def select_sources(self, triple_pattern):
if not self.__full_stats:
get_metadata(self.__sources, triple_pattern)
true_sources = set([
name.replace("tpf@", "")
for name in triple_pattern.sources.keys()
])
else:
true_sources = set(
self.__full_stats.predicate_counts(triple_pattern[1]).keys())
self.processed_predicates += 1
sources = self.__stats.predicate_counts(triple_pattern[1])
delta = true_sources - set(sources).intersection(true_sources)
#print(len(delta))
if len(sources) == 0:
if triple_pattern[1].isuri():
self.missed_predicate_cnt += 1
self.missed_predicates[triple_pattern[1]] = list(delta)
elif len(delta) > 0 and triple_pattern[1].isuri():
self.missed_sources[triple_pattern[1]] = delta
triple_pattern.sources = sources
def select_sources(self, triple_pattern):
if self.__stats and triple_pattern[1].isuri():
predicate = triple_pattern[1].value.replace("<",
"").replace(">", "")
# Update stats
#card, auth_stats = get_metadata_tpf_stats(self.__sources, triple_pattern)
#self.__stats.update_authorities(predicate, auth_stats)
#return card
return get_metadata(self.__sources, triple_pattern)
else:
return get_metadata(self.__sources, triple_pattern)
def optimize_bgp(self, triple_patterns):
for tp in triple_patterns:
if isinstance(tp, TriplePattern):
# For each server, we need one requests to get the metadata
self.planning_requests += len(self.sources)
get_metadata(self.sources, tp)
# Compute E_star for completeness and max cost
E_star = self.compute_completeness(triple_patterns)
self.max_cost = self.compute_cost(triple_patterns)
# Prune Sources
if self.prune_sources:
triple_patterns = self.prune_relevant_source(triple_patterns)
# Compute E and completness
E = self.compute_completeness(triple_patterns)
try:
bgp_comp = E / E_star
except ZeroDivisionError:
bgp_comp = 1.0
self.decompostion_completeness.append(bgp_comp)
# Decomposition
if self.decomposer:
# Compute Decomposition
decomposition = LDFF_Decomposer.get_decomposition(triple_patterns)
else:
decomposition = triple_patterns
# Compute Cost
try:
bgp_cost = self.compute_cost(decomposition) / self.max_cost
except ZeroDivisionError:
bgp_cost = 1.0
self.decompostion_cost.append(bgp_cost)
# Get plan
plan = self.decompostion_to_plan(decomposition)
return plan
def select_sources(self, triple_pattern):
predicate = triple_pattern[1].value[1:-1]
if triple_pattern.variable_position == 5:
sources = self.__stats.predicate_counts(predicate)
triple_pattern.sources = sources
triple_pattern.cardinality = sum(sources.values())
return triple_pattern.cardinality
else:
relevant_sources = self.__stats.sources_by_predicate(predicate)
sources = relevant_sources if len(
relevant_sources) > 0 else self.sources
return get_metadata(sources, triple_pattern)
def capability_aware_decomp(decomposition, sparql_exclusive=True):
D = []
for id, subquery in decomposition.items():
for source in subquery[1]:
if source.startswith("sparql@") and len(subquery[0]) > 1:
count = get_metadata([source], subquery[0])
if count > 0:
S_c = (BGP(subquery[0]), source, count)
D.append(S_c)
else:
for triple_pattern in subquery[0]:
S_c = (triple_pattern, source,
triple_pattern.sources[source])
D.append(S_c)
else:
for triple_pattern in subquery[0]:
if source in triple_pattern.sources.keys():
S_c = (triple_pattern, source,
triple_pattern.sources[source])
D.append(S_c)
return D
def get_decomposition(triple_patterns):
Q = triple_patterns
change = True
while change:
for sq_i, sq_j in combinations(Q, 2):
if isinstance(sq_i, Filter) or isinstance(sq_j, Filter):
continue
if sq_i.compatible(sq_j):
if len(sq_i.sources.keys()) == 1 and len(sq_j.sources.keys()) == 1 \
and len(set(sq_i.sources.keys()).intersection(set(sq_j.sources.keys()))) == 1:
if sq_i.sources.keys()[0].startswith("sparql@"):
Q.remove(sq_j)
Q.remove(sq_i)
if isinstance(sq_i, BGP):
tps = sq_i.triple_patterns
else:
tps = [sq_i]
if isinstance(sq_j, BGP):
tps.extend(sq_j.triple_patterns)
else:
tps.append(sq_j)
new_sq = BGP(tps)
Q.append(new_sq)
break
else:
change = False
# Update Cardinalities if a BGP was created
for sq in Q:
if isinstance(sq, BGP):
