Load and save run times
[plstackapi.git] / planetstack / observer / event_loop.py
1 import time
2 import traceback
3 import commands
4 import threading
5 import json
6
7 from datetime import datetime
8 from collections import defaultdict
9 from core.models import *
10 from django.db.models import F, Q
11 from openstack.manager import OpenStackManager
12 from util.logger import Logger, logging, logger
13 #from timeout import timeout
14
15 debug_mode = False
16
17 logger = Logger(logfile='observer.log', level=logging.INFO)
18
19 class StepNotReady(Exception):
20         pass
21
22 def toposort(g, steps):
23         reverse = {}
24
25         for k,v in g.items():
26                 for rk in v:
27                         try:
28                                 reverse[rk].append(k)
29                         except:
30                                 reverse[rk]=k
31
32         sources = []
33         for k,v in g.items():
34                 if not reverse.has_key(k):
35                         sources.append(k)
36
37
38         for k,v in reverse.iteritems():
39                 if (not v):
40                         sources.append(k)
41
42         order = []
43         marked = []
44         while sources:
45                 n = sources.pop()
46                 try:
47                         for m in g[n]:
48                                 if m not in marked:
49                                         sources.append(m)
50                                         marked.append(m)
51                 except KeyError:
52                         pass
53                 if (n in steps):
54                         order.append(n)
55
56         return order
57
58 class PlanetStackObserver:
59         sync_steps = ['SyncNetworks','SyncNetworkSlivers','SyncSites','SyncSitePrivileges','SyncSlices','SyncSliceMemberships','SyncSlivers','SyncSliverIps']
60
61         def __init__(self):
62                 self.manager = OpenStackManager()
63                 # The Condition object that gets signalled by Feefie events
64                 self.load_sync_steps()
65                 self.event_cond = threading.Condition()
66                 self.load_enacted()
67
68         def wait_for_event(self, timeout):
69                 self.event_cond.acquire()
70                 self.event_cond.wait(timeout)
71                 self.event_cond.release()
72                 
73         def wake_up(self):
74                 logger.info('Wake up routine called. Event cond %r'%self.event_cond)
75                 self.event_cond.acquire()
76                 self.event_cond.notify()
77                 self.event_cond.release()
78
79         def load_sync_steps(self):
80                 dep_path = Config().pl_dependency_path
81                 try:
82                         # This contains dependencies between records, not sync steps
83                         self.model_dependency_graph = json.loads(open(dep_path).read())
84                 except Exception,e:
85                         raise e
86
87                 backend_path = Config().backend_dependency_path
88                 try:
89                         # This contains dependencies between backend records
90                         self.backend_dependency_graph = json.loads(open(backend_path).read())
91                 except Exception,e:
92                         raise e
93
94                 provides_dict = {}
95                 for s in sync_steps:
96                         for m in s.provides:
97                                 provides_dict[m]=s.__name__
98                                 
99                 step_graph = {}
100                 for k,v in model_dependency_graph.iteritems():
101                         try:
102                                 source = provides_dict[k]
103                                 for m in v:
104                                         try:
105                                                 dest = provides_dict[m]
106                                         except KeyError:
107                                                 pass
108                                                 # no deps, pass
109                                         step_graph[source]=dest
110                                         
111                         except KeyError:
112                                 pass
113                                 # no dependencies, pass
114                 
115                 if (backend_dependency_graph):
116                         backend_dict = {}
117                         for s in sync_steps:
118                                 for m in s.serves:
119                                         backend_dict[m]=s.__name__
120                                         
121                         for k,v in backend_dependency_graph.iteritems():
122                                 try:
123                                         source = backend_dict[k]
124                                         for m in v:
125                                                 try:
126                                                         dest = backend_dict[m]
127                                                 except KeyError:
128                                                         # no deps, pass
129                                                         pass
130                                                 step_graph[source]=dest
131                                                 
132                                 except KeyError:
133                                         pass
134                                         # no dependencies, pass
135
136                 dependency_graph = step_graph
137
138                 self.ordered_steps = toposort(dependency_graph, steps)
139                 self.load_run_times()
140                 
141
142         def check_duration(self):
143                 try:
144                         if (duration > S.deadline):
145                                 logger.info('Sync step %s missed deadline, took %.2f seconds'%(S.name,duration))
146                 except AttributeError:
147                         # S doesn't have a deadline
148                         pass
149
150         def update_run_time(self, step):
151                 self.last_run_times[step.name]=time.time()
152
153         def check_schedule(self, step):
154                 time_since_last_run = time.time() - self.last_run_times[step.name]
155                 try:
156                         if (time_since_last_run < step.requested_interval):
157                                 raise StepNotReady
158                 except AttributeError:
159                         logger.info('Step %s does not have requested_interval set'%step.name)
160                         raise StepNotReady
161         
162         def load_run_times(self):
163                 try:
164                         jrun_times = open('/tmp/observer_run_times').read()
165                         self.last_run_times = json.loads(jrun_times)
166                 except:
167                         self.last_run_times={}
168                         for e in self.ordered_steps:
169                                 self.last_run_times[e.name]=0
170
171
172
173         def save_run_times(self):
174                 run_times = json.dumps(self.last_run_times)
175                 open('/tmp/observer_run_times','w').write(run_times)
176
177         def check_class_dependency(self, step, failed_steps):
178                 for failed_step in failed_steps:
179                         if (failed_step in self.dependency_graph[step.name]):
180                                 raise StepNotReady
181
182         def run(self):
183                 if not self.manager.enabled or not self.manager.has_openstack:
184                         return
185
186                 while True:
187                         try:
188                                 logger.info('Waiting for event')
189                                 tBeforeWait = time.time()
190                                 self.wait_for_event(timeout=300)
191                                 logger.info('Observer woke up')
192
193                                 # Set of whole steps that failed
194                                 failed_steps = []
195
196                                 # Set of individual objects within steps that failed
197                                 failed_step_objects = []
198
199                                 for S in self.ordered_steps:
200                                         start_time=time.time()
201                                         
202                                         sync_step = S()
203                                         sync_step.dependencies = self.dependencies[sync_step.name]
204                                         sync_step.debug_mode = debug_mode
205
206                                         should_run = False
207                                         try:
208                                                 # Various checks that decide whether
209                                                 # this step runs or not
210                                                 self.check_class_dependency(sync_step, failed_steps) # dont run Slices if Sites failed
211                                                 self.check_schedule(sync_step) # dont run sync_network_routes if time since last run < 1 hour
212                                                 should_run = True
213                                         except StepNotReady:
214                                                 logging.info('Step not ready: %s'%sync_step.name)
215                                                 failed_steps.add(sync_step)
216                                         except:
217                                                 failed_steps.add(sync_step)
218
219                                         if (should_run):
220                                                 try:
221                                                         duration=time.time() - start_time
222
223                                                         # ********* This is the actual sync step
224                                                         failed_objects = sync_step(failed=failed_step_objects)
225
226
227                                                         check_deadline(sync_step, duration)
228                                                         failed_step_objects.extend(failed_objects)
229                                                         self.update_run_time(sync_step)
230                                                 except:
231                                                         failed_steps.add(S)
232                                 self.save_run_times()
233                         except:
234                                 logger.log_exc("Exception in observer run loop")
235                                 traceback.print_exc()