Many resources are needed to download a project. Please understand that we have to compensate our server costs. Thank you in advance. Project price only 1 $
You can buy this project and download/modify it how often you want.
/* This file is part of VoltDB.
* Copyright (C) 2008-2018 VoltDB Inc.
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU Affero General Public License as
* published by the Free Software Foundation, either version 3 of the
* License, or (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU Affero General Public License for more details.
*
* You should have received a copy of the GNU Affero General Public License
* along with VoltDB. If not, see .
*/
package org.voltdb.plannodes;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.Collection;
import java.util.HashMap;
import java.util.HashSet;
import java.util.Iterator;
import java.util.LinkedHashMap;
import java.util.List;
import java.util.Map;
import java.util.Map.Entry;
import java.util.Set;
import java.util.TreeMap;
import java.util.regex.Matcher;
import java.util.regex.Pattern;
import org.json_voltpatches.JSONArray;
import org.json_voltpatches.JSONException;
import org.json_voltpatches.JSONObject;
import org.json_voltpatches.JSONString;
import org.json_voltpatches.JSONStringer;
import org.voltdb.catalog.Database;
import org.voltdb.compiler.DatabaseEstimates;
import org.voltdb.compiler.ScalarValueHints;
import org.voltdb.expressions.AbstractExpression;
import org.voltdb.expressions.AbstractSubqueryExpression;
import org.voltdb.expressions.TupleValueExpression;
import org.voltdb.planner.PlanStatistics;
import org.voltdb.planner.PlanningErrorException;
import org.voltdb.planner.StatsField;
import org.voltdb.planner.parseinfo.StmtTableScan;
import org.voltdb.planner.parseinfo.StmtTargetTableScan;
import org.voltdb.types.PlanNodeType;
import org.voltdb.types.SortDirectionType;
public abstract class AbstractPlanNode implements JSONString, Comparable {
/**
* Internal PlanNodeId counter. Note that this member is static, which means
* all PlanNodes will have a unique id
*/
private static int NEXT_PLAN_NODE_ID = 1;
// Keep this flag turned off in production or when testing user-accessible EXPLAIN output or when
// using EXPLAIN output to validate plans.
protected static boolean m_verboseExplainForDebugging = false; // CODE REVIEWER! KEEP false on master!
public static boolean enableVerboseExplainForDebugging()
{
boolean was = m_verboseExplainForDebugging;
m_verboseExplainForDebugging = true;
return was;
}
public static boolean disableVerboseExplainForDebugging()
{
boolean was = m_verboseExplainForDebugging;
m_verboseExplainForDebugging = false;
return was;
}
public static void restoreVerboseExplainForDebugging(boolean was) { m_verboseExplainForDebugging = was; }
/*
* IDs only need to be unique for a single plan.
* Reset between plans
*/
public static final void resetPlanNodeIds() {
NEXT_PLAN_NODE_ID = 1;
}
public enum Members {
ID,
PLAN_NODE_TYPE,
INLINE_NODES,
CHILDREN_IDS,
PARENT_IDS,
OUTPUT_SCHEMA;
}
protected int m_id = -1;
protected List m_children = new ArrayList<>();
protected List m_parents = new ArrayList<>();
protected Set m_dominators = new HashSet<>();
// TODO: planner accesses this data directly. Should be protected.
protected List m_outputColumnHints = new ArrayList<>();
protected long m_estimatedOutputTupleCount = 0;
protected long m_estimatedProcessedTupleCount = 0;
protected boolean m_hasComputedEstimates = false;
// The output schema for this node
protected boolean m_hasSignificantOutputSchema;
protected NodeSchema m_outputSchema;
/**
* Some PlanNodes can take advantage of inline PlanNodes to perform
* certain additional tasks while performing their main operation, rather than
* having to re-read tuples from intermediate results
*/
protected Map m_inlineNodes =
new LinkedHashMap<>();
protected boolean m_isInline = false;
/**
* The textual explanation of why the plan may fail to have a deterministic result or effect when replayed.
*/
protected String m_nondeterminismDetail = "the query result does not guarantee a consistent ordering";
/**
* Instantiates a new plan node.
*/
protected AbstractPlanNode() {
m_id = NEXT_PLAN_NODE_ID++;
}
public int resetPlanNodeIds(int nextId) {
nextId = overrideId(nextId);
for (AbstractPlanNode inNode : getInlinePlanNodes().values()) {
// Inline nodes also need their ids to be overridden to make sure
// the subquery node ids are also globaly unique
nextId = inNode.resetPlanNodeIds(nextId);
}
for (int i = 0; i < getChildCount(); i++) {
AbstractPlanNode child = getChild(i);
assert(child != null);
nextId = child.resetPlanNodeIds(nextId);
}
return nextId;
}
public int overrideId(int newId) {
m_id = newId++;
// Override subqueries ids
Collection subqueries = findAllSubquerySubexpressions();
for (AbstractExpression expr : subqueries) {
assert(expr instanceof AbstractSubqueryExpression);
AbstractSubqueryExpression subquery = (AbstractSubqueryExpression) expr;
// overrideSubqueryNodeIds(newId) will get an NPE if the subquery
// has not been planned, presumably the effect of hitting a bug
// earlier in the planner. If that happens again, it MAY be useful
// to preempt those cases here and single-step through a replay of
// findAllSubquerySubexpressions. Determining where in the parent
// plan this subquery expression was found MAY provide a clue
// as to why the subquery was not planned. It has helped before.
