def union_subplans(self, plans):
op = Xunion(self.id_operator, self.eddies, inputs=len(plans))
self.operators.append(op)
union_vars = set()
height = 0
total_res = 0
for plan in plans:
union_vars.update(plan.variables)
height = max(height, plan.height)
total_res += plan.total_res
height += 1
tree_plan = TreePlan(op, union_vars, None, self.sources, plans, None,
height, total_res)
# Update signature of tuples.
self.operators_sym.update({self.id_operator: False})
self.operators_desc[self.id_operator] = {}
for source in tree_plan.sources:
self.operators_desc[self.id_operator].update({source: 0})
self.eofs_desc[source] = self.eofs_desc[source] | pow(
2, self.id_operator)
self.source_by_operator[
source] = self.source_by_operator[source] | pow(
2, self.id_operator)
self.plan_order[self.id_operator] = tree_plan.height
self.operators_vars[self.id_operator] = tree_plan.vars
self.id_operator += 1
return tree_plan
def create_triple_pattern_plan(self, triple_pattern):
self.eofs_desc.update({self.source_id: 0})
self.sources[self.source_id] = triple_pattern.variables
self.source_by_operator[self.source_id] = pow(2, self.id_operator)
self.operators_desc.setdefault(self.id_operator,
{})[self.source_id] = 0
left = IndependentOperator(self.source_id,
self.source,
triple_pattern,
self.source_by_operator,
triple_pattern.variables,
self.eddies,
self.source_by_operator,
sparql_limit=self.sparql_limit)
left.total_res = triple_pattern.cardinality
op = Xunion(self.id_operator, self.eddies, inputs=1)
self.operators.append(op)
# Update signature of tuples.
self.operators_sym.update({self.id_operator: False})
self.operators_desc[self.id_operator] = {}
source = self.source_id
self.operators_desc[self.id_operator].update({source: 0})
self.eofs_desc[source] = self.eofs_desc[source] | pow(
2, self.id_operator)
self.source_by_operator[
source] = self.source_by_operator[source] | pow(
2, self.id_operator)
self.plan_order[self.id_operator] = 1
self.operators_vars[self.id_operator] = left.vars
tree_plan = TreePlan(op, left.vars, left.vars, self.sources, left,
None, 1, left.total_res)
self.source_id += 1
self.independent_sources += 1
self.id_operator += 1
return tree_plan
def add_order_by(self, tree):
op = Xorderby(self.id_operator, self.eddies, self.query.order_by)
self.operators.append(op)
tree = TreePlan(op, tree.vars, tree.join_vars, tree.sources, tree,
None, tree.height + 1, tree.total_res)
# Update signature of tuples.
self.operators_sym.update({self.id_operator: False})
self.operators_desc[self.id_operator] = {}
for source in tree.sources:
self.operators_desc[self.id_operator].update({source: 0})
self.eofs_desc[source] = self.eofs_desc[source] | pow(
2, self.id_operator)
self.sources_desc[source] = self.sources_desc[source] | pow(
2, self.id_operator)
self.plan_order[self.id_operator] = tree.height
self.operators_vars[self.id_operator] = tree.vars
self.id_operator += 1
return tree
def join_subplans(self,
left,
right,
join_type=None,
card=-1,
logial_plan=None):
# Set Operator ID
if logial_plan:
self.operator_id2logical_plan[self.id_operator] = logial_plan
logial_plan.operator_id = self.id_operator
# Get Metadata for operator
if isinstance(left, TriplePattern):
# Get cardinality; Query only if necessary
left_card = left.count if not left.count is None else get_metadata(
self.source, left)
else:
left_card = left.total_res
if isinstance(right, TriplePattern):
# Get cardinality; Query only if necessary
right_card = right.count if not right.count is None else get_metadata(
self.source, right)
else:
right_card = right.total_res
# Pre-decided Join Type
if join_type:
xn_join = True if (issubclass(join_type, Xnjoin)) else False
xn_optional = issubclass(join_type, Xnoptional)
xg_optional = issubclass(join_type, Xgoptional)
if xn_join or xn_optional:
# Switch sides for NLJ
if left_card > right_card:
