Revision b7309a0d
| b/daemons/ganeti-watcher | ||
|---|---|---|
| 30 | 30 |
import os |
| 31 | 31 |
import sys |
| 32 | 32 |
import time |
| 33 |
import fcntl |
|
| 34 |
import errno |
|
| 35 | 33 |
import logging |
| 36 | 34 |
from optparse import OptionParser |
| 37 | 35 |
|
| ... | ... | |
| 46 | 44 |
|
| 47 | 45 |
|
| 48 | 46 |
MAXTRIES = 5 |
| 49 |
BAD_STATES = ['stopped']
|
|
| 50 |
HELPLESS_STATES = ['(node down)']
|
|
| 47 |
BAD_STATES = ['ERROR_down']
|
|
| 48 |
HELPLESS_STATES = ['ERROR_nodedown']
|
|
| 51 | 49 |
NOTICE = 'NOTICE' |
| 52 | 50 |
ERROR = 'ERROR' |
| 53 | 51 |
KEY_RESTART_COUNT = "restart_count" |
| ... | ... | |
| 238 | 236 |
"""Get a list of instances on this cluster. |
| 239 | 237 |
|
| 240 | 238 |
""" |
| 241 |
fields = ["name", "oper_state", "admin_state"]
|
|
| 239 |
fields = ["name", "status", "admin_state"]
|
|
| 242 | 240 |
|
| 243 | 241 |
if with_secondaries is not None: |
| 244 | 242 |
fields.append("snodes")
|
| ... | ... | |
| 327 | 325 |
try: |
| 328 | 326 |
logging.info("Activating disks for instance %s", instance.name)
|
| 329 | 327 |
instance.ActivateDisks() |
| 330 |
except Exception, err: |
|
| 331 |
logging.error(str(err), exc_info=True) |
|
| 328 |
except Exception: |
|
| 329 |
logging.exception("Error while activating disks for instance %s",
|
|
| 330 |
instance.name) |
|
| 332 | 331 |
|
| 333 | 332 |
# Keep changed boot IDs |
| 334 | 333 |
for name in check_nodes: |
| ... | ... | |
| 339 | 338 |
|
| 340 | 339 |
""" |
| 341 | 340 |
for instance in self.instances: |
| 342 |
# Don't care about manually stopped instances |
|
| 343 |
if not instance.autostart: |
|
| 344 |
continue |
|
| 345 |
|
|
| 346 | 341 |
if instance.state in BAD_STATES: |
| 347 | 342 |
n = notepad.NumberOfRestartAttempts(instance) |
| 348 | 343 |
|
| ... | ... | |
| 361 | 356 |
instance.name, last) |
| 362 | 357 |
instance.Restart() |
| 363 | 358 |
self.started_instances.add(instance.name) |
| 364 |
except Exception, err:
|
|
| 365 |
logging.error(str(err), exc_info=True)
|
|
| 359 |
except Exception: |
|
| 360 |
logging.exception("Erro while restarting instance %s", instance.name)
|
|
| 366 | 361 |
|
| 367 | 362 |
notepad.RecordRestartAttempt(instance) |
| 368 | 363 |
elif instance.state in HELPLESS_STATES: |
| ... | ... | |
| 373 | 368 |
notepad.RemoveInstance(instance) |
| 374 | 369 |
logging.info("Restart of %s succeeded", instance.name)
|
| 375 | 370 |
|
| 376 |
def VerifyDisks(self): |
|
| 371 |
@staticmethod |
|
| 372 |
def VerifyDisks(): |
|
| 377 | 373 |
"""Run gnt-cluster verify-disks. |
| 378 | 374 |
|
| 379 | 375 |
""" |
Also available in: Unified diff