//REDO to debug*/ if (subquery instanceof SelectSubqueryExpression) {
//REDO to debug*/ CompiledPlan subqueryPlan = ((SelectSubqueryExpression)subquery)
//REDO to debug*/ .getSubqueryScan().getBestCostPlan();
//REDO to debug*/ if (subqueryPlan == null) {
//REDO to debug*/ findAllSubquerySubexpressions();
//REDO to debug*/ }
//REDO to debug*/ }
newId = subquery.overrideSubqueryNodeIds(newId);
}
return newId;
}
/**
* Create a PlanNode that clones the configuration information but
* is not inserted in the plan graph and has a unique plan node id.
*/
protected void produceCopyForTransformation(AbstractPlanNode copy) {
copy.m_outputSchema = m_outputSchema;
copy.m_hasSignificantOutputSchema = m_hasSignificantOutputSchema;
copy.m_outputColumnHints = m_outputColumnHints;
copy.m_estimatedOutputTupleCount = m_estimatedOutputTupleCount;
copy.m_estimatedProcessedTupleCount = m_estimatedProcessedTupleCount;
// clone is not yet implemented for every node.
assert(m_inlineNodes.size() == 0);
assert(m_isInline == false);
// the api requires the copy is not (yet) connected
assert (copy.m_parents.size() == 0);
assert (copy.m_children.size() == 0);
}
public abstract PlanNodeType getPlanNodeType();
/**
* Generate the output schema for this node based on the
* output schemas of its children. The generated schema consists of
* the complete set of columns but is not yet ordered.
*
* Right now it's best to call this on every node after it gets added
* and linked to the top of the current plan graph.
* FIXME: "it's best to call this" means "to be on the paranoid safe side".
* It used to be that there was a hacky dependency in some non-critical aggregate code,
* so it would crash if generateOutputSchema had not been run earlier on its subtree.
* Historically, there may have been other dependencies like this, too,
* but they are mostly gone or otherwise completely avoidable.
* This means that one definitive depth-first recursive call that combines the effects
* of resolveColumnIndexes and then generateOutputSchema should suffice,
* if applied to a complete plan tree just before it gets fragmentized.
* The newest twist is that most of this repeated effort goes into generating
* redundant pass-thorugh structures that get ignored by the serializer.
*
* Many nodes will need to override this method in order to take whatever
* action is appropriate (so, joins will combine two schemas, projections
* will already have schemas defined and do nothing, etc).
* They should set m_hasSignificantOutputSchema to true so that the serialization knows
* not to ignore their work.
*
* @param db A reference to the Database object from the catalog.
*/
public void generateOutputSchema(Database db)
{
// default behavior: just copy the input schema
// to the output schema
assert(m_children.size() == 1);
AbstractPlanNode childNode = m_children.get(0);
childNode.generateOutputSchema(db);
// Replace the expressions in our children's columns with TVEs. When
// we resolve the indexes in these TVEs they will point back at the
// correct input column, which we are assuming that the child node
// has filled in with whatever expression was here before the replacement.
// Output schemas defined using this standard algorithm
// are just cached "fillers" that satisfy the legacy
// resolveColumnIndexes/generateOutputSchema/getOutputSchema protocol
// until it can be fixed up -- see the FIXME comment on generateOutputSchema.
m_hasSignificantOutputSchema = false;
m_outputSchema = childNode.getOutputSchema().copyAndReplaceWithTVE();
}
/**
* Recursively iterate through the plan and resolve the column_idx value for
* every TupleValueExpression in every AbstractExpression in every PlanNode.
* Few enough common cases so we force every AbstractPlanNode subclass to
* implement this. After index resolution, this method also sorts
* the columns in the output schema appropriately, depending upon what
* sort of node it is, so that its parent will be able to resolve
* its indexes successfully.
*
* Should get called on the plan graph after any optimizations but before
* the plan gets fragmented.
* FIXME: This needs to be reworked with generateOutputSchema to eliminate redundancies.
*/
public abstract void resolveColumnIndexes();
protected void resolveSubqueryColumnIndexes() {
// Possible subquery expressions
Collection exprs = findAllSubquerySubexpressions();
for (AbstractExpression expr: exprs) {
((AbstractSubqueryExpression) expr).resolveColumnIndexes();
}
}
public void validate() throws Exception {
//
// Make sure our children have us listed as their parents
//
for (AbstractPlanNode child : m_children) {
if (!child.m_parents.contains(this)) {
throw new Exception("ERROR: The child PlanNode '" + child.toString() + "' does not " +
"have its parent PlanNode '" + toString() + "' in its parents list");
}
child.validate();
}
//
// Inline PlanNodes
//
if (!m_inlineNodes.isEmpty()) {
for (AbstractPlanNode node : m_inlineNodes.values()) {
//
// Make sure that we're not attached to some kind of tree somewhere...
//
if (!node.m_children.isEmpty()) {
throw new Exception("ERROR: The inline PlanNode '" + node + "' has children inside of PlanNode '" + this + "'");
} else if (!node.m_parents.isEmpty()) {
throw new Exception("ERROR: The inline PlanNode '" + node + "' has parents inside of PlanNode '" + this + "'");
} else if (!node.isInline()) {
throw new Exception("ERROR: The inline PlanNode '" + node + "' was not marked as inline for PlanNode '" + this + "'");
} else if (!node.getInlinePlanNodes().isEmpty()) {
throw new Exception("ERROR: The inline PlanNode '" + node + "' has its own inline PlanNodes inside of PlanNode '" + this + "'");
}
node.validate();
}
}
}
public boolean hasReplicatedResult()
{
Map tablesRead = new TreeMap<>();
getTablesAndIndexes(tablesRead, null);
for (StmtTableScan tableScan : tablesRead.values()) {
if ( ! tableScan.getIsReplicated()) {
return false;
}
}
return true;
}
/**
* Recursively build sets of read tables read and index names used.