tmp = left
left = right
right = tmp
# Decide based in heursitics
else:
# Decide Join Type: xn = NLJ, FJ = SHJ
if isinstance(left, IndependentOperator):
xn_join = True if left_card < (right_card / 100.0) else False
else:
xn_join = True if left_card <= right_card else False
# Joins Variable info
join_vars = set(left.variables).intersection(right.variables)
all_variables = set(left.variables).union(right.variables)
# If the subplans have no varibale in common,
# always place a Hash Join to handle the Cross-Product
if len(join_vars) == 0:
xn_join = False
# Tree Plans as Leafs
if isinstance(left, TreePlan):
leaf_left = left
for source in left.sources.keys():
self.source_by_operator[
source] = self.source_by_operator[source] | pow(
2, self.id_operator)
self.operators_desc.setdefault(self.id_operator,
{})[source] = 0
self.operators_desc.setdefault(self.id_operator,
{})[self.source_id] = 0
if isinstance(right, TreePlan):
leaf_right = right
for source in right.sources.keys():
self.source_by_operator[
source] = self.source_by_operator[source] | pow(
2, self.id_operator)
self.operators_desc.setdefault(self.id_operator,
{})[source] = 1
if xn_join and isinstance(left, TreePlan) and isinstance(
right, TreePlan):
print("Invalid plan")
if xn_optional and isinstance(left, TreePlan) and isinstance(
right, TreePlan):
print("Invalid plan")
# Operator Leafs
if isinstance(left, TriplePattern) or isinstance(left, BGP):
self.eofs_desc.update({self.source_id: 0})
self.sources[self.source_id] = left.variables
self.source_by_operator[self.source_id] = pow(2, self.id_operator)
self.eofs_desc[self.source_id] = pow(2, self.id_operator)
self.operators_desc.setdefault(self.id_operator,
{})[self.source_id] = 0
# Base on operator, create operator
# If SHJ(FJ), use IO
# Or if it is a NLJ(XN) and left_plan is a TP, then use IO
if (not xn_join) or (xn_join and (isinstance(right, TriplePattern)
or isinstance(right, BGP))):
leaf_left = IndependentOperator(self.source_id,
self.source,
left,
self.source_by_operator,
left.variables,
self.eddies,
self.source_by_operator,
sparql_limit=self.sparql_limit)
self.independent_sources += 1
elif (xn_join or xn_optional) and isinstance(right, TreePlan):
leaf_left = DependentOperator(self.source_id, self.source,
left, self.source_by_operator,
left.variables,
self.source_by_operator)
self.dependent_sources += 1
leaf_left.total_res = left_card
self.source_id += 1
if isinstance(right, TriplePattern) or isinstance(right, BGP):
self.eofs_desc.update({self.source_id: 0})
self.sources[self.source_id] = right.variables
self.source_by_operator[self.source_id] = pow(2, self.id_operator)
self.eofs_desc[self.source_id] = pow(2, self.id_operator)
self.operators_desc.setdefault(self.id_operator,
{})[self.source_id] = 1
# Base on operator, create operator
if xn_join or xn_optional:
leaf_right = DependentOperator(self.source_id, self.source,
right, self.source_by_operator,
right.variables,
self.source_by_operator)
self.dependent_sources += 1
else:
leaf_right = IndependentOperator(
self.source_id,
self.source,
right,
self.source_by_operator,
right.variables,
self.eddies,
self.source_by_operator,
sparql_limit=self.sparql_limit)
self.independent_sources += 1
leaf_right.total_res = right_card
self.source_id += 1
self.operators_vars[self.id_operator] = join_vars
self.plan_order[self.id_operator] = max(leaf_left.height,
leaf_right.height)
# Place Join
if xn_join: # NLJ
#if isinstance(left, TreePlan) and isinstance(right, TriplePattern) and self.poly: # First condition only
# needed for poly bind join
if (isinstance(right, TriplePattern)
or isinstance(right, BGP)) and self.poly:
logger.debug("Placing Poly XN Join")
op = Poly_Xnjoin(self.id_operator,
join_vars,
self.eddies,
brtpf_mappings=self.brtpf_mappings,
sparql_mappings=self.sparql_mappings)
