def pickJoinOrder(self, plan):
relations = plan.relations()
fieldDict = self.obtainFieldDict(plan)
(joinTablesDict, selectTablesDict) = self.getExprDicts(plan, fieldDict)
# makes dicts that maps a list of relations to exprs involving that list
# then in system R we will build opt(A,B) Join C using join exprs involving A,C and B,C
# and on top of it the select exprs that involve 2 tables A,C or B,C
isGroupBy = True if plan.root.operatorType() == "GroupBy" else False
outputSchema = plan.schema()
self.reportPlanCount = 0
worklist = []
for r in relations:
table = TableScan(r,self.db.relationSchema(r))
table.prepare(self.db)
if (r,) in selectTablesDict:
selectExprs = selectTablesDict[(r,)]
selectString = self.combineSelects(selectExprs)
select = Select(table,selectString)
select.prepare(self.db)
worklist.append(Plan(root=select))
else:
worklist.append(Plan(root=table))
while(len(worklist) > 1):
combos = itertools.combinations(worklist,2)
bestJoin = None
sourcePair = None
for pair in combos:
op1 = pair[0].root
op2 = pair[1].root
selectExpr = self.createExpression(pair[0].relations(), pair[1].relations(), selectTablesDict)
joinExpr = self.createExpression(pair[0].relations(), pair[1].relations(), joinTablesDict)
join1BnljOp = Join(op1, op2, expr=joinExpr, method="block-nested-loops" )
join2BnljOp = Join(op2, op1, expr=joinExpr, method="block-nested-loops" )
join1NljOp = Join(op1, op2, expr=joinExpr, method="nested-loops" )
join2NljOp = Join(op2, op1, expr=joinExpr, method="nested-loops" )
if selectExpr == "True":
full1BnljOp = join1BnljOp
full2BnljOp = join2BnljOp
full1NljOp = join1NljOp
full2NljOp = join2NljOp
else:
full1BnljOp = Select(join1BnljOp, selectExpr)
full2BnljOp = Select(join2BnljOp, selectExpr)
full1NljOp = Select(join1NljOp, selectExpr)
full2NljOp = Select(join2NljOp, selectExpr)
joinList = [full1BnljOp, full2BnljOp, full1NljOp, full2NljOp]
for j in joinList:
joinplan = Plan(root=j)
joinplan.prepare(self.db)
joinplan.sample(100)
if bestJoin == None or joinplan.cost(True) < bestJoin.cost(True):
bestJoin = joinplan
sourcePair = pair
self.reportPlanCount += 4
self.clearSampleFiles()
worklist.remove(sourcePair[0])
worklist.remove(sourcePair[1])
worklist.append(bestJoin)
# after System R algorithm
newPlan = worklist[0]
if isGroupBy:
newGroupBy = GroupBy(newPlan.root, groupSchema=plan.root.groupSchema, \
aggSchema=plan.root.aggSchema, groupExpr=plan.root.groupExpr, \
aggExprs=plan.root.aggExprs, \
groupHashFn=plan.root.groupHashFn)
newGroupBy.prepare(self.db)
newPlan = Plan(root=newGroupBy)
if set(outputSchema.schema()) != set(newPlan.schema().schema()):
projectDict = {}
for f, t in outputSchema.schema():
projectDict[f] = (f, t)
currRoot = newPlan.root
project = Project(currRoot, projectDict)
project.prepare(self.db)
newPlan = Plan(root=project)
return newPlan
def pickJoinOrder(self, plan):
relations = plan.relations()
fieldDict = self.obtainFieldDict(plan)
(joinTablesDict, selectTablesDict) = self.getExprDicts(plan, fieldDict)
# makes dicts that maps a list of relations to exprs involving that list
# then in system R we will build opt(A,B) Join C using join exprs involving A,C and B,C
# and on top of it the select exprs that involve 2 tables A,C or B,C
isGroupBy = True if plan.root.operatorType() == "GroupBy" else False
outputSchema = plan.schema()
self.reportPlanCount = 0
worklist = []
for r in relations:
table = TableScan(r, self.db.relationSchema(r))
table.prepare(self.db)
if (r, ) in selectTablesDict:
selectExprs = selectTablesDict[(r, )]
selectString = self.combineSelects(selectExprs)
select = Select(table, selectString)
select.prepare(self.db)
worklist.append(Plan(root=select))
else:
worklist.append(Plan(root=table))
while (len(worklist) > 1):
combos = itertools.combinations(worklist, 2)
