def hold_on_submit(hook_config, job):
'''
Hold every job so that we can process it properly in the periodic hook.
'''
pbs.logmsg(pbs.LOG_DEBUG,
"cycle_sub_hook - holding job %s with hold_type 'so'" % job.id)
job.Hold_Types = pbs.hold_types("so")
def placement_hook(hook_config, job):
if not get_select(job):
# pbs 18 seems to treat host booleans as strings, which is causing this very annoying workaround.
#job.Resource_List["ungrouped"] = "true"
if job.Resource_List["slot_type"]:
job.Resource_List["slot_type"] = job.Resource_List["slot_type"]
# Check to see if job is interactive
if job.interactive:
debug("Job is interactive")
return
debug(
"The job doesn't have a select statement, it doesn't have any placement requirements."
)
debug("Place a hold on the job")
job.Hold_Types = pbs.hold_types("so")
return
if validate_groupid_placement(job):
_, select_dict = parse_select(job)
if "ungrouped" not in select_dict:
set_select_key(job, "ungrouped", "false")
slot_type = select_dict.get("slot_type")
if slot_type:
set_select_key(job, "slot_type", slot_type)
debug("Using the grouped slot_type as a resource (%s)." %
slot_type)
def placement_hook(hook_config, job):
if not get_select(job):
# pbs 18 seems to treat host booleans as strings, which is causing this very annoying workaround.
#job.Resource_List["ungrouped"] = "true"
if job.Resource_List["slot_type"]:
job.Resource_List["slot_type"] = job.Resource_List["slot_type"]
# Check to see if job is interactive
if job.interactive:
debug("Job is interactive")
return
debug(
"The job doesn't have a select statement, it doesn't have any placement requirements."
)
debug("Place a hold on the job")
job.Hold_Types = pbs.hold_types("so")
return
debug("The job has a select statement.")
debug("Place a hold on the job")
job.Hold_Types = pbs.hold_types("so")
return
'''
def hold_on_submit(hook_config, event):
"""
Hold every job so that we can process it properly in the periodic hook.
"""
job = event.job
if job.Resource_List["nodes"]:
job.Resource_List[]
debug(
(
"holding job %s with hold_type 'so' because it uses -l nodes syntax."
+ " It will be updated and released shortly."
)
% job.id
)
job.Hold_Types = pbs.hold_types("so")
event.accept()
return
return
new_select = selspec.increment_chunks(1)
e.job.Resource_List["select"] = new_select
pbs.logmsg(pbs.LOG_DEBUG, "job's select spec changed to %s" % new_select)
elif e.type == pbs.EXECJOB_LAUNCH:
if 'PBS_NODEFILE' not in e.env:
e.accept()
# add a log entry in primary mom logs
pbs.logmsg(pbs.LOG_DEBUG, "Executing launch")
# print out the vnode_list[] values
for vn in e.vnode_list:
v = e.vnode_list[vn]
pbs.logjobmsg(e.job.id, "launch: found vnode_list[" + v.name + "]")
# print out the vnodes in vnode_list_fail[] and offline them
for vn in e.vnode_list_fail:
v = e.vnode_list_fail[vn]
pbs.logjobmsg(
e.job.id, "launch: found vnode_list_fail[" + v.name + "]")
v.state = pbs.ND_OFFLINE
# prune the job's vnodes to satisfy the select spec in resource 'site'
# and vnodes in vnode_list_fail[] are not used.
if e.job.in_ms_mom():
pj = e.job.release_nodes(keep_select=e.job.Resource_List["site"])
if pj is None:
e.job.Hold_Types = pbs.hold_types("s")
e.job.rerun()
e.reject("unsuccessful at LAUNCH")
new_select = selspec.increment_chunks(1)
e.job.Resource_List["select"] = new_select
pbs.logmsg(pbs.LOG_DEBUG, "job's select spec changed to %s" % new_select)
elif e.type == pbs.EXECJOB_LAUNCH:
if 'PBS_NODEFILE' not in e.env:
e.accept()
# add a log entry in primary mom logs
pbs.logmsg(pbs.LOG_DEBUG, "Executing launch")
# print out the vnode_list[] values
for vn in e.vnode_list:
v = e.vnode_list[vn]
pbs.logjobmsg(e.job.id, "launch: found vnode_list[" + v.name + "]")
# print out the vnodes in vnode_list_fail[] and offline them
for vn in e.vnode_list_fail:
v = e.vnode_list_fail[vn]
pbs.logjobmsg(e.job.id,
"launch: found vnode_list_fail[" + v.name + "]")
v.state = pbs.ND_OFFLINE
# prune the job's vnodes to satisfy the select spec in resource 'site'
# and vnodes in vnode_list_fail[] are not used.
if e.job.in_ms_mom():
pj = e.job.release_nodes(keep_select=e.job.Resource_List["site"])
if pj is None:
e.job.Hold_Types = pbs.hold_types("s")
e.job.rerun()
e.reject("unsuccessful at LAUNCH")