#logger.debug("Placing Poly Bind Join")
#op = Poly_Bind_Join(self.id_operator, join_vars, self.eddies, left_card=card)
else:
logger.debug("Placing XN Join")
op = Poly_Xnjoin(self.id_operator,
join_vars,
self.eddies,
brtpf_mappings=1,
sparql_mappings=1)
#op = Xnjoin(self.id_operator, join_vars, self.eddies)
self.operators_sym.update({self.id_operator: True})
# If Right side has to be DP
if not isinstance(leaf_right, DependentOperator):
# Switch Leafs
tmp = leaf_right
leaf_right = leaf_left
leaf_left = tmp
# Update operators_descs for current operator id
for key, value in self.operators_desc[
self.id_operator].items():
# Leaf Right is now the DP and needs to be input Right, i.e. 1
if key == leaf_right.sources.keys()[0]:
self.operators_desc[self.id_operator][key] = 1
# All other will be on the left_plan input
else:
self.operators_desc[self.id_operator][key] = 0
elif not xn_optional and not xg_optional: # SHJ
#op = Fjoin(self.id_operator, join_vars, self.eddies)
if isinstance(left, TreePlan) and isinstance(
right, TriplePattern) and self.poly:
# Place Polymorphic Hash Join Operator
op = Fjoin(self.id_operator, join_vars, self.eddies)
#logger.debug("Placing Poly FJoin")
#op = Poly_Fjoin(self.id_operator, join_vars, self.eddies, leaf_left, leaf_right)
else:
op = Fjoin(self.id_operator, join_vars, self.eddies)
self.operators_sym.update({self.id_operator: False})
elif not xg_optional: # XN Optional
op = Xnoptional(self.id_operator, left.variables, right.variables,
self.eddies)
#op = Xnjoin(self.id_operator, join_vars, self.eddies)
self.operators_sym.update({self.id_operator: True})
else: # XG Optional
op = Xgoptional(self.id_operator, left.variables, right.variables,
self.eddies)
self.operators_sym.update({self.id_operator: False})
# Add Operator
self.operators.append(op)
tree_height = max(leaf_left.height, leaf_right.height) + 1
#tree_sources = {k: v for k, v in self.sources.items()}
# 2020-03-04: Changed here to route everything properly
tree_sources = dict(leaf_left.sources)
tree_sources.update(dict(leaf_right.sources))
# Create Tree Plan
join_card = card
tree_plan = TreePlan(op, all_variables, join_vars, tree_sources,
leaf_left, leaf_right, tree_height, join_card)
if isinstance(op, Xnjoin) and isinstance(
leaf_left, TreePlan) and isinstance(leaf_right, TreePlan):
raise Exception
self.id_operator += 1
return tree_plan
def create_plan_original(self, query, eddies, source):
# Plan structures.
tree_height = 0
id_operator = 0
operators = []
operators_desc = {}
plan_order = {}
operators_vars = {}
ordered_subtrees = []
independent_sources = 0
eofs_operators_desc = {}
operators_sym = {}
sources_desc = {}
eofs_desc = {}
subtrees = []
# Create initial signatures and leaves of the plan.
for subquery in query.where.left.triple_patterns:
sources_desc.update({id_operator: 0})
eofs_desc.update({id_operator: 0})
leaf = IndependentOperator(id_operator, source, subquery, sources_desc, subquery.get_variables(), eddies, eofs_desc)
leaf.total_res = get_metadata(leaf.server, leaf.query)
subtrees.append(leaf)
ordered_subtrees.append(leaf.total_res)
id_operator += 1
# Order leaves depending on the cardinality of fragments.
keydict = dict(zip(subtrees, ordered_subtrees))
subtrees.sort(key=keydict.get)
# Stage 1: Generate left_plan-linear index nested stars.
stars = []
id_operator = 0
while len(subtrees) > 0:
to_delete = []
star_tree = subtrees.pop(0)
star_vars = star_tree.vars
tree_height = 0
independent_sources = independent_sources + 1
for j in range(0, len(subtrees)):
subtree_j = subtrees[j]
join_variables = set(star_vars) & set(subtree_j.join_vars)
all_variables = set(star_tree.vars) | set(subtree_j.vars)
# Case: There is a join.
if len(join_variables) > 0:
to_delete.append(subtree_j)
# Update signatures.
sources = {}
sources.update(star_tree.sources)
sources.update(subtree_j.sources)
