from ganeti import query
from ganeti import runtime
from ganeti import pathutils
+from ganeti import ht
CLIENT_REQUEST_WORKERS = 16
EXIT_NODESETUP_ERROR = constants.EXIT_NODESETUP_ERROR
+def _LogNewJob(status, info, ops):
+ """Log information about a recently submitted job.
+
+ """
+ op_summary = utils.CommaJoin(op.Summary() for op in ops)
+
+ if status:
+ logging.info("New job with id %s, summary: %s", info, op_summary)
+ else:
+ logging.info("Failed to submit job, reason: '%s', summary: %s",
+ info, op_summary)
+
+
class ClientRequestWorker(workerpool.BaseWorker):
# pylint: disable=W0221
def RunTask(self, server, message, client):
# TODO: Rewrite to not exit in each 'if/elif' branch
if method == luxi.REQ_SUBMIT_JOB:
- logging.info("Received new job")
+ logging.info("Receiving new job")
(job_def, ) = args
ops = [opcodes.OpCode.LoadOpCode(state) for state in job_def]
- return queue.SubmitJob(ops)
+ job_id = queue.SubmitJob(ops)
+ _LogNewJob(True, job_id, ops)
+ return job_id
elif method == luxi.REQ_SUBMIT_MANY_JOBS:
- logging.info("Received multiple jobs")
+ logging.info("Receiving multiple jobs")
(job_defs, ) = args
jobs = []
for ops in job_defs:
jobs.append([opcodes.OpCode.LoadOpCode(state) for state in ops])
- return queue.SubmitManyJobs(jobs)
+ job_ids = queue.SubmitManyJobs(jobs)
+ for ((status, job_id), ops) in zip(job_ids, jobs):
+ _LogNewJob(status, job_id, ops)
+ return job_ids
elif method == luxi.REQ_CANCEL_JOB:
(job_id, ) = args
logging.info("Received job cancel request for %s", job_id)
return queue.CancelJob(job_id)
+ elif method == luxi.REQ_CHANGE_JOB_PRIORITY:
+ (job_id, priority) = args
+ logging.info("Received request to change priority for job %s to %s",
+ job_id, priority)
+ return queue.ChangeJobPriority(job_id, priority)
+
elif method == luxi.REQ_ARCHIVE_JOB:
(job_id, ) = args
logging.info("Received job archive request for %s", job_id)
op = opcodes.OpGroupQuery(names=names, output_fields=fields)
return self._Query(op)
+ elif method == luxi.REQ_QUERY_NETWORKS:
+ (names, fields, use_locking) = args
+ logging.info("Received network query request for %s", names)
+ if use_locking:
+ raise errors.OpPrereqError("Sync queries are not allowed",
+ errors.ECODE_INVAL)
+ op = opcodes.OpNetworkQuery(names=names, output_fields=fields)
+ return self._Query(op)
+
elif method == luxi.REQ_QUERY_EXPORTS:
(nodes, use_locking) = args
if use_locking:
elif method == luxi.REQ_SET_WATCHER_PAUSE:
(until, ) = args
- if until is None:
- logging.info("Received request to no longer pause the watcher")
- else:
- if not isinstance(until, (int, float)):
- raise TypeError("Duration must be an integer or float")
-
- if until < time.time():
- raise errors.GenericError("Unable to set pause end time in the past")
-
- logging.info("Received request to pause the watcher until %s", until)
-
- return _SetWatcherPause(until)
+ return _SetWatcherPause(context, until)
else:
logging.info("Received invalid request '%s'", method)
self.glm = locking.GanetiLockManager(
self.cfg.GetNodeList(),
self.cfg.GetNodeGroupList(),
- self.cfg.GetInstanceList())
+ self.cfg.GetInstanceList(),
+ self.cfg.GetNetworkList())
self.cfg.SetContext(self)
self.glm.remove(locking.LEVEL_NODE_RES, name)
-def _SetWatcherPause(until):
+def _SetWatcherPause(context, until):
"""Creates or removes the watcher pause file.
+ @type context: L{GanetiContext}
+ @param context: Global Ganeti context
@type until: None or int
@param until: Unix timestamp saying until when the watcher shouldn't run
"""
+ node_names = context.cfg.GetNodeList()
+
if until is None:
- utils.RemoveFile(pathutils.WATCHER_PAUSEFILE)
+ logging.info("Received request to no longer pause watcher")
else:
- utils.WriteFile(pathutils.WATCHER_PAUSEFILE,
- data="%d\n" % (until, ))
+ if not ht.TNumber(until):
+ raise TypeError("Duration must be numeric")
+
+ if until < time.time():
+ raise errors.GenericError("Unable to set pause end time in the past")
+
+ logging.info("Received request to pause watcher until %s", until)
+
+ result = context.rpc.call_set_watcher_pause(node_names, until)
+
+ errmsg = utils.CommaJoin("%s (%s)" % (node_name, nres.fail_msg)
+ for (node_name, nres) in result.items()
+ if nres.fail_msg and not nres.offline)
+ if errmsg:
+ raise errors.OpExecError("Watcher pause was set where possible, but failed"
+ " on the following node(s): %s" % errmsg)
return until