/**
* Checks that the column permutation + select list + static partition exprs + dynamic partition
* exprs collectively cover exactly all columns in the target table (not more of fewer).
*/
private void checkColumnCoverage(
ArrayList<Column> selectExprTargetColumns,
Set<String> mentionedColumnNames,
int numSelectListExprs,
int numStaticPartitionExprs)
throws AnalysisException {
boolean isHBaseTable = (table_ instanceof HBaseTable);
int numClusteringCols = isHBaseTable ? 0 : table_.getNumClusteringCols();
// Check that all columns are mentioned by the permutation and partition clauses
if (selectExprTargetColumns.size() + numStaticPartitionExprs != table_.getColumns().size()) {
// We've already ruled out too many columns in the permutation and partition clauses
// by checking that there are no duplicates and that every column mentioned actually
// exists. So all columns aren't mentioned in the query. If the unmentioned columns
// include partition columns, this is an error.
List<String> missingColumnNames = Lists.newArrayList();
for (Column column : table_.getColumns()) {
if (!mentionedColumnNames.contains(column.getName())) {
// HBase tables have a single row-key column which is always in position 0. It
// must be mentioned, since it is invalid to set it to NULL (which would
// otherwise happen by default).
if (isHBaseTable && column.getPosition() == 0) {
throw new AnalysisException(
"Row-key column '"
+ column.getName()
+ "' must be explicitly mentioned in column permutation.");
}
if (column.getPosition() < numClusteringCols) {
missingColumnNames.add(column.getName());
}
}
}
if (!missingColumnNames.isEmpty()) {
throw new AnalysisException(
"Not enough partition columns mentioned in query. Missing columns are: "
+ Joiner.on(", ").join(missingColumnNames));
}
}
// Expect the selectListExpr to have entries for every target column
if (selectExprTargetColumns.size() != numSelectListExprs) {
String comparator = (selectExprTargetColumns.size() < numSelectListExprs) ? "fewer" : "more";
String partitionClause =
(partitionKeyValues_ == null) ? "returns" : "and PARTITION clause return";
// If there was no column permutation provided, the error is that the select-list
// has the wrong number of expressions compared to the number of columns in the
// table. If there was a column permutation, then the mismatch is between the
// select-list and the permutation itself.
if (columnPermutation_ == null) {
int totalColumnsMentioned = numSelectListExprs + numStaticPartitionExprs;
throw new AnalysisException(
String.format(
"Target table '%s' has %s columns (%s) than the SELECT / VALUES clause %s"
+ " (%s)",
table_.getFullName(),
comparator,
table_.getColumns().size(),
partitionClause,
totalColumnsMentioned));
} else {
String partitionPrefix =
(partitionKeyValues_ == null) ? "mentions" : "and PARTITION clause mention";
throw new AnalysisException(
String.format(
"Column permutation %s %s columns (%s) than "
+ "the SELECT / VALUES clause %s (%s)",
partitionPrefix,
comparator,
selectExprTargetColumns.size(),
partitionClause,
numSelectListExprs));
}
}
}
/**
* Performs three final parts of the analysis: 1. Checks type compatibility between all
* expressions and their targets
*
* <p>2. Populates partitionKeyExprs with type-compatible expressions, in Hive partition-column
* order, for all partition columns
*
* <p>3. Populates resultExprs_ with type-compatible expressions, in Hive column order, for all
* expressions in the select-list. Unmentioned columns are assigned NULL literal expressions.
*
* <p>If necessary, adds casts to the expressions to make them compatible with the type of the
* corresponding column.
*
* @throws AnalysisException If an expression is not compatible with its target column
*/
private void prepareExpressions(
List<Column> selectExprTargetColumns,
List<Expr> selectListExprs,
Table tbl,
Analyzer analyzer)
throws AnalysisException {
// Temporary lists of partition key exprs and names in an arbitrary order.