operators_desc[id_operator] = {}
operators_vars[id_operator] = join_variables
eofs_operators_desc[id_operator] = {}
# The current tree is the left_plan argument of the plan.
for source in star_tree.sources.keys():
if len(set(sources[source]) & join_variables) > 0:
# TODO: Change the next 0 for len of something
operators_desc[id_operator].update({source: 0})
sources_desc[source] = sources_desc[source] | pow(2, id_operator)
# TODO: check this.
eofs_operators_desc[id_operator].update({source: 0})
eofs_desc[source] = eofs_desc[source] | pow(2, id_operator)
# The subtree j is the right_plan argument of the plan.
for source in subtree_j.sources.keys():
if len(set(sources[source]) & join_variables) > 0:
# TODO: Change the next q for len of something
operators_desc[id_operator].update({source: 1})
sources_desc[source] = sources_desc[source] | pow(2, id_operator)
# TODO: check this.
eofs_operators_desc[id_operator].update({source: 1})
eofs_desc[source] = eofs_desc[source] | pow(2, id_operator)
plan_order[id_operator] = tree_height
operators_vars[id_operator] = join_variables
tree_height = tree_height + 1
# Place physical operator estimating cardinality.
if isinstance(star_tree, IndependentOperator):
res = self.estimate_card(star_tree.total_res, subtree_j.total_res)
# Place a Nested Loop join.
if star_tree.total_res < (subtree_j.total_res / 100.0):
subtree_j = DependentOperator(subtree_j.sources, subtree_j.server, subtree_j.query,
subtree_j.sources_desc, subtree_j.vars, subtree_j.total_res)
op = Xnjoin(id_operator, join_variables, eddies)
operators.append(op)
star_tree = TreePlan(op, all_variables, join_variables, sources,
star_tree, subtree_j, tree_height, 0)
operators_sym.update({id_operator: False})
# Place a Symmetric Hash join.
else:
op = Fjoin(id_operator, join_variables, eddies)
operators.append(op)
star_tree = TreePlan(op, all_variables, join_variables, sources,
star_tree, subtree_j, tree_height, res)
independent_sources = independent_sources + 1
operators_sym.update({id_operator: True})
else:
# TODO: new change here
res = self.estimate_card(star_tree.total_res, subtree_j.total_res)
#res = (2.0 * star_tree.total_res * subtree_j.total_res) / (star_tree.total_res + subtree_j.total_res)
#res = (star_tree.total_res + subtree_j.total_res) / 2
if (star_tree.total_res / float(subtree_j.total_res) < 0.30) or (subtree_j.total_res > 100*1000 and star_tree.total_res < 100*1000) or (subtree_j.total_res < 100*5):
subtree_j = DependentOperator(subtree_j.sources, subtree_j.server, subtree_j.query,
subtree_j.sources_desc, subtree_j.vars, subtree_j.total_res)
op = Xnjoin(id_operator, join_variables, eddies)
operators.append(op)
star_tree = TreePlan(op, all_variables, join_variables, sources,
star_tree, subtree_j, tree_height)
operators_sym.update({id_operator: False})
else:
op = Fjoin(id_operator, join_variables, eddies)
operators.append(op)
star_tree = TreePlan(op, all_variables,
join_variables, sources, star_tree, subtree_j, tree_height, res)
independent_sources = independent_sources + 1
operators_sym.update({id_operator: True})
id_operator += 1
# Add current tree to the list of stars and
# remove from the list of subtrees to process.
stars.append(star_tree)
for elem in to_delete:
subtrees.remove(elem)
# Stage 2: Build bushy tree to combine SSGs with common variables.
while len(stars) > 1:
subtree_i = stars.pop(0)
for j in range(0, len(stars)):
subtree_j = stars[j]
all_variables = set(subtree_i.vars) | set(subtree_j.vars)
join_variables = set(subtree_i.join_vars) & set(subtree_j.join_vars)
# Case: There is a join between stars.
if len(join_variables) > 0:
# Update signatures.
sources = {}
sources.update(subtree_i.sources)
sources.update(subtree_j.sources)
operators_desc[id_operator] = {}
operators_vars[id_operator] = join_variables
eofs_operators_desc[id_operator] = {}
for source in subtree_i.sources.keys():
# This models the restriction: a tuple must have the join
# variable instantiated to be routed to a certain join.
if len(set(sources[source]) & join_variables) > 0:
# TODO: Change the next 0 for len of something
operators_desc[id_operator].update({source: 0})
sources_desc[source] = sources_desc[source] | pow(2, id_operator)
# TODO: Check this.
eofs_operators_desc[id_operator].update({source: 0})
eofs_desc[source] = eofs_desc[source] | pow(2, id_operator)
for source in subtree_j.sources.keys():
# This models the restriction: a tuple must have the join
# variable instantiated to be routed to a certain join.
if len(set(sources[source]) & join_variables) > 0:
# TODO: Change the next 1 for len of something
operators_desc[id_operator].update({source: 1})
sources_desc[source] = sources_desc[source] | pow(2, id_operator)
# TODO: Check this.
eofs_operators_desc[id_operator].update({source: 1})
eofs_desc[source] = eofs_desc[source] | pow(2, id_operator)
plan_order[id_operator] = max(subtree_i.height, subtree_j.height)
stars.pop(j)
# Place physical operators between stars.
if isinstance(subtree_j, IndependentOperator):
res = self.estimate_card(star_tree.total_res, subtree_j.total_res)
# This case models a satellite, therefore apply cardinality estimation.
if subtree_i.total_res < (subtree_j.total_res/100.0):
subtree_j = DependentOperator(subtree_j.sources, subtree_j.server, subtree_j.query,
subtree_j.sources_desc, subtree_j.vars, subtree_j.total_res)
op = Xnjoin(id_operator, join_variables, eddies)
operators.append(op)
stars.append(TreePlan(op, all_variables,
join_variables, sources, subtree_i, subtree_j,
max(subtree_i.height, subtree_j.height, res)))
# Adjust number of asynchronous leaves.
independent_sources = independent_sources - 1
operators_sym.update({id_operator: False})
else:
op = Fjoin(id_operator, join_variables, eddies)
operators.append(op)
stars.append(TreePlan(op, all_variables, join_variables,
sources, subtree_i, subtree_j,
max(subtree_i.height, subtree_j.height, res)))
operators_sym.update({id_operator: True})
else:
res = (subtree_i.total_res + subtree_j.total_res) / 2
op = Fjoin(id_operator, join_variables, eddies)
operators.append(op)
stars.append(TreePlan(op, all_variables, join_variables,
sources, subtree_i, subtree_j,
max(subtree_i.height, subtree_j.height, res)))
operators_sym.update({id_operator: True})
id_operator += 1
break
if len(subtrees) % 2 == 0:
tree_height += 1
tree_height += 1
tree = stars.pop()
# Adds the projection operator to the plan.
if query.projection:
op = Xproject(id_operator, query.projection, eddies)
operators.append(op)
tree = TreePlan(op,
tree.vars, tree.join_vars, tree.sources, tree, None, tree_height+1, tree.total_res)
# Update signature of tuples.
operators_sym.update({id_operator: False})
operators_desc[id_operator] = {}
eofs_operators_desc[id_operator] = {}
for source in tree.sources:
operators_desc[id_operator].update({source: 0})
eofs_operators_desc[id_operator].update({source: 0})
eofs_desc[source] = eofs_desc[source] | pow(2, id_operator)
sources_desc[source] = sources_desc[source] | pow(2, id_operator)
plan_order[id_operator] = tree_height
operators_vars[id_operator] = tree.vars
id_operator += 1
tree_height += 1
# Adds the distinct operator to the plan.
if query.distinct:
op = Xdistinct(id_operator, eddies)
operators.append(op)
tree = TreePlan(op, tree.vars, tree.join_vars, tree.sources,
tree, None, tree_height + 1, tree.total_res)
# Update signature of tuples.
operators_sym.update ({id_operator: False})
operators_desc[id_operator] = {}
eofs_operators_desc[id_operator] = {}
for source in tree.sources:
operators_desc[id_operator].update({source: 0})
eofs_operators_desc[id_operator].update({source: 0})
eofs_desc[source] = eofs_desc[source] | pow(2, id_operator)
sources_desc[source] = sources_desc[source] | pow(2, id_operator)
plan_order[id_operator] = tree_height
operators_vars[id_operator] = tree.vars
id_operator += 1
tree_height += 1
physical_plan = Plan(query_tree=tree, tree_height=tree.height,
operators_desc=operators_desc, sources_desc=sources_desc,
plan_order=plan_order, operators_vars=operators_vars,
independent_sources=independent_sources,
operators_sym=operators_sym, operators=operators)
return physical_plan