*
* @param tablesRead Set of table aliases read potentially added to at each recursive level.
* @param indexes Set of index names used in the plan tree
* Only the current fragment is of interest.
*/
public void getTablesAndIndexes(Map tablesRead,
Collection indexes)
{
for (AbstractPlanNode node : m_inlineNodes.values()) {
node.getTablesAndIndexes(tablesRead, indexes);
}
for (AbstractPlanNode node : m_children) {
node.getTablesAndIndexes(tablesRead, indexes);
}
getTablesAndIndexesFromSubqueries(tablesRead, indexes);
}
/**
* Collect read tables read and index names used in the current node subquery expressions.
*
* @param tablesRead Set of table aliases read potentially added to at each recursive level.
* @param indexes Set of index names used in the plan tree
* Only the current node is of interest.
*/
protected void getTablesAndIndexesFromSubqueries(Map tablesRead,
Collection indexes) {
for(AbstractExpression expr : findAllSubquerySubexpressions()) {
assert(expr instanceof AbstractSubqueryExpression);
AbstractSubqueryExpression subquery = (AbstractSubqueryExpression) expr;
AbstractPlanNode subqueryNode = subquery.getSubqueryNode();
assert(subqueryNode != null);
subqueryNode.getTablesAndIndexes(tablesRead, indexes);
}
}
/**
* Recursively find the target table name for a DML statement.
* The name will be attached to the AbstractOperationNode child
* of a Send Node, in all cases, so the "recursion" can be very limited.
* Most plan nodes can quickly stub out this recursion and return null.
* @return
*/
@SuppressWarnings("static-method")
public String getUpdatedTable() {
return null;
}
/**
* Does the (sub)plan guarantee an identical result/effect when "replayed"
* against the same database state, such as during replication or CL recovery.
* @return
*/
public boolean isOrderDeterministic() {
// Leaf nodes need to re-implement this test.
assert(m_children != null);
for (AbstractPlanNode child : m_children) {
if (! child.isOrderDeterministic()) {
m_nondeterminismDetail = child.m_nondeterminismDetail;
return false;
}
}
return true;
}
/**
* Does the plan guarantee a result sorted according to the required sort order.
* The default implementation delegates the question to its child if there is only one child.
*
*@param sortExpressions list of ordering columns expressions
*@param sortDirections list of corresponding sort orders
*
* @return TRUE if the node's output table is ordered. FALSE otherwise
*/
public boolean isOutputOrdered (List sortExpressions, List sortDirections) {
assert(sortExpressions.size() == sortDirections.size());
if (m_children.size() == 1) {
return m_children.get(0).isOutputOrdered(sortExpressions, sortDirections);
}
return false;
}
/**
* Accessor for description of plan non-determinism.
* @return the field
*/
public String nondeterminismDetail() {
return m_nondeterminismDetail;
}
@Override
public final String toString() {
return getPlanNodeType() + "[" + m_id + "]";
}
/**
* Called to compute cost estimates and statistics on a plan graph. Computing of the costs
* should be idempotent, but updating the PlanStatistics instance isn't, so this should
* be called once per finished graph, and once per PlanStatistics instance.
* TODO(XIN): It takes at least 14% planner CPU. Optimize it.
*/
public final void computeEstimatesRecursively(PlanStatistics stats,
DatabaseEstimates estimates,
ScalarValueHints[] paramHints)
{
assert(stats != null);
m_outputColumnHints.clear();
m_estimatedOutputTupleCount = 0;
// recursively compute and collect stats from children
long childOutputTupleCountEstimate = 0;
for (AbstractPlanNode child : m_children) {
child.computeEstimatesRecursively(stats, estimates, paramHints);
m_outputColumnHints.addAll(child.m_outputColumnHints);
childOutputTupleCountEstimate += child.m_estimatedOutputTupleCount;
}
// make sure any inlined scans (for NLIJ mostly) are costed as well
for (Entry entry : m_inlineNodes.entrySet()) {
AbstractPlanNode inlineNode = entry.getValue();
if (inlineNode instanceof AbstractScanPlanNode) {
inlineNode.computeCostEstimates(0, estimates, paramHints);
}
}
computeCostEstimates(childOutputTupleCountEstimate, estimates, paramHints);
stats.incrementStatistic(0, StatsField.TUPLES_READ, m_estimatedProcessedTupleCount);
}
/**
* Given the number of tuples expected as input to this node, compute an estimate
* of the number of tuples read/processed and the number of tuples output.
* This will be called by
* {@see AbstractPlanNode#computeEstimatesRecursively(PlanStatistics, Cluster, Database, DatabaseEstimates, ScalarValueHints[])}.
*/
protected void computeCostEstimates(long childOutputTupleCountEstimate,
DatabaseEstimates estimates,
ScalarValueHints[] paramHints)
{
m_estimatedOutputTupleCount = childOutputTupleCountEstimate;
m_estimatedProcessedTupleCount = childOutputTupleCountEstimate;
}
public long getEstimatedOutputTupleCount() {
return m_estimatedOutputTupleCount;
}
public long getEstimatedProcessedTupleCount() {
return m_estimatedProcessedTupleCount;
}
/**
* Gets the id.
*
* @return the id
*/
public Integer getPlanNodeId() {
return m_id;
}
/**
* Get this PlanNode's output schema
* FIXME: This needs to be reworked with generateOutputSchema to eliminate redundancies.
* In short, if generateOutputSchema was called definitively ONCE and returned the child's
* effective outputSchema to its parent -- possibly without even caching it as m_outputSchema,
* m_outputSchema could be used to cache only significant non-redundant output schemas.
