def get_best_node(self, service):
"""Returns the 'least loaded' node currently available, increments the
active count on that node, and decrements the slots currently available
"""
nodes = self._get_nodes_table(service)
service = self._get_service_id(service)
where = [
nodes.c.service == service, nodes.c.available > 0,
nodes.c.capacity > nodes.c.current_load, nodes.c.downed == 0
]
query = select([nodes]).where(and_(*where))
if self._is_sqlite:
# sqlite doesn't have the 'log' funtion, and requires
# coercion to a float for the sorting to work.
query = query.order_by(nodes.c.current_load * 1.0 /
nodes.c.capacity)
else:
# using log() increases floating-point precision on mysql
# and thus makes the sorting more accurate.
query = query.order_by(
sqlfunc.log(nodes.c.current_load) /
sqlfunc.log(nodes.c.capacity))
query = query.limit(1)
res = self._safe_execute(query)
one = res.fetchone()
if one is None:
# unable to get a node
res.close()
raise BackendError('unable to get a node')
nodeid = one.id
node = str(one.node)
res.close()
# updating the table
where = [nodes.c.service == service, nodes.c.node == node]
where = and_(*where)
fields = {
'available': nodes.c.available - 1,
'current_load': nodes.c.current_load + 1
}
query = update(nodes, where, fields)
con = self._safe_execute(query, close=True)
con.close()
return nodeid, node
def get_best_node(self, service):
"""Returns the 'least loaded' node currently available, increments the
active count on that node, and decrements the slots currently available
"""
nodes = self._get_nodes_table(service)
service = self._get_service_id(service)
where = [nodes.c.service == service,
nodes.c.available > 0,
nodes.c.capacity > nodes.c.current_load,
nodes.c.downed == 0]
query = select([nodes]).where(and_(*where))
if self._is_sqlite:
# sqlite doesn't have the 'log' funtion, and requires
# coercion to a float for the sorting to work.
query = query.order_by(nodes.c.current_load * 1.0 /
nodes.c.capacity)
else:
# using log() increases floating-point precision on mysql
# and thus makes the sorting more accurate.
query = query.order_by(sqlfunc.log(nodes.c.current_load) /
sqlfunc.log(nodes.c.capacity))
query = query.limit(1)
res = self._safe_execute(query)
one = res.fetchone()
if one is None:
# unable to get a node
res.close()
raise BackendError('unable to get a node')
nodeid = one.id
node = str(one.node)
res.close()
# updating the table
where = [nodes.c.service == service, nodes.c.node == node]
where = and_(*where)
fields = {'available': nodes.c.available - 1,
'current_load': nodes.c.current_load + 1}
query = update(nodes, where, fields)
con = self._safe_execute(query, close=True)
con.close()
return nodeid, node
def get_best_node(self, service):
"""Returns the 'least loaded' node currently available, increments the
active count on that node, and decrements the slots currently available
"""
nodes = self._get_nodes_table(service)
service = self._get_service_id(service)
# Pick the least-loaded node that has available slots.
where = [nodes.c.service == service,
nodes.c.available > 0,
nodes.c.capacity > nodes.c.current_load,
nodes.c.downed == 0,
nodes.c.backoff == 0]
query = select([nodes]).where(and_(*where))
if self._is_sqlite:
# sqlite doesn't have the 'log' funtion, and requires
# coercion to a float for the sorting to work.
query = query.order_by(nodes.c.current_load * 1.0 /
nodes.c.capacity)
else:
# using log() increases floating-point precision on mysql
# and thus makes the sorting more accurate.
query = query.order_by(sqlfunc.log(nodes.c.current_load) /
sqlfunc.log(nodes.c.capacity))
query = query.limit(1)
# We may have to re-try the query if we need to release more capacity.
# This loop allows a maximum of five retries before bailing out.
for _ in xrange(5):
res = self._safe_execute(query)
row = res.fetchone()
res.close()
if row is None:
# Try to release additional capacity from any nodes
# that are not fully occupied.
where = and_(nodes.c.service == service,
nodes.c.available <= 0,
nodes.c.capacity > nodes.c.current_load,
nodes.c.downed == 0)
fields = {
'available': sqlfunc.min(
nodes.c.capacity * self.capacity_release_rate,
nodes.c.capacity - nodes.c.current_load
),
}
res = self._safe_execute(update(nodes, where, fields))
res.close()
if res.rowcount == 0:
break
# Did we succeed in finding a node?
if row is None:
raise BackendError('unable to get a node')
nodeid = row.id
node = str(row.node)