bestJoin = None
sourcePair = None
for pair in combos:
op1 = pair[0].root
op2 = pair[1].root
selectExpr = self.createExpression(pair[0].relations(),
pair[1].relations(),
selectTablesDict)
joinExpr = self.createExpression(pair[0].relations(),
pair[1].relations(),
joinTablesDict)
join1BnljOp = Join(op1,
op2,
expr=joinExpr,
method="block-nested-loops")
join2BnljOp = Join(op2,
op1,
expr=joinExpr,
method="block-nested-loops")
join1NljOp = Join(op1,
op2,
expr=joinExpr,
method="nested-loops")
join2NljOp = Join(op2,
op1,
expr=joinExpr,
method="nested-loops")
if selectExpr == "True":
full1BnljOp = join1BnljOp
full2BnljOp = join2BnljOp
full1NljOp = join1NljOp
full2NljOp = join2NljOp
else:
full1BnljOp = Select(join1BnljOp, selectExpr)
full2BnljOp = Select(join2BnljOp, selectExpr)
full1NljOp = Select(join1NljOp, selectExpr)
full2NljOp = Select(join2NljOp, selectExpr)
joinList = [full1BnljOp, full2BnljOp, full1NljOp, full2NljOp]
for j in joinList:
joinplan = Plan(root=j)
joinplan.prepare(self.db)
joinplan.sample(100)
if bestJoin == None or joinplan.cost(True) < bestJoin.cost(
True):
bestJoin = joinplan
sourcePair = pair
self.reportPlanCount += 4
self.clearSampleFiles()
worklist.remove(sourcePair[0])
worklist.remove(sourcePair[1])
worklist.append(bestJoin)
# after System R algorithm
newPlan = worklist[0]
if isGroupBy:
newGroupBy = GroupBy(newPlan.root, groupSchema=plan.root.groupSchema, \
aggSchema=plan.root.aggSchema, groupExpr=plan.root.groupExpr, \
aggExprs=plan.root.aggExprs, \
groupHashFn=plan.root.groupHashFn)
newGroupBy.prepare(self.db)
newPlan = Plan(root=newGroupBy)
if set(outputSchema.schema()) != set(newPlan.schema().schema()):
projectDict = {}
for f, t in outputSchema.schema():
projectDict[f] = (f, t)
currRoot = newPlan.root
project = Project(currRoot, projectDict)
project.prepare(self.db)
newPlan = Plan(root=project)
return newPlan
def pickJoinOrder(self, plan):
relations = plan.relations()
fieldDict = self.obtainFieldDict(plan)
(joinTablesDict, selectTablesDict) = self.getExprDicts(plan, fieldDict)
# makes dicts that maps a list of relations to exprs involving that list
# then in system R we will build opt(A,B) Join C using join exprs involving A,C and B,C
# and on top of it the select exprs that involve 2 tables A,C or B,C
isGroupBy = True if plan.root.operatorType() == "GroupBy" else False
outputSchema = plan.schema()
optDict = {}
self.reportPlanCount = 0
for npass in range(1, len(relations) + 1):
if npass == 1:
for r in relations:
table = TableScan(r,self.db.relationSchema(r))
if (r,) in selectTablesDict:
selectExprs = selectTablesDict[(r,)]
selectString = self.combineSelects(selectExprs)
select = Select(table,selectString)
optDict[(r,)] = Plan(root=select)
else:
optDict[(r,)] = Plan(root=table)
self.reportPlanCount += 1
else:
combinations = itertools.combinations(relations,npass)
for c in combinations:
fullList = sorted(c)
clist = self.getCombos(fullList)
bestJoin = None
for subcombo in clist:
complement = self.getComplement(fullList, subcombo)
leftOps = optDict[tuple(complement)].root
rightOps = optDict[tuple(subcombo)].root
selectExpr = self.createExpression(complement, subcombo, selectTablesDict)
joinExpr = self.createExpression(complement, subcombo, joinTablesDict)
joinBnljOp = Join(leftOps, rightOps, expr=joinExpr, method="block-nested-loops" )
fullBnljOp = Select(joinBnljOp, selectExpr)
if selectExpr == "True":
joinBnlj = Plan(root=joinBnljOp)
else:
joinBnlj = Plan(root=fullBnljOp)
joinBnlj.prepare(self.db)
joinBnlj.sample(100)
joinNljOp = Join(leftOps, rightOps, expr=joinExpr, method="nested-loops" )
fullNljOp = Select(joinNljOp, selectExpr)
if selectExpr == "True":
joinNlj = Plan(root=joinNljOp)
else:
joinNlj = Plan(root=fullNljOp)
joinNlj.prepare(self.db)
joinNlj.sample(100)