sq.cardinality = get_metadata(sq.sources.keys(), sq)
return Q
def iterative_dynamic_programming1(self, triple_patterns):
if len(triple_patterns) == 1:
# For each server, we need one requests to get the metadata
self.planning_requests += len(self.sources)
get_metadata(self.sources, triple_patterns[0])
return LogicalPlan(triple_patterns[0])
best_row = False
opt_plan = {}
toDo = set()
k = min(len(triple_patterns), self.k)
if self.adaptive_k and len(triple_patterns) >= 6:
k = 2
for index, triple_pattern in enumerate(triple_patterns):
# For each server, we need one requests to get the metadata
self.planning_requests += len(self.sources)
get_metadata(self.sources, triple_pattern)
accessPlan = set([LogicalPlan(triple_pattern)])
opt_plan[(triple_pattern, )] = accessPlan
toDo.add(triple_pattern)
while len(toDo) > 1:
k = min(k, len(toDo))
for i in range(2, k + 1):
for S in combinations(toDo, i):
opt_plan[S] = set()
for O in self.true_subset(S):
try:
opt_plan_O = opt_plan[O]
S_minus_O = tuple(set(S).difference(set(O)))
opt_plan_S_minus_O = opt_plan.get(S_minus_O, None)
if not opt_plan_S_minus_O or not opt_plan_O:
continue
for opt_plan_o in opt_plan_O:
for opt_plan_s_minus_o in opt_plan_S_minus_O:
join_vars = opt_plan_o + opt_plan_s_minus_o
if join_vars > 0:
join_plans = self.joinPlans(
opt_plan_o, opt_plan_s_minus_o)
join_plans_S = opt_plan[S].union(
join_plans)
opt_plan[S] = self.best_n_plans(
list(join_plans_S), self.top_t)
#opt_plan[S] = join_plans_S
except Exception as e:
raise e
best_plans = []
V = set()
for key, values in opt_plan.items():
for value in values:
k_len = len(key)
if k_len == k and value and set(key).issubset(toDo):
V.add(key)
rob = value.cost
best_plans.append((value, value.cost, rob, key))
if len(best_plans) == 0:
raise Exception("IDP Error: No best plan")
for v in V:
del opt_plan[v]
try:
if len(best_plans) > 0:
# In intermediate steps of IDP: Take best plan only
best_plan = sorted(best_plans, key=lambda x:
(x[1], x[2]))[0]
tps = best_plan[3]
opt_plan[(tps, )] = set([best_plan[0]])
if best_row:
best_plans.remove(best_plan)
# Best Row
for bp in best_plans:
if bp[3] == tps:
opt_plan[(tps, )].add(bp[0])
# Remove triple patterns from todo list
for tp in tps:
toDo.remove(tp)
toDo.add(best_plan[3])
except Exception as e:
raise e
tmp_plans = []
for plan in best_plans:
cost = plan[0].cost
rob = plan[0].average_cost(self.robust_model)
tmp_plans.append((plan, cost, rob, cost / rob))
#print((plan, cost, rob, cost/rob))
cheap_plan = sorted(tmp_plans, key=lambda x: (x[1], x[3]))[0]
# Decision rule for robust plan
self.robust_over_cost = False
rob_cost_ratio = cheap_plan[1] / cheap_plan[2]
self.cost_robust_ratio = rob_cost_ratio
#logger.debug("{} {}".format(self.cost_robust_ratio, len(tmp_plans) ))
if len(tmp_plans) > 1:
tmp_plans.remove(cheap_plan)
plans_over_thrshld = filter(
lambda x: x[3] >= self.robustness_threshold, tmp_plans)
if not plans_over_thrshld or len(plans_over_thrshld) == 0:
plans_over_thrshld = tmp_plans
robust_plan = sorted(plans_over_thrshld,
key=lambda x: (x[1], x[2]))[0]
else:
robust_plan = cheap_plan
# What is the cost ratio of the cheapest and the most robust plan
cost_cost_ratio = cheap_plan[1] / robust_plan[1]
self.cost_cost_ratio = cost_cost_ratio
self.robust_over_cost = rob_cost_ratio <= self.robustness_threshold and cost_cost_ratio >= self.cost_threshold
if self.enable_robustplan and self.robust_over_cost:
logger.debug("IDP: Robust Plan over Cheapest Plan")
return robust_plan[0][0]
return cheap_plan[0][0]
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 optimize_subquery(self, triples):
subtrees = []
for triple in triples:
# For each server, we need one requests to get the metadata
self.planning_requests += len(self.sources)
get_metadata(self.sources, triple)
leaf = LogicalPlan(triple)
subtrees.append(leaf)
subtrees.sort(key=lambda x: x.cardinality)
stars = []
while len(subtrees) > 0:
to_delete = []
star_tree = subtrees.pop(0)
star_vars = star_tree.variables
for j in range(0, len(subtrees)):
subtree_j = subtrees[j]
join_variables = set(star_vars).intersection(subtree_j.variables)
# Case: There is a join.
if len(join_variables) > 0:
to_delete.append(subtree_j)
# Place physical operator estimating cardinality.
if star_tree.is_triple_pattern:
res = self.estimate_card(star_tree.cardinality, subtree_j.cardinality)
# Place a Nested Loop join.
# Paper; if (tpi.count / tpi.pagesize) <= s.count then
#if star_tree.total_res < (subtree_j.total_res / 100.0):
if star_tree.cardinality < (subtree_j.cardinality / 100.0):
join_type = Xnjoin
# If NLJ is placed, set res = 0/1 to force NLJs later
res = 1
# Place a Symmetric Hash join.
else:
join_type = Fjoin
else:
res = self.estimate_card(star_tree.cardinality, subtree_j.cardinality)
if (star_tree.cardinality / float(subtree_j.cardinality) < 0.30) or (subtree_j.cardinality > 100*1000 and star_tree.cardinality < 100*1000) or (subtree_j.cardinality < 100*5):
join_type = Xnjoin
# If NLJ is placed, set res = 0/1 to force NLJs later
res = 1
else:
join_type = Fjoin
star_tree = LogicalPlan(star_tree, subtree_j, join_type)
star_tree.cardinality = res
# 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)
star_vars = subtree_i.variables
for j in range(0, len(stars)):
subtree_j = stars[j]
join_variables = set(star_vars).intersection(subtree_j.variables)
# Case: There is a join between stars.
if len(join_variables) > 0:
stars.pop(j)
res = self.estimate_card(star_tree.cardinality, subtree_j.cardinality)
# Place physical operators between stars.
if subtree_j.is_triple_pattern:
# This case models a satellite, therefore apply cardinality estimation.
if subtree_i.cardinality < (subtree_j.cardinality / 100.0):
join_type = Xnjoin
else:
join_type = Fjoin
else:
res = (subtree_i.cardinality + subtree_j.cardinality) / 2
join_type = Fjoin
star_tree = LogicalPlan(subtree_i, subtree_j, join_type)
star_tree.cardinality = res
stars.append(star_tree)
break
tree = stars.pop()
return tree
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
query_str = open(
"/Users/larsheling/Documents/Development/crop.nosync/queries/fedbench/LD9.rq"
).read()
query_parsed = parse_new(query_str)
fhandler = logging.FileHandler('../../logs/{}.log'.format(queryname), 'w')
fhandler.setLevel(logging.INFO)
logger.addHandler(fhandler)
tps = get_tps(query_parsed.body.triples)
id2tp = {}
index = 0
for tp in tps:
if isinstance(tp, TriplePattern):
get_metadata(sources, tp)
id2tp[index] = tp
index += 1
print("Got Metadata")
tp_sources = set()
graph = {}
for i in range(index):
for j in range(i, index):
if i != j:
a = id2tp[i]
b = id2tp[j]
a_just_tp = True if False in [
s_i.startswith("sparql@") for s_i in a.sources.keys()
] else False
def create_plan(self, query):
triple_patterns = list(query.where.left.triple_patterns)
for triple_pattern in triple_patterns:
get_metadata(self.source, triple_pattern)
return self.create_best(triple_patterns, query)