# Update the node to reflect the new assignment.
# This is a little racy with concurrent assignments, but no big deal.
where = [nodes.c.service == service, nodes.c.node == node]
where = and_(*where)
fields = {'available': nodes.c.available - 1,
'current_load': nodes.c.current_load + 1}
query = update(nodes, where, fields)
con = self._safe_execute(query, close=True)
con.close()
return nodeid, node
def get_best_node(self, service, email):
"""Returns the 'least loaded' node currently available, increments the
active count on that node, and decrements the slots currently available
"""
nodes = self._get_nodes_table(service)
service = self._get_service_id(service)
query = select([nodes])
send_to_spanner = self.should_allocate_to_spanner(email)
if send_to_spanner:
query = query.where(nodes.c.id == self._spanner_node_id)
else:
# Pick the least-loaded node that has available slots.
query = query.where(and_(
nodes.c.service == service,
nodes.c.available > 0,
nodes.c.capacity > nodes.c.current_load,
nodes.c.downed == 0,
nodes.c.backoff == 0
))
if self._is_sqlite:
# sqlite doesn't have the 'log' funtion, and requires
# coercion to a float for the sorting to work.
query = query.order_by(
nodes.c.current_load * 1.0 /
nodes.c.capacity)
else:
# using log() increases floating-point precision on mysql
# and thus makes the sorting more accurate.
query = query.order_by(
sqlfunc.log(nodes.c.current_load) /
sqlfunc.log(nodes.c.capacity))
query = query.limit(1)
# We may have to re-try the query if we need to release more capacity.
# This loop allows a maximum of five retries before bailing out.
for _ in xrange(5):
res = self._safe_execute(query)
row = res.fetchone()
res.close()
if row is None:
# Try to release additional capacity from any nodes
# that are not fully occupied.
where = and_(nodes.c.service == service,
nodes.c.available <= 0,
nodes.c.capacity > nodes.c.current_load,
nodes.c.downed == 0)
fields = {
'available': self._sqlfunc_min(
nodes.c.capacity * self.capacity_release_rate,
nodes.c.capacity - nodes.c.current_load
),
}
res = self._safe_execute(update(nodes, where, fields))
res.close()
if res.rowcount == 0:
break
else:
break
# Did we succeed in finding a node?
if row is None:
raise BackendError('unable to get a node')
nodeid = row.id
node = str(row.node)
# Update the node to reflect the new assignment.
# This is a little racy with concurrent assignments, but no big
# deal.
where = [nodes.c.service == service, nodes.c.node == node]
where = and_(*where)
fields = {'current_load': nodes.c.current_load + 1}
if not send_to_spanner:
fields['available'] = self._sqlfunc_max(nodes.c.available - 1, 0)
query = update(nodes, where, fields)
con = self._safe_execute(query, close=True)
con.close()
return nodeid, node
q = q.filter (DumpFileEntry.binm10 < 16.)
q = q.filter (SimulationEntry.contains (session, [tag for tag in namespace.tags]))
fig, axes = plt.subplots (len (namespace.caches), 1, sharex = True)
try:
axes [0]
except TypeError:
axes = [axes]
if namespace.checkcache:
cache (session, q.add_entity (SimulationEntry).filter (DumpFileEntry.state == "presn").all (), {cache: getattr (kepler_utils.database.cache, cache) for cache in namespace.caches}, states = [state for state in namespace.states])
for cache, axis in zip (namespace.caches, axes):
colors = cycle (["blue", "green", "red"])
for state in namespace.states:
qs = q.filter (DumpFileEntry.state == state)
cign = CorePlot (qs, axis)
binkey = getattr (DumpFileEntry, namespace.binkey)
if namespace.binlog:
binkey = func.log (binkey)
s, e = cign.plotScatter (getattr (DumpFileEntry, namespace.binkey), cache, binkey = binkey, errorbars = True, color = next (colors), label = state)
axis.set_ylabel (cache)
if namespace.binlog:
axis.set_xscale ("log")
axis.set_xlabel (namespace.binkey)
axis.legend (loc = "upper left")
# CUSTOM SELECT FOR ACTIVITY CLIFFS THAT WILL BE MAPPED AGAINST A CLASS
A1, A2 = metadata.tables['chembl.activities'].alias(
), metadata.tables['chembl.activities'].alias()
FP1, FP2 = metadata.tables['chembl.compound_rdfps'].alias(
), metadata.tables['chembl.compound_rdfps'].alias()
join = A1.join(A2, A2.c.assay_id == A1.c.assay_id).join(
FP1, FP1.c.molregno == A1.c.molregno).join(FP2,
FP2.c.molregno == A2.c.molregno)
delta_tanimoto = 1 - func.rdkit.tanimoto_sml(FP1.c.circular_fp,
FP2.c.circular_fp)
delta_activity = func.abs(
func.log(A1.c.standard_value) -
func.log(A2.c.standard_value)).label('delta_activity')
sali = (delta_activity / delta_tanimoto).label('sali')
whereclause = and_(A1.c.activity_id != A2.c.activity_id,
A1.c.molregno < A2.c.molregno,
A1.c.standard_type == A2.c.standard_type,
A1.c.standard_value > 0, A2.c.standard_value > 0,
A1.c.standard_flag > 0, A2.c.standard_flag > 0,
A1.c.standard_units == 'nM', A2.c.standard_units == 'nM',
A1.c.relation == '=', A1.c.relation == '=', sali >= 1.5)
activity_cliffs = select([
A1.c.assay_id,
A1.c.activity_id.label('activity_bgn_id'),
A2.c.activity_id.label('activity_end_id'),