* For now, the m_hasSignificantOutputSchema flag is checked separately to determine whether
* m_outputSchema is worth looking at.
* @return the NodeSchema which represents this node's output schema
*/
public NodeSchema getOutputSchema()
{
return m_outputSchema;
}
/**
* If the output schema of this node is a cheap copy of
* some other node's schema, and we decide to change the
* output true output schema, we can change this to the
* updated schema.
*
* @param childSchema
*/
private void setOutputSchema(NodeSchema childSchema) {
assert( ! m_hasSignificantOutputSchema);
m_outputSchema = childSchema;
}
/**
* Find the true output schema. This may be in some child
* node. This seems to be the search order when constructing
* a plan node in the EE.
*
* There are several cases.
* 1.) If the child has an output schema, and if it is
* not a copy of one of its children's schemas,
* that's the one we want. We know it's a copy
* if m_hasSignificantOutputSchema is false.
* 2.) If the child has no significant output schema but it
* has an inline projection node, then
* a.) If it does not have an inline insert
* node then the output schema of the child is
* the output schema of the inline projection node.
* b.) If the output schema has an inline insert node
* then the output schema is the usual DML output
* schema, which will be the schema of the inline
* insert node. I don't think we will ever see.
* this case in this function. This function is
* only called from the microoptimizer to remove
* projection nodes. So we don't see a projection
* node on top of a node with an inlined insert node.
* 3.) Otherwise, the output schema is the output schema
* of the child's first child. We should be able to
* follow the first children until we get something
* usable.
*
* Just for the record, if the child node has an inline
* insert and a projection node, the projection node's
* output schema is the schema of the tuples we will be
* inserting into the target table. The output schema of
* the child node will be the output schema of the insert
* node, which will be the usual DML schema. This has one
* long integer column counting the number of rows inserted.
*
* @param resetBack If this is true, we walk back the
* chain of parent plan nodes, updating
* the output schema in ancestor nodes
* with the true output schema. If we
* somehow change the true output schema
* we want to be able to change all the
* ones which are copies of the true
* output schema.
* @return The true output schema. This will never return null.
*/
public final NodeSchema getTrueOutputSchema(boolean resetBack) throws PlanningErrorException {
AbstractPlanNode child;
NodeSchema answer = null;
//
// Note: This code is translated from the C++ code in
// AbstractPlanNode::getOutputSchema. It's considerably
// different there, but I think this has the corner
// cases covered correctly.
for (child = this;
child != null;
child = (child.getChildCount() == 0) ? null : child.getChild(0)) {
NodeSchema childSchema;
if (child.m_hasSignificantOutputSchema) {
childSchema = child.getOutputSchema();
assert(childSchema != null);
answer = childSchema;
break;
}
AbstractPlanNode childProj = child.getInlinePlanNode(PlanNodeType.PROJECTION);
if (childProj != null) {
AbstractPlanNode schemaSrc = null;
AbstractPlanNode inlineInsertNode = childProj.getInlinePlanNode(PlanNodeType.INSERT);
if (inlineInsertNode != null) {
schemaSrc = inlineInsertNode;
} else {
schemaSrc = childProj;
}
childSchema = schemaSrc.getOutputSchema();
if (childSchema != null) {
answer = childSchema;
break;
}
}
}
if (child == null) {
// We've gone to the end of the plan. This is a
// failure in the EE.
assert(false);
throw new PlanningErrorException("AbstractPlanNode with no true output schema. Please notify VoltDB Support.");
}
// Trace back the chain of parents and reset the
// output schemas of the parent. These will all be
// exactly the same. Note that the source of the
// schema may be an inline plan node. So we need
// to set the child's output schema to be the answer.
// If the schema source is the child node itself, this will
// set the the output schema to itself, so no harm
// will be done.
if (resetBack) {
do {
if (child instanceof AbstractJoinPlanNode) {
// In joins with inlined aggregation, the inlined
// aggregate node is the one that determines the schema.
// (However, the enclosing join node still has its
// "m_hasSignificantOutputSchema" bit set.)
//
// The method resolveColumnIndexes will overwrite
// a join node's schema if there is aggregation. In order
// to avoid undoing the work we've done here, we must
// also update the inlined aggregate node.
AggregatePlanNode aggNode = AggregatePlanNode.getInlineAggregationNode(child);
if (aggNode != null) {
aggNode.setOutputSchema(answer);
}
}
if (! child.m_hasSignificantOutputSchema) {
child.setOutputSchema(answer);
}
child = (child.getParentCount() == 0) ? null : child.getParent(0);
} while (child != null);
}
return answer;
}
/**
* Add a child and link this node child's parent.
* @param child The node to add.
*/
public void addAndLinkChild(AbstractPlanNode child) {
assert(child != null);
m_children.add(child);
child.m_parents.add(this);
}
/**
* Used to re-link the child without changing the order.
*
* This is called by PushDownLimit and RemoveUnnecessaryProjectNodes.
* @param index
* @param child
*/
public void setAndLinkChild(int index, AbstractPlanNode child) {
assert(child != null);
m_children.set(index, child);
child.m_parents.add(this);
}
/** Remove child from this node.
* @param child to remove.
*/
public void unlinkChild(AbstractPlanNode child) {
assert(child != null);
m_children.remove(child);
child.m_parents.remove(this);
}
/**
* Replace an existing child with a new one preserving the child's position.
* @param oldChild The node to replace.
* @param newChild The new node.