List<Expr> tmpPartitionKeyExprs = new ArrayList<Expr>();
List<String> tmpPartitionKeyNames = new ArrayList<String>();
int numClusteringCols = (tbl instanceof HBaseTable) ? 0 : tbl.getNumClusteringCols();
// Check dynamic partition columns for type compatibility.
for (int i = 0; i < selectListExprs.size(); ++i) {
Column targetColumn = selectExprTargetColumns.get(i);
Expr compatibleExpr = checkTypeCompatibility(targetColumn, selectListExprs.get(i));
if (targetColumn.getPosition() < numClusteringCols) {
// This is a dynamic clustering column
tmpPartitionKeyExprs.add(compatibleExpr);
tmpPartitionKeyNames.add(targetColumn.getName());
}
selectListExprs.set(i, compatibleExpr);
}
// Check static partition columns, dynamic entries in partitionKeyValues will already
// be in selectExprTargetColumns and therefore are ignored in this loop
if (partitionKeyValues_ != null) {
for (PartitionKeyValue pkv : partitionKeyValues_) {
if (pkv.isStatic()) {
// tableColumns is guaranteed to exist after the earlier analysis checks
Column tableColumn = table_.getColumn(pkv.getColName());
Expr compatibleExpr = checkTypeCompatibility(tableColumn, pkv.getValue());
tmpPartitionKeyExprs.add(compatibleExpr);
tmpPartitionKeyNames.add(pkv.getColName());
}
}
}
// Reorder the partition key exprs and names to be consistent with the target table
// declaration. We need those exprs in the original order to create the corresponding
// Hdfs folder structure correctly.
for (Column c : table_.getColumns()) {
for (int j = 0; j < tmpPartitionKeyNames.size(); ++j) {
if (c.getName().equals(tmpPartitionKeyNames.get(j))) {
partitionKeyExprs_.add(tmpPartitionKeyExprs.get(j));
break;
}
}
}
Preconditions.checkState(partitionKeyExprs_.size() == numClusteringCols);
// Make sure we have stats for partitionKeyExprs
for (Expr expr : partitionKeyExprs_) {
expr.analyze(analyzer);
}
// Finally, 'undo' the permutation so that the selectListExprs are in Hive column
// order, and add NULL expressions to all missing columns.
for (Column tblColumn : table_.getColumnsInHiveOrder()) {
boolean matchFound = false;
for (int i = 0; i < selectListExprs.size(); ++i) {
if (selectExprTargetColumns.get(i).getName().equals(tblColumn.getName())) {
resultExprs_.add(selectListExprs.get(i));
matchFound = true;
break;
}
}
// If no match is found, either the column is a clustering column with a static
// value, or it was unmentioned and therefore should have a NULL select-list
// expression.
if (!matchFound) {
if (tblColumn.getPosition() >= numClusteringCols) {
// Unmentioned non-clustering columns get NULL literals with the appropriate
// target type because Parquet cannot handle NULL_TYPE (IMPALA-617).
resultExprs_.add(NullLiteral.create(tblColumn.getType()));
}
}
}
// TODO: Check that HBase row-key columns are not NULL? See IMPALA-406
if (needsGeneratedQueryStatement_) {
// Build a query statement that returns NULL for every column
List<SelectListItem> selectListItems = Lists.newArrayList();
for (Expr e : resultExprs_) {
selectListItems.add(new SelectListItem(e, null));
}
SelectList selectList = new SelectList(selectListItems);
queryStmt_ = new SelectStmt(selectList, null, null, null, null, null, null);
queryStmt_.analyze(analyzer);
}
}
@Override
public void analyze(Analyzer analyzer) throws AnalysisException {
if (isExplain_) analyzer.setIsExplain();
try {
if (withClause_ != null) withClause_.analyze(analyzer);
} catch (AnalysisException e) {
// Ignore AnalysisExceptions if tables are missing to ensure the maximum number
// of missing tables can be collected before failing analyze().
if (analyzer.getMissingTbls().isEmpty()) throw e;
}
List<Expr> selectListExprs = null;
if (!needsGeneratedQueryStatement_) {
try {
// Use a child analyzer for the query stmt to properly scope WITH-clause
// views and to ignore irrelevant ORDER BYs.
Analyzer queryStmtAnalyzer = new Analyzer(analyzer);
queryStmt_.analyze(queryStmtAnalyzer);
if (analyzer.containsSubquery()) {
Preconditions.checkState(queryStmt_ instanceof SelectStmt);
StmtRewriter.rewriteStatement((SelectStmt) queryStmt_, queryStmtAnalyzer);
queryