public static void computeStatistics() {
Iterator<Integer> tableIt = Database.getCatalog().tableIdIterator();
System.out.println("Computing table stats.");
while (tableIt.hasNext()) {
int tableid = tableIt.next();
TableStats s = new TableStats(tableid, IOCOSTPERPAGE);
setTableStats(Database.getCatalog().getTableName(tableid), s);
}
System.out.println("Done.");
}
/**
* Returns the TupleDesc with field names from the underlying HeapFile, prefixed with the
* tableAlias string from the constructor. This prefix becomes useful when joining tables
* containing a field(s) with the same name.
*
* @return the TupleDesc with field names from the underlying HeapFile, prefixed with the
* tableAlias string from the constructor.
*/
public TupleDesc getTupleDesc() {
TupleDesc tup = Database.getCatalog().getTupleDesc(id);
int length = tup.numFields();
String[] field = new String[length];
Type[] types = new Type[length];
for (int i = 0; i < length; i++) {
types[i] = tup.getFieldType(i);
field[i] = alias + "." + tup.getFieldName(i);
}
return new TupleDesc(types, field);
}
public void open() throws DbException, TransactionAbortedException {
// some code goes here
if (i_pos != null) {
i = i_pos;
} else {
Catalog gc = Database.getCatalog();
HeapFile file = (HeapFile) gc.getDbFile(tableid);
i = file.iterator(tid);
}
i.open();
}
/**
* Create a new TableStats object, that keeps track of statistics on each column of a table
*
* @param tableid The table over which to compute statistics
* @param ioCostPerPage The cost per page of IO. This doesn't differentiate between
* sequential-scan IO and disk seeks.
*/
public TableStats(int tableid, int ioCostPerPage) {
// For this function, we use the DbFile for the table in question,
// then scan through its tuples and calculate the values that you
// to build the histograms.
// TODO: Fill out the rest of the constructor.
// Feel free to change anything already written, it's only a guideline
this.ioCostPerPage = ioCostPerPage;
DbFile file = Database.getCatalog().getDbFile(tableid);
tupleDesc = file.getTupleDesc();
numPages = ((HeapFile) file).numPages();
numTuples = 0;
int numFields = tupleDesc.numFields();
// TODO: what goes here?
statistics = new ArrayList<Object>();
for (int i = 0; i < numFields; i++) {
if (Type.INT_TYPE.equals(tupleDesc.getFieldType(i))) {
statistics.add(new IntStatistics(NUM_HIST_BINS));
} else {
statistics.add(new StringHistogram(NUM_HIST_BINS));
}
}
final DbFileIterator iter = file.iterator(null);
try {
iter.open();
while (iter.hasNext()) {
Tuple t = iter.next();
numTuples++;
// TODO: and here?
for (int i = 0; i < numFields; i++) {
if (Type.INT_TYPE.equals(tupleDesc.getFieldType(i))) {
((IntStatistics) statistics.get(i)).addValue(((IntField) t.getField(i)).getValue());
} else {
((StringHistogram) statistics.get(i))
.addValue(((StringField) t.getField(i)).getValue());
}
}
}
iter.close();
} catch (DbException e) {
e.printStackTrace();
} catch (TransactionAbortedException e) {
e.printStackTrace();
}
}
private void createAliasedTd() {
Catalog gc = Database.getCatalog();
TupleDesc old_td = gc.getTupleDesc(tableid);
String[] newFieldAr = new String[old_td.numFields()];
Type[] typeAr = new Type[old_td.numFields()];
String field = null;
for (int i = 0; i < newFieldAr.length; i++) {
field = old_td.getFieldName(i);
if (alias == null) {
alias = "null";
} else if (field == null) {
field = "null";
}
newFieldAr[i] = alias + "." + field;
typeAr[i] = old_td.getFieldType(i);
}
td = new TupleDesc(typeAr, newFieldAr);
}
/**
* Create a HeapPage from a set of bytes of data read from disk. The format of a HeapPage is a set
* of header bytes indicating the slots of the page that are in use, some number of tuple slots.
* Specifically, the number of tuples is equal to:
*
* <p>floor((BufferPool.PAGE_SIZE*8) / (tuple size * 8 + 1))
*
* <p>where tuple size is the size of tuples in this database table, which can be determined via
* {@link Catalog#getTupleDesc}. The number of 8-bit header words is equal to:
*
* <p>ceiling(no. tuple slots / 8)
*
* <p>
*
* @see Database#getCatalog
* @see Catalog#getTupleDesc
* @see BufferPool#PAGE_SIZE
*/
public HeapPage(HeapPageId id, byte[] data) throws IOException {
this.pid = id;
this.td = Database.getCatalog().getTupleDesc(id.getTableId());
this.numSlots = getNumTuples();
DataInputStream dis = new DataInputStream(new ByteArrayInputStream(data));
// allocate and read the header slots of this page
header = new byte[getHeaderSize()];
for (int i = 0; i < header.length; i++) header[i] = dis.readByte();
try {
// allocate and read the actual records of this page
tuples = new Tuple[numSlots];
for (int i = 0; i < tuples.length; i++) tuples[i] = readNextTuple(dis, i);
} catch (NoSuchElementException e) {
e.printStackTrace();
}
dis.close();
setBeforeImage();
}
/**
* Create a HeapPage from a set of bytes of data read from disk. The format of a HeapPage is a set
* of 32-bit header words indicating the slots of the page that are in use, plus
* (BufferPool.PAGE_SIZE/tuple size) tuple slots, where tuple size is the size of tuples in this
* database table, which can be determined via {@link Catalog#getTupleDesc}.
*
* <p>The number of 32-bit header words is equal to:
*
* <p>(no. tuple slots / 32) + 1
*
* <p>
*
* @see Database#getCatalog
* @see Catalog#getTupleDesc
* @see BufferPool#PAGE_SIZE
*/
public HeapPage(HeapPageId id, byte[] data) throws IOException {
this.pid = id;
this.td = Database.getCatalog().getTupleDesc(id.tableid());
// this.numSlots = (BufferPool.PAGE_SIZE) / (td.getSize());
this.numSlots = (BufferPool.PAGE_SIZE * 8) / ((td.getSize() * 8) + 1);
// System.out.println(this.numSlots);
DataInputStream dis = new DataInputStream(new ByteArrayInputStream(data));
// allocate and read the header slots of this page
header = new Header(dis);
try {
// allocate and read the actual records of this page
tuples = new Tuple[numSlots];
for (int i = 0; i < numSlots; i++) {
tuples[i] = readNextTuple(dis, i);
}
} catch (NoSuchElementException e) {
// e.printStackTrace();
}
dis.close();
}
/**
* Returns the TupleDesc with field names from the underlying HeapFile, prefixed with the
* tableAlias string from the constructor. This prefix becomes useful when joining tables
* containing a field(s) with the same name.
*
* @return the TupleDesc with field names from the underlying HeapFile, prefixed with the
* tableAlias string from the constructor.
*/
public TupleDesc getTupleDesc() {
// some code goes here
TupleDesc td = Database.getCatalog().getTupleDesc(tableid);
Iterator<TDItem> tdIter = td.iterator();
int size = td.numFields();
Type[] typeAr = new Type[size];
String[] fieldAr = new String[size];
String aliasString = this.tableAlias;
TDItem item;
Type fieldType;
String fieldName;
int count = 0;
if (aliasString == null) {
aliasString = "null";
}
// for (int i = 0; i < size; i++){
// item = tdIter.next();
// fieldType = item.fieldType;
// fieldName = item.fieldName;
while (tdIter.hasNext()) {
item = tdIter.next();
fieldType = item.fieldType;
fieldName = item.fieldName;
if (fieldName == null) {
fieldName = "null";
}
typeAr[count] = fieldType;
fieldAr[count] = aliasString + "." + fieldName; // "null.null case may occur"
count++;
}
return new TupleDesc(typeAr, fieldAr);
}
/**
* This is a helper method that computes the cost and cardinality of joining joinToRemove to
* joinSet (joinSet should contain joinToRemove), given that all of the subsets of size
* joinSet.size() - 1 have already been computed and stored in PlanCache pc.
*
* @param stats table stats for all of the tables, referenced by table names rather than alias
* (see {@link #orderJoins})
* @param filterSelectivities the selectivities of the filters over each of the tables (where
* tables are indentified by their alias or name if no alias is given)
* @param joinToRemove the join to remove from joinSet
* @param joinSet the set of joins being considered
* @param bestCostSoFar the best way to join joinSet so far (minimum of previous invocations of
* computeCostAndCardOfSubplan for this joinSet, from returned CostCard)
* @param pc the PlanCache for this join; should have subplans for all plans of size
* joinSet.size()-1
* @return A {@link CostCard} objects desribing the cost, cardinality, optimal subplan
* @throws ParsingException when stats, filterSelectivities, or pc object is missing tables
* involved in join
*/
@SuppressWarnings("unchecked")
private CostCard computeCostAndCardOfSubplan(
HashMap<String, TableStats> stats,
HashMap<String, Double> filterSelectivities,
LogicalJoinNode joinToRemove,
Set<LogicalJoinNode> joinSet,
double bestCostSoFar,
PlanCache pc)
throws ParsingException {
LogicalJoinNode j = joinToRemove;
Vector<LogicalJoinNode> prevBest;
if (this.p.getTableId(j.t1Alias) == null)
throw new ParsingException("Unknown table " + j.t1Alias);
if (this.p.getTableId(j.t2Alias) == null)
throw new ParsingException("Unknown table " + j.t2Alias);
String table1Name = Database.getCatalog().getTableName(this.p.getTableId(j.t1Alias));
String table2Name = Database.getCatalog().getTableName(this.p.getTableId(j.t2Alias));
String table1Alias = j.t1Alias;
String table2Alias = j.t2Alias;
Set<LogicalJoinNode> news = (Set<LogicalJoinNode>) ((HashSet<LogicalJoinNode>) joinSet).clone();
news.remove(j);
double t1cost, t2cost;
int t1card, t2card;
boolean leftPkey, rightPkey;
if (news.isEmpty()) { // base case -- both are base relations
prevBest = new Vector<LogicalJoinNode>();
t1cost = stats.get(table1Name).estimateScanCost();
t1card = stats.get(table1Name).estimateTableCardinality(filterSelectivities.get(j.t1Alias));
leftPkey = isPkey(j.t1Alias, j.f1PureName);
t2cost = table2Alias == null ? 0 : stats.get(table2Name).estimateScanCost();
t2card =
table2Alias == null
? 0
: stats.get(table2Name).estimateTableCardinality(filterSelectivities.get(j.t2Alias));
rightPkey = table2Alias == null ? false : isPkey(table2Alias, j.f2PureName);
} else {
// news is not empty -- figure best way to join j to news
prevBest = pc.getOrder(news);
// possible that we have not cached an answer, if subset
// includes a cross product
if (prevBest == null) {
return null;
}
double prevBestCost = pc.getCost(news);
int bestCard = pc.getCard(news);
// estimate cost of right subtree
if (doesJoin(prevBest, table1Alias)) { // j.t1 is in prevBest
t1cost = prevBestCost; // left side just has cost of whatever
// left
// subtree is
t1card = bestCard;
leftPkey = hasPkey(prevBest);
t2cost = j.t2Alias == null ? 0 : stats.get(table2Name).estimateScanCost();
t2card =
j.t2Alias == null
? 0
: stats
.get(table2Name)
.estimateTableCardinality(filterSelectivities.get(j.t2Alias));
rightPkey = j.t2Alias == null ? false : isPkey(j.t2Alias, j.f2PureName);