* @return true if the child was replaced
*/
public boolean replaceChild(AbstractPlanNode oldChild, AbstractPlanNode newChild) {
assert(oldChild != null);
assert(newChild != null);
int idx = 0;
for (AbstractPlanNode child : m_children) {
if (child.equals(oldChild)) {
oldChild.m_parents.clear();
setAndLinkChild(idx, newChild);
return true;
}
++idx;
}
return false;
}
public void replaceChild(int oldChildIdx, AbstractPlanNode newChild) {
assert(oldChildIdx >= 0 && oldChildIdx < getChildCount());
assert(newChild != null);
AbstractPlanNode oldChild = m_children.get(oldChildIdx);
assert(oldChild != null);
oldChild.m_parents.clear();
setAndLinkChild(oldChildIdx, newChild);
}
/**
* Gets the children.
* @return the children
*/
public int getChildCount() {
return m_children.size();
}
/**
* @param index
* @return The child node of this node at a given index or null if none exists.
*/
public AbstractPlanNode getChild(int index) {
return m_children.get(index);
}
public void clearChildren() {
m_children.clear();
}
public boolean hasChild(AbstractPlanNode receive) {
return m_children.contains(receive);
}
/**
* Gets the parents.
* @return the parents
*/
public int getParentCount() {
return m_parents.size();
}
public AbstractPlanNode getParent(int index) {
return m_parents.get(index);
}
public void clearParents() {
m_parents.clear();
}
public void removeFromGraph() {
disconnectParents();
disconnectChildren();
}
public void disconnectParents() {
for (AbstractPlanNode parent : m_parents)
parent.m_children.remove(this);
m_parents.clear();
}
public void disconnectChildren() {
for (AbstractPlanNode child : m_children)
child.m_parents.remove(this);
m_children.clear();
}
/** Interject the provided node between this node and this node's current children */
public void addIntermediary(AbstractPlanNode node) {
// transfer this node's children to node
Iterator it = m_children.iterator();
while (it.hasNext()) {
AbstractPlanNode child = it.next();
it.remove(); // remove this.child from m_children
assert child.getParentCount() == 1;
child.clearParents(); // and reset child's parents list
node.addAndLinkChild(child); // set node.child and child.parent
}
// and add node to this node's children
assert(m_children.size() == 0);
addAndLinkChild(node);
}
/**
* @return The map of inlined nodes.
*/
public Map getInlinePlanNodes() {
return m_inlineNodes;
}
/**
* @param node
*/
public void addInlinePlanNode(AbstractPlanNode node) {
node.m_isInline = true;
m_inlineNodes.put(node.getPlanNodeType(), node);
node.m_children.clear();
node.m_parents.clear();
}
/**
*
* @param type
*/
public void removeInlinePlanNode(PlanNodeType type) {
if (m_inlineNodes.containsKey(type)) {
m_inlineNodes.remove(type);
}
}
/**
*
* @param type
* @return An inlined node of the given type or null if none.
*/
public AbstractPlanNode getInlinePlanNode(PlanNodeType type) {
return m_inlineNodes.get(type);
}
/**
*
* @return Is this node inlined in another node.
*/
public Boolean isInline() {
return m_isInline;
}
public boolean isSubQuery() {
return false;
}
public boolean hasSubquery() {
if (isSubQuery()) {
return true;
}
for (AbstractPlanNode n : m_children) {
if (n.hasSubquery()) {
return true;
}
}
for (AbstractPlanNode inlined : m_inlineNodes.values()) {
if (inlined.hasSubquery()) {
return true;
}
}
return false;
}
/**
* Refer to the override implementation on NestLoopIndexJoin node.
* @param tableName
* @return whether this node has an inlined index scan node or not.
*/
public boolean hasInlinedIndexScanOfTable(String tableName) {
for (int i = 0; i < getChildCount(); i++) {
AbstractPlanNode child = getChild(i);
if (child.hasInlinedIndexScanOfTable(tableName) == true) {
return true;
}
}
return false;
}
/**
* @return the dominator list for a node
*/
public Set getDominators() {
return m_dominators;
}
/**
* Initialize a hashset for each node containing that node's dominators
* (the set of predecessors that *always* precede this node in a traversal
* of the plan-graph in reverse-execution order (from root to leaves)).
*/
public void calculateDominators() {
HashSet visited = new HashSet<>();
calculateDominators_recurse(visited);
}
private void calculateDominators_recurse(HashSet visited) {
if (visited.contains(this)) {
assert(false): "do not expect loops in plangraph.";
return;
}
visited.add(this);
m_dominators.clear();
m_dominators.add(this);
// find nodes that are in every parent's dominator set.
HashMap union = new HashMap<>();
for (AbstractPlanNode n : m_parents) {
for (AbstractPlanNode d : n.getDominators()) {
if (union.containsKey(d))
union.put(d, union.get(d) + 1);
else
union.put(d, 1);
}
}
for (AbstractPlanNode pd : union.keySet() ) {
if (union.get(pd) == m_parents.size())
m_dominators.add(pd);
}
for (AbstractPlanNode n : m_children)
n.calculateDominators_recurse(visited);
}
/**
* @param type plan node type to search for
* @return a list of nodes that are eventual successors of this node of the desired type
*/
public ArrayList findAllNodesOfType(PlanNodeType type) {
HashSet visited = new HashSet<>();
ArrayList collected = new ArrayList<>();
findAllNodesOfType_recurse(type, null, collected, visited);
return collected;
}
/**
* @param pnClass plan node class to search for
* @return a list of nodes that are eventual successors of this node of the desired class
*/
public ArrayList findAllNodesOfClass(Class< ? extends AbstractPlanNode> pnClass) {
HashSet visited = new HashSet<>();
ArrayList collected = new ArrayList<>();
findAllNodesOfType_recurse(null, pnClass, collected, visited);
return collected;
}
private void findAllNodesOfType_recurse(PlanNodeType type, Class< ? extends AbstractPlanNode> pnClass, ArrayList collected,
HashSet visited)
{
if (visited.contains(this)) {
assert(false): "do not expect loops in plangraph.";
return;
}
visited.add(this);
if (getPlanNodeType() == type) {
collected.add(this);
} else if (pnClass != null && pnClass.isAssignableFrom(getClass())) {
collected.add(this);
}
for (AbstractPlanNode child : m_children)
child.findAllNodesOfType_recurse(type, pnClass, collected, visited);
for (AbstractPlanNode inlined : m_inlineNodes.values())
inlined.findAllNodesOfType_recurse(type, pnClass, collected, visited);
}
final public Collection findAllSubquerySubexpressions() {
Set collected = new HashSet<>();
findAllExpressionsOfClass(AbstractSubqueryExpression.class, collected);
return collected;
}
/**
* Collect a unique list of expressions of a given type that this node has including its inlined nodes
* @param aeClass AbstractExpression class to search for
* @param collection set to populate with expressions that this node has
*/
protected void findAllExpressionsOfClass(Class< ? extends AbstractExpression> aeClass,
Set collected) {
// Check the inlined plan nodes
for (AbstractPlanNode inlineNode: getInlinePlanNodes().values()) {
// For inline node we MUST go recursive to its children!!!!!