if joinBnlj.cost(True) < joinNlj.cost(True):
if bestJoin == None or joinBnlj.cost(True) < bestJoin.cost(True):
bestJoin = joinBnlj
else:
if bestJoin == None or joinNlj.cost(True) < bestJoin.cost(True):
bestJoin = joinNlj
self.reportPlanCount += 2
self.clearSampleFiles()
optDict[tuple(fullList)] = bestJoin
# after System R algorithm
newPlan = optDict[tuple(sorted(relations))]
if isGroupBy:
newGroupBy = GroupBy(newPlan.root, groupSchema=plan.root.groupSchema, \
aggSchema=plan.root.aggSchema, groupExpr=plan.root.groupExpr, \
aggExprs=plan.root.aggExprs, \
groupHashFn=plan.root.groupHashFn)
newGroupBy.prepare(self.db)
newPlan = Plan(root=newGroupBy)
if set(outputSchema.schema()) != set(newPlan.schema().schema()):
projectDict = {}
for f, t in outputSchema.schema():
projectDict[f] = (f, t)
currRoot = newPlan.root
project = Project(currRoot, projectDict)
project.prepare(self.db)
newPlan = Plan(root=project)
return newPlan
def pickJoinOrder(self, plan):
relations = plan.relations()
fieldDict = self.obtainFieldDict(plan)
(joinTablesDict, selectTablesDict) = self.getExprDicts(plan, fieldDict)
# makes dicts that maps a list of relations to exprs involving that list
# then in system R we will build opt(A,B) Join C using join exprs involving A,C and B,C
# and on top of it the select exprs that involve 2 tables A,C or B,C
isGroupBy = True if plan.root.operatorType() == "GroupBy" else False
outputSchema = plan.schema()
optDict = {}
self.reportPlanCount = 0
for npass in range(1, len(relations) + 1):
if npass == 1:
for r in relations:
table = TableScan(r, self.db.relationSchema(r))
if (r, ) in selectTablesDict:
selectExprs = selectTablesDict[(r, )]
selectString = self.combineSelects(selectExprs)
select = Select(table, selectString)
optDict[(r, )] = Plan(root=select)
else:
optDict[(r, )] = Plan(root=table)
self.reportPlanCount += 1
else:
combinations = itertools.combinations(relations, npass)
for c in combinations:
fullList = sorted(c)
clist = self.getCombos(fullList)
bestJoin = None
for subcombo in clist:
complement = self.getComplement(fullList, subcombo)
leftOps = optDict[tuple(complement)].root
rightOps = optDict[tuple(subcombo)].root
selectExpr = self.createExpression(
complement, subcombo, selectTablesDict)
joinExpr = self.createExpression(
complement, subcombo, joinTablesDict)
joinBnljOp = Join(leftOps,
rightOps,
expr=joinExpr,
method="block-nested-loops")
fullBnljOp = Select(joinBnljOp, selectExpr)
if selectExpr == "True":
joinBnlj = Plan(root=joinBnljOp)
else:
joinBnlj = Plan(root=fullBnljOp)
joinBnlj.prepare(self.db)
joinBnlj.sample(100)
joinNljOp = Join(leftOps,
rightOps,
expr=joinExpr,
method="nested-loops")
fullNljOp = Select(joinNljOp, selectExpr)
if selectExpr == "True":
joinNlj = Plan(root=joinNljOp)
else:
joinNlj = Plan(root=fullNljOp)
joinNlj.prepare(self.db)
joinNlj.sample(100)
if joinBnlj.cost(True) < joinNlj.cost(True):
if bestJoin == None or joinBnlj.cost(
True) < bestJoin.cost(True):
bestJoin = joinBnlj
else:
if bestJoin == None or joinNlj.cost(
True) < bestJoin.cost(True):
bestJoin = joinNlj
self.reportPlanCount += 2
self.clearSampleFiles()
optDict[tuple(fullList)] = bestJoin
# after System R algorithm
newPlan = optDict[tuple(sorted(relations))]
if isGroupBy:
newGroupBy = GroupBy(newPlan.root, groupSchema=plan.root.groupSchema, \
aggSchema=plan.root.aggSchema, groupExpr=plan.root.groupExpr, \
aggExprs=plan.root.aggExprs, \
groupHashFn=plan.root.groupHashFn)
newGroupBy.prepare(self.db)
newPlan = Plan(root=newGroupBy)
if set(outputSchema.schema()) != set(newPlan.schema().schema()):
projectDict = {}
for f, t in outputSchema.schema():
projectDict[f] = (f, t)
currRoot = newPlan.root
project = Project(currRoot, projectDict)
project.prepare(self.db)
newPlan = Plan(root=project)
return newPlan