} else if (doesJoin(prevBest, j.t2Alias)) { // j.t2 is in prevbest
// (both
// shouldn't be)
t2cost = prevBestCost; // left side just has cost of whatever
// left
// subtree is
t2card = bestCard;
rightPkey = hasPkey(prevBest);
t1cost = stats.get(table1Name).estimateScanCost();
t1card = stats.get(table1Name).estimateTableCardinality(filterSelectivities.get(j.t1Alias));
leftPkey = isPkey(j.t1Alias, j.f1PureName);
} else {
// don't consider this plan if one of j.t1 or j.t2
// isn't a table joined in prevBest (cross product)
return null;
}
}
// case where prevbest is left
double cost1 = estimateJoinCost(j, t1card, t2card, t1cost, t2cost);
LogicalJoinNode j2 = j.swapInnerOuter();
double cost2 = estimateJoinCost(j2, t2card, t1card, t2cost, t1cost);
if (cost2 < cost1) {
boolean tmp;
j = j2;
cost1 = cost2;
tmp = rightPkey;
rightPkey = leftPkey;
leftPkey = tmp;
}
if (cost1 >= bestCostSoFar) return null;
CostCard cc = new CostCard();
cc.card = estimateJoinCardinality(j, t1card, t2card, leftPkey, rightPkey, stats);
cc.cost = cost1;
cc.plan = (Vector<LogicalJoinNode>) prevBest.clone();
cc.plan.addElement(j); // prevbest is left -- add new join to end
return cc;
}
public SeqScan(TransactionId tid, int tableid) {
this(tid, tableid, Database.getCatalog().getTableName(tableid));
}
/**
* @return return the table name of the table the operator scans. This should be the actual name
* of the table in the catalog of the database
*/
public String getTableName() {
return Database.getCatalog().getTableName(id);
}
/**
* Creates a sequential scan over the specified table as a part of the specified transaction.
*
* @param tid The transaction this scan is running as a part of.
* @param tableid the table to scan.
* @param tableAlias the alias of this table (needed by the parser); the returned tupleDesc should
* have fields with name tableAlias.fieldName (note: this class is not responsible for
* handling a case where tableAlias or fieldName are null. It shouldn't crash if they are, but
* the resulting name can be null.fieldName, tableAlias.null, or null.null).
*/
public SeqScan(TransactionId tid, int tableid, String tableAlias) {
alias = tableAlias;
id = tableid;
dbiter = Database.getCatalog().getDatabaseFile(id).iterator(tid);
}
/**
* Static method to generate a byte array corresponding to an empty HeapPage. Used to add new,
* empty pages to the file. Passing the results of this method to the HeapPage constructor will
* create a HeapPage with no valid tuples in it.
*
* @param tableid The id of the table that this empty page will belong to.
* @return The returned ByteArray.
*/
public static byte[] createEmptyPageData(int tableid) {
TupleDesc td = Database.getCatalog().getTupleDesc(tableid);
// int hb = (((BufferPool.PAGE_SIZE / td.getSize()) / 32) +1) * 4;
int len = BufferPool.PAGE_SIZE; // + hb;
return new byte[len]; // all 0
}
/**
* This is a helper method that computes the cost and cardinality of joining a LogicalJoinNode j
* to the current greedy plan we have built up.
*
* @param j the join to try adding to our plan
* @param plan the current plan we have built so far from the greedy algorithm, a Vector of
* LogicalJoinNodes that we've so far chosen.
* @param planCardinalities given the join order from plan, we also keep track of how large joined
* tables are, so we can help estimate the cardinality and cost of this next join
* @param planCosts given the join order from plan, we also keep track of how expensive executing
* some joins are, so we can help estimate the cardinality and cost of this next join
* @param stats table stats for all of the tables, referenced by table names rather than alias
* (see {@link #orderGreedyJoins(HashMap, HashMap)})
* @param filterSelectivities the selectivities of the filters over each of the tables (where
* tables are indentified by their alias or name if no alias is given)
* @return A {@link CostCard} objects desribing the cost, cardinality, optimal subplan
* @throws ParsingException when stats, filterSelectivities, or pc object is missing tables
* involved in join
*/
private CostCard costGreedyJoin(
LogicalJoinNode j,
Vector<LogicalJoinNode> plan,
Vector<Integer> planCardinalities,
Vector<Double> planCosts,
HashMap<String, TableStats> stats,
HashMap<String, Double> filterSelectivities)
throws ParsingException {
if (this.p.getTableId(j.t1Alias) == null)
throw new ParsingException("Unknown table " + j.t1Alias);
if (this.p.getTableId(j.t2Alias) == null)
throw new ParsingException("Unknown table " + j.t2Alias);
String table1Name = Database.getCatalog().getTableName(this.p.getTableId(j.t1Alias));
String table2Name = Database.getCatalog().getTableName(this.p.getTableId(j.t2Alias));
String table1Alias = j.t1Alias;
String table2Alias = j.t2Alias;
double t1cost, t2cost;
int t1card, t2card;
boolean leftPkey, rightPkey;
// estimate cost of right subtree
if (doesJoin(plan, table1Alias)) { // j.t1 is in plan already
CostCard c = getCostCard(plan, planCardinalities, planCosts, table1Alias);
t1cost = c.cost; // left side just has cost of whatever left subtree is
t1card = c.card;
leftPkey = hasPkey(plan);
t2cost = j.t2Alias == null ? 0 : stats.get(table2Name).estimateScanCost();
t2card =
j.t2Alias == null
? 0
: stats.get(table2Name).estimateTableCardinality(filterSelectivities.get(j.t2Alias));
rightPkey = j.t2Alias == null ? false : isPkey(j.t2Alias, j.f2PureName);
} else if (doesJoin(plan, j.t2Alias)) { // j.t2 is in plan
// (else if since both j.t1 and j.t2 shouldn't both be)
CostCard c = getCostCard(plan, planCardinalities, planCosts, table2Alias);
t2cost = c.cost;
t2card = c.card;
rightPkey = hasPkey(plan);
t1cost = stats.get(table1Name).estimateScanCost();
t1card = stats.get(table1Name).estimateTableCardinality(filterSelectivities.get(j.t1Alias));
leftPkey = isPkey(j.t1Alias, j.f1PureName);
} else { // Neither is a plan, both are just single tables
t1cost = stats.get(table1Name).estimateScanCost();
t1card = stats.get(table1Name).estimateTableCardinality(filterSelectivities.get(j.t1Alias));
leftPkey = isPkey(j.t1Alias, j.f1PureName);
t2cost = table2Alias == null ? 0 : stats.get(table2Name).estimateScanCost();
t2card =
table2Alias == null
? 0
: stats.get(table2Name).estimateTableCardinality(filterSelectivities.get(j.t2Alias));
rightPkey = table2Alias == null ? false : isPkey(table2Alias, j.f2PureName);
}
double cost1 = estimateJoinCost(j, t1card, t2card, t1cost, t2cost);
LogicalJoinNode j2 = j.swapInnerOuter();
double cost2 = estimateJoinCost(j2, t2card, t1card, t2cost, t1cost);
if (cost2 < cost1) {
boolean tmp;
j = j2;
cost1 = cost2;
tmp = rightPkey;
rightPkey = leftPkey;
leftPkey = tmp;
}
CostCard cc = new CostCard();
cc.card = estimateJoinCardinality(j, t1card, t2card, leftPkey, rightPkey, stats);
cc.cost = cost1;
return cc;
}
/**
* Return true if field is a primary key of the specified table, false otherwise
*
* @param tableAlias The alias of the table in the query
* @param field The pure name of the field
*/
private boolean isPkey(String tableAlias, String field) {
int tid1 = p.getTableId(tableAlias);
String pkey1 = Database.getCatalog().getPrimaryKey(tid1);
return pkey1.equals(field);
}
/**
* Helper function to display a Swing window with a tree representation of the specified list of
* joins. See {@link #orderJoins}, which may want to call this when the analyze flag is true.
*
* @param js the join plan to visualize
* @param pc the PlanCache accumulated whild building the optimal plan
* @param stats table statistics for base tables
* @param selectivities the selectivities of the filters over each of the tables (where tables are
* indentified by their alias or name if no alias is given)
*/
private void printJoins(
Vector<LogicalJoinNode> js,
PlanCache pc,
HashMap<String, TableStats> stats,
HashMap<String, Double> selectivities) {
JFrame f = new JFrame("Join Plan for " + p.getQuery());
// Set the default close operation for the window,
// or else the program won't exit when clicking close button
f.setDefaultCloseOperation(WindowConstants.DISPOSE_ON_CLOSE);
f.setVisible(true);
f.setSize(300, 500);
HashMap<String, DefaultMutableTreeNode> m = new HashMap<String, DefaultMutableTreeNode>();
// int numTabs = 0;
// int k;
DefaultMutableTreeNode root = null, treetop = null;
HashSet<LogicalJoinNode> pathSoFar = new HashSet<LogicalJoinNode>();
boolean neither;
System.out.println(js);
for (LogicalJoinNode j : js) {
pathSoFar.add(j);
System.out.println("PATH SO FAR = " + pathSoFar);
String table1Name = Database.getCatalog().getTableName(this.p.getTableId(j.t1Alias));
String table2Name = Database.getCatalog().getTableName(this.p.getTableId(j.t2Alias));
// Double c = pc.getCost(pathSoFar);
neither = true;
root =
new DefaultMutableTreeNode(
"Join "
+ j
+ " (Cost ="
+ pc.getCost(pathSoFar)
+ ", card = "
+ pc.getCard(pathSoFar)
+ ")");
DefaultMutableTreeNode n = m.get(j.t1Alias);
if (n == null) { // never seen this table before
n =
new DefaultMutableTreeNode(
j.t1Alias
+ " (Cost = "
+ stats.get(table1Name).estimateScanCost()
+ ", card = "
+ stats.get(table1Name).estimateTableCardinality(selectivities.get(j.t1Alias))
+ ")");
root.add(n);
} else {
// make left child root n
root.add(n);
neither = false;
}
m.put(j.t1Alias, root);
n = m.get(j.t2Alias);
if (n == null) { // never seen this table before
n =
new DefaultMutableTreeNode(
j.t2Alias == null
? "Subplan"
: (j.t2Alias
+ " (Cost = "
+ stats.get(table2Name).estimateScanCost()
+ ", card = "
+ stats
.get(table2Name)
.estimateTableCardinality(selectivities.get(j.t2Alias))
+ ")"));
root.add(n);
} else {
// make right child root n
root.add(n);
neither = false;
}
m.put(j.t2Alias, root);
// unless this table doesn't join with other tables,
// all tables are accessed from root
if (!neither) {
for (String key : m.keySet()) {
m.put(key, root);
}
}
treetop = root;
}
JTree tree = new JTree(treetop);
JScrollPane treeView = new JScrollPane(tree);
tree.setShowsRootHandles(true);
// Set the icon for leaf nodes.
ImageIcon leafIcon = new ImageIcon("join.jpg");
DefaultTreeCellRenderer renderer = new DefaultTreeCellRenderer();
renderer.setOpenIcon(leafIcon);
renderer.setClosedIcon(leafIcon);
tree.setCellRenderer(renderer);
f.setSize(300, 500);
f.add(treeView);
for (int i = 0; i < tree.getRowCount(); i++) {
tree.expandRow(i);
}
if (js.size() == 0) {
f.add(new JLabel("No joins in plan."));
}
f.pack();
}
public void open() throws DbException, TransactionAbortedException {
// some code goes here
fileIt = Database.getCatalog().getDbFile(tableid).iterator(tid);
fileIt.open();
}