inlineNode.findAllExpressionsOfClass(aeClass, collected);
}
// and the output column expressions if there were no projection
NodeSchema schema = getOutputSchema();
if (schema != null) {
schema.addAllSubexpressionsOfClassFromNodeSchema(collected, aeClass);
}
}
/**
* @param type plan node type to search for
* @return whether a node of that type is contained in the plan tree
*/
public boolean hasAnyNodeOfType(PlanNodeType type) {
if (getPlanNodeType() == type) {
return true;
}
for (AbstractPlanNode n : m_children) {
if (n.hasAnyNodeOfType(type)) {
return true;
}
}
for (AbstractPlanNode inlined : m_inlineNodes.values()) {
if (inlined.hasAnyNodeOfType(type)) {
return true;
}
}
return false;
}
/**
* @param pnClass plan node class to search for
* @return whether a node of that type is contained in the plan tree
*/
public boolean hasAnyNodeOfClass(Class< ? extends AbstractPlanNode> pnClass) {
if (pnClass.isAssignableFrom(getClass())) {
return true;
}
for (AbstractPlanNode n : m_children) {
if (n.hasAnyNodeOfClass(pnClass)) {
return true;
}
}
for (AbstractPlanNode inlined : m_inlineNodes.values()) {
if (inlined.hasAnyNodeOfClass(pnClass)) {
return true;
}
}
return false;
}
@Override
public int compareTo(AbstractPlanNode other) {
int diff = 0;
// compare child nodes
HashMap nodesById = new HashMap<>();
for (AbstractPlanNode node : m_children)
nodesById.put(node.getPlanNodeId(), node);
for (AbstractPlanNode node : other.m_children) {
AbstractPlanNode myNode = nodesById.get(node.getPlanNodeId());
diff = myNode.compareTo(node);
if (diff != 0) return diff;
}
// compare inline nodes
HashMap> inlineNodesById =
new HashMap<>();
for (Entry e : m_inlineNodes.entrySet())
inlineNodesById.put(e.getValue().getPlanNodeId(), e);
for (Entry e : other.m_inlineNodes.entrySet()) {
Entry myE = inlineNodesById.get(e.getValue().getPlanNodeId());
if (myE.getKey() != e.getKey()) return -1;
diff = myE.getValue().compareTo(e.getValue());
if (diff != 0) return diff;
}
diff = m_id - other.m_id;
return diff;
}
// produce a file that can imported into graphviz for easier visualization
public String toDOTString() {
StringBuilder sb = new StringBuilder();
// id [label=id: value-type ];
// id -> child_id;
// id -> child_id;
sb.append(m_id).append(" [label=\"").append(m_id).append(": ").append(getPlanNodeType()).append("\" ");
sb.append(getValueTypeDotString());
sb.append("];\n");
for (AbstractPlanNode node : m_inlineNodes.values()) {
sb.append(m_id).append(" -> ").append(node.getPlanNodeId().intValue()).append(";\n");
sb.append(node.toDOTString());
}
for (AbstractPlanNode node : m_children) {
sb.append(m_id).append(" -> ").append(node.getPlanNodeId().intValue()).append(";\n");
}
return sb.toString();
}
// maybe not worth polluting
private String getValueTypeDotString() {
PlanNodeType pnt = getPlanNodeType();
if (isInline()) {
return "fontcolor=\"white\" style=\"filled\" fillcolor=\"red\"";
}
if (pnt == PlanNodeType.SEND || pnt == PlanNodeType.RECEIVE || pnt == PlanNodeType.MERGERECEIVE) {
return "fontcolor=\"white\" style=\"filled\" fillcolor=\"black\"";
}
return "";
}
@Override
public String toJSONString() {
JSONStringer stringer = new JSONStringer();
try
{
stringer.object();
toJSONString(stringer);
stringer.endObject();
}
catch (JSONException e)
{
e.printStackTrace();
throw new RuntimeException(e.getMessage(), e);
}
return stringer.toString();
}
public void toJSONString(JSONStringer stringer) throws JSONException {
stringer.keySymbolValuePair(Members.ID.name(), m_id);
stringer.keySymbolValuePair(Members.PLAN_NODE_TYPE.name(), getPlanNodeType().toString());
if (m_inlineNodes.size() > 0) {
PlanNodeType types[] = new PlanNodeType[m_inlineNodes.size()];
int i = 0;
for (PlanNodeType type : m_inlineNodes.keySet()) {
types[i++] = type;
}
Arrays.sort(types);
stringer.key(Members.INLINE_NODES.name()).array();
for (PlanNodeType type : types) {
AbstractPlanNode node = m_inlineNodes.get(type);
assert(node != null);
stringer.value(node);
}
stringer.endArray();
}
if (m_children.size() > 0) {
stringer.key(Members.CHILDREN_IDS.name()).array();
for (AbstractPlanNode node : m_children) {
