch of database changes (a GetSlivers(), a loan, a record).
113 It may be necessary in the future to do something smarter."""
115 # delete expired records
117 for name, rec in self.items():
118 if rec.get('expires', now) < now: del self[name]
120 self._compute_effective_rspecs()
123 x = coresched.CoreSched()
126 logger.log_exc("database: exception while doing core sched")
128 # create and destroy accounts as needed
129 logger.verbose("database: sync : fetching accounts")
130 existing_acct_names = account.all()
131 for name in existing_acct_names:
133 logger.verbose("database: sync : ensure_destroy'ing %s"%name)
134 account.get(name).ensure_destroyed()
135 for name, rec in self.iteritems():
136 # protect this; if anything fails for a given sliver
137 # we still need the other ones to be handled
139 sliver = account.get(name)
140 logger.verbose("database: sync : looping on %s (shell account class from pwd %s)" %(name,sliver._get_class()))
141 # Make sure we refresh accounts that are running
142 if rec['instantiation'] == 'plc-instantiated':
143 logger.verbose ("database: sync : ensure_create'ing 'instantiation' sliver %s"%name)
144 sliver.ensure_created(rec)
145 elif rec['instantiation'] == 'nm-controller':
146 logger.verbose ("database: sync : ensure_create'ing 'nm-controller' sliver %s"%name)
147 sliver.ensure_created(rec)
148 # Back door to ensure PLC overrides Ticket in delegation.
149 elif rec['instantiation'] == 'delegated' and sliver._get_class() != None:
150 # if the ticket has been delivered and the nm-controller started the slice
151 # update rspecs and keep them up to date.
152 if sliver.is_running():
153 logger.verbose ("database: sync : ensure_create'ing 'delegated' sliver %s"%name)
154 sliver.ensure_created(rec)
156 logger.log_exc("database: sync failed to handle sliver",name=name)
158 # Wake up bwmom to update limits.
160 global dump_requested
161 dump_requested = True
166 """The database dumper daemon.
167 When it starts up, it populates the database with the last dumped database.
168 It proceeds to handle dump requests forever."""
170 global dump_requested
173 while not dump_requested: db_cond.wait()
174 db_pickle = cPickle.dumps(db, cPickle.HIGHEST_PROTOCOL)
175 dump_requested = False
178 tools.write_file(DB_FILE, lambda f: f.write(db_pickle))
179 logger.log_database(db)
181 logger.log_exc("database.start: failed to pickle/dump")
185 try: db = cPickle.load(f)
188 logger.log ("database: Could not load %s -- starting from a fresh database"%DB_FILE)
191 logger.log_exc("database: failed in start")
193 logger.log('database.start')
194 tools.as_daemon_thread(run)