Adam Sadovsky | 62e37f3 | 2014-10-23 18:14:20 -0700 | [diff] [blame] | 1 | #!/usr/bin/python2.7 |
| 2 | |
Adam Sadovsky | 78f2245 | 2014-10-23 18:50:04 -0700 | [diff] [blame] | 3 | """Playground GCE monitoring script. |
Adam Sadovsky | 62e37f3 | 2014-10-23 18:14:20 -0700 | [diff] [blame] | 4 | |
Adam Sadovsky | 78f2245 | 2014-10-23 18:50:04 -0700 | [diff] [blame] | 5 | This needs to run on a GCE VM with the replica pool service account scope |
| 6 | (https://www.googleapis.com/auth/ndev.cloudman). |
| 7 | |
| 8 | You also need to enable preview in gcloud: |
| 9 | $ gcloud components update preview |
| 10 | |
| 11 | Then add it to your crontab, e.g. |
| 12 | */10 * * * * gcloud preview replica-pools --zone us-central1-a replicas --pool playground-pool list|monitor.py |
| 13 | """ |
| 14 | |
Adam Sadovsky | 62e37f3 | 2014-10-23 18:14:20 -0700 | [diff] [blame] | 15 | import datetime |
| 16 | import subprocess |
| 17 | import sys |
| 18 | import yaml |
| 19 | |
| 20 | DESIRED = 2 |
| 21 | MAX_ALIVE_MIN = 60 |
Adam Sadovsky | 78f2245 | 2014-10-23 18:50:04 -0700 | [diff] [blame] | 22 | POOL = 'playground-pool' |
Adam Sadovsky | 62e37f3 | 2014-10-23 18:14:20 -0700 | [diff] [blame] | 23 | |
| 24 | |
Adam Sadovsky | 78f2245 | 2014-10-23 18:50:04 -0700 | [diff] [blame] | 25 | def RunCommand(*args): |
| 26 | cmd = ['gcloud', 'preview', 'replica-pools', '--zone', 'us-central1-a'] |
| 27 | cmd.extend(args) |
| 28 | subprocess.check_call(cmd) |
Adam Sadovsky | 62e37f3 | 2014-10-23 18:14:20 -0700 | [diff] [blame] | 29 | |
| 30 | |
Adam Sadovsky | 78f2245 | 2014-10-23 18:50:04 -0700 | [diff] [blame] | 31 | def ResizePool(size): |
| 32 | RunCommand('resize', '--new-size', str(size), POOL) |
Adam Sadovsky | 62e37f3 | 2014-10-23 18:14:20 -0700 | [diff] [blame] | 33 | |
| 34 | |
Adam Sadovsky | 78f2245 | 2014-10-23 18:50:04 -0700 | [diff] [blame] | 35 | def ShouldRestart(replica): |
| 36 | if replica['status']['state'] == 'PERMANENTLY_FAILING': |
| 37 | print 'Replica %s failed: %s' % ( |
| 38 | replica['name'], replica['status']['details']) |
| 39 | return True |
| 40 | return IsTooOld(replica) |
Adam Sadovsky | 62e37f3 | 2014-10-23 18:14:20 -0700 | [diff] [blame] | 41 | |
| 42 | |
Adam Sadovsky | 78f2245 | 2014-10-23 18:50:04 -0700 | [diff] [blame] | 43 | def IsTooOld(replica): |
| 44 | start_text = replica['status']['vmStartTime'] |
| 45 | if start_text: |
| 46 | start = yaml.load(start_text) |
| 47 | uptime = datetime.datetime.now() - start |
| 48 | return uptime.seconds > MAX_ALIVE_MIN * 60 |
| 49 | |
| 50 | |
| 51 | def RestartReplica(replica): |
| 52 | print 'Restarting replica ' + replica['name'] |
| 53 | ResizePool(DESIRED + 1) |
| 54 | RunCommand('replicas', '--pool', POOL, 'delete', replica['name']) |
| 55 | |
| 56 | |
| 57 | def MaybeRestartReplica(replica): |
| 58 | if ShouldRestart(replica): |
| 59 | RestartReplica(replica) |
Adam Sadovsky | 62e37f3 | 2014-10-23 18:14:20 -0700 | [diff] [blame] | 60 | |
| 61 | |
| 62 | def main(): |
Adam Sadovsky | 78f2245 | 2014-10-23 18:50:04 -0700 | [diff] [blame] | 63 | replicas = yaml.load_all(sys.stdin.read()) |
| 64 | for replica in replicas: |
| 65 | MaybeRestartReplica(replica) |
Adam Sadovsky | 62e37f3 | 2014-10-23 18:14:20 -0700 | [diff] [blame] | 66 | |
| 67 | |
Adam Sadovsky | 78f2245 | 2014-10-23 18:50:04 -0700 | [diff] [blame] | 68 | if __name__ == '__main__': |
| 69 | main() |