stringer.value(node.getPlanNodeId().intValue());
}
stringer.endArray();
}
outputSchemaToJSON(stringer);
}
private void outputSchemaToJSON(JSONStringer stringer) throws JSONException {
if (m_hasSignificantOutputSchema) {
stringer.key(Members.OUTPUT_SCHEMA.name());
stringer.array();
for (int colNo = 0; colNo < m_outputSchema.size(); colNo += 1) {
SchemaColumn column = m_outputSchema.getColumn(colNo);
column.toJSONString(stringer, true, colNo);
}
stringer.endArray();
}
}
public String toExplainPlanString() {
StringBuilder sb = new StringBuilder();
explainPlan_recurse(sb, "");
String fullExpalinString = sb.toString();
// Extract subqueries into a map to explain them separately. Each subquery is
// surrounded by the 'Subquery_[SubqueryId]' tags. Example:
// Subquery_1SEQUENTIAL SCAN of "R1"Subquery_1
Pattern subqueryPattern = Pattern.compile(
String.format("(%s)([0-9]+)(.*)(\\s*)%s(\\2)", AbstractSubqueryExpression.SUBQUERY_TAG, AbstractSubqueryExpression.SUBQUERY_TAG),
Pattern.DOTALL);
Map subqueries = new TreeMap<>();
String topStmt = extractExplainedSubquries(fullExpalinString, subqueryPattern, subqueries);
StringBuilder fullSb = new StringBuilder(topStmt);
for (Map.Entry subquery : subqueries.entrySet()) {
fullSb.append("\n").append(subquery.getKey()).append('\n').append(subquery.getValue());
}
return fullSb.toString();
}
private String extractExplainedSubquries(String explainedSubquery, Pattern pattern, Map subqueries) {
Matcher matcher = pattern.matcher(explainedSubquery);
int pos = 0;
StringBuilder sb = new StringBuilder();
// Find all the subqueries from the input string
while(matcher.find()) {
sb.append(explainedSubquery.substring(pos, matcher.end(2)));
pos = matcher.end();
// Recurse into the subquery string to extract its own subqueries if any
String nextExplainedStmt = extractExplainedSubquries(matcher.group(3), pattern, subqueries);
subqueries.put(AbstractSubqueryExpression.SUBQUERY_TAG + matcher.group(2), nextExplainedStmt);
}
// Append the rest of the input string
if (pos < explainedSubquery.length()) {
sb.append(explainedSubquery.substring(pos));
}
return sb.toString();
}
public void explainPlan_recurse(StringBuilder sb, String indent) {
String extraIndent = " ";
// Except when verbosely debugging,
// skip projection nodes basically (they're boring as all get out)
boolean skipCurrentNode = ! m_verboseExplainForDebugging
&& getPlanNodeType() == PlanNodeType.PROJECTION;
if (skipCurrentNode) {
extraIndent = "";
}
else {
if ( ! m_skipInitalIndentationForExplain) {
sb.append(indent);
}
String nodePlan = explainPlanForNode(indent);
sb.append(nodePlan);
if (m_verboseExplainForDebugging && m_outputSchema != null) {
sb.append(indent + " " + m_outputSchema.toExplainPlanString());
}
sb.append("\n");
}
// Agg < Proj < Limit < Scan
// Order the inline nodes with integer in ascending order
TreeMap sort_inlineNodes =
new TreeMap<>();
// every inline plan node is unique
int ii = 4;
for (AbstractPlanNode inlineNode : m_inlineNodes.values()) {
if (inlineNode instanceof AggregatePlanNode) {
sort_inlineNodes.put(0, inlineNode);
} else if (inlineNode instanceof ProjectionPlanNode) {
sort_inlineNodes.put(1, inlineNode);
} else if (inlineNode instanceof LimitPlanNode) {
sort_inlineNodes.put(2, inlineNode);
} else if (inlineNode instanceof AbstractScanPlanNode) {
sort_inlineNodes.put(3, inlineNode);
} else {
// any other inline nodes currently ? --xin
sort_inlineNodes.put(ii++, inlineNode);
}
}
// inline nodes with ascending order as their integer keys
for (AbstractPlanNode inlineNode : sort_inlineNodes.values()) {
// don't bother with inlined projections
if (( ! m_verboseExplainForDebugging) &&
(inlineNode.getPlanNodeType() == PlanNodeType.PROJECTION)) {
continue;
}
inlineNode.setSkipInitalIndentationForExplain(true);
sb.append(indent + extraIndent + "inline ");
inlineNode.explainPlan_recurse(sb, indent + extraIndent);
}
for (AbstractPlanNode node : m_children) {
// inline nodes shouldn't have children I hope
assert(m_isInline == false);
if (skipCurrentNode) {
// If the current node is skipped, I would like to pass the skip indentation
// flag on to the next level.
node.setSkipInitalIndentationForExplain(m_skipInitalIndentationForExplain);
}
node.explainPlan_recurse(sb, indent + extraIndent);
}
}
private boolean m_skipInitalIndentationForExplain = false;
public void setSkipInitalIndentationForExplain(boolean skip) {
m_skipInitalIndentationForExplain = skip;
}
protected abstract String explainPlanForNode(String indent);
public ArrayList getScanNodeList () {
HashSet visited = new HashSet<>();
ArrayList collected = new ArrayList<>();
getScanNodeList_recurse( collected, visited);
return collected;
}
//postorder adding scan nodes
public void getScanNodeList_recurse(ArrayList collected,
HashSet visited) {
if (visited.contains(this)) {
assert(false): "do not expect loops in plangraph.";
return;
}
visited.add(this);
for (AbstractPlanNode n : m_children) {
n.getScanNodeList_recurse(collected, visited);
}
for (AbstractPlanNode node : m_inlineNodes.values()) {
node.getScanNodeList_recurse(collected, visited);
}
}
public ArrayList getPlanNodeList () {
HashSet visited = new HashSet<>();
ArrayList collected = new ArrayList<>();
getPlanNodeList_recurse( collected, visited);
return collected;
}
//postorder add nodes
public void getPlanNodeList_recurse(ArrayList collected,
HashSet visited) {
if (visited.contains(this)) {
assert(false): "do not expect loops in plangraph.";
return;
}
visited.add(this);
for (AbstractPlanNode n : m_children) {
n.getPlanNodeList_recurse(collected, visited);
}
collected.add(this);
}
abstract protected void loadFromJSONObject(JSONObject obj, Database db)
throws JSONException;
protected static void loadBooleanArrayFromJSONObject(JSONObject jobj, String key, List target) throws JSONException {
if ( ! jobj.isNull(key)) {
JSONArray jarray = jobj.getJSONArray(key);
int numCols = jarray.length();
for (int ii = 0; ii < numCols; ++ii) {
target.add(jarray.getBoolean(ii));
}
}
}
protected static void booleanArrayToJSONString(JSONStringer stringer, String key, List array) throws JSONException {
stringer.key(key).array();
for (Boolean arrayElement : array) {
stringer.value(arrayElement);
}
stringer.endArray();
}
protected static NodeSchema loadSchemaFromJSONObject(JSONObject jobj,
String jsonKey) throws JSONException {
NodeSchema nodeSchema = new NodeSchema();
JSONArray jarray = jobj.getJSONArray(jsonKey);
int size = jarray.length();
for (int i = 0; i < size; ++i) {
nodeSchema.addColumn(SchemaColumn.fromJSONObject(
jarray.getJSONObject(i)) );
}
return nodeSchema;
}
protected final void helpLoadFromJSONObject(JSONObject jobj, Database db)
throws JSONException {
assert(jobj != null);
m_id = jobj.getInt(Members.ID.name());
JSONArray jarray = null;
//load inline nodes
if ( ! jobj.isNull(Members.INLINE_NODES.name())) {
jarray = jobj.getJSONArray( Members.INLINE_NODES.name() );
PlanNodeTree pnt = new PlanNodeTree();
pnt.loadPlanNodesFromJSONArrays(jarray, db);
List list = pnt.getNodeList();
for( AbstractPlanNode pn : list ) {
m_inlineNodes.put( pn.getPlanNodeType(), pn);
}
}
//children and parents list loading implemented in planNodeTree.loadFromJsonArray
// load the output schema if it was marked significant.
if ( ! jobj.isNull(Members.OUTPUT_SCHEMA.name())) {
m_hasSignificantOutputSchema = true;
m_outputSchema = loadSchemaFromJSONObject(jobj,
Members.OUTPUT_SCHEMA.name());
}
}
/**
* @param jobj
* @param key
* @return
* @throws JSONException
*/
List loadStringListMemberFromJSON(JSONObject jobj, String key)
throws JSONException {
if (jobj.isNull(key)) {
return null;
}
JSONArray jarray = jobj.getJSONArray(key);
int numElems = jarray.length();
List result = new ArrayList<>(numElems);
for (int ii = 0; ii < numElems; ++ii) {
result.add(jarray.getString(ii));
}
return result;
}
/**
* @param stringer
* @param key
* @param stringList
* @throws JSONException
*/
void toJSONStringArrayString(JSONStringer stringer, String key,
List stringList) throws JSONException {
stringer.key(key).array();
for (String elem : stringList) {
stringer.value(elem);
}
stringer.endArray();
}
/**
* @param jobj
* @param key
* @return
* @throws JSONException
*/
int[] loadIntArrayMemberFromJSON(JSONObject jobj, String key)
throws JSONException {
if (jobj.isNull(key)) {
return null;
}
JSONArray jarray = jobj.getJSONArray(key);
int numElems = jarray.length();
int[] result = new int[numElems];
for (int ii = 0; ii < numElems; ++ii) {
result[ii] = jarray.getInt(ii);
}
return result;
}
/**
* @param stringer
* @param key
* @param intArray
* @throws JSONException
*/
void toJSONIntArrayString(JSONStringer stringer, String key, int[] intArray) throws JSONException {
stringer.key(key).array();
for (int i : intArray) {
stringer.value(i);
}
stringer.endArray();
}
public boolean reattachFragment(AbstractPlanNode child) {
for (AbstractPlanNode pn : m_children) {
if (pn.reattachFragment(child)) {
return true;
}
}
return false;
}
public boolean planNodeClassNeedsProjectionNode() {
return true;
}
/**
* When a project node is added to the top of the plan, we need to adjust
* the differentiator field of TVEs to reflect differences in the scan
* schema vs the storage schema of a table, so that fields with duplicate names
* produced by expanding "SELECT *" can resolve correctly.
*
* We recurse until we find either a join node or a scan node.
*
* @param existing differentiator field of a TVE
* @return new differentiator value
*/
public void adjustDifferentiatorField(TupleValueExpression tve) {
assert (m_children.size() == 1);
m_children.get(0).adjustDifferentiatorField(tve);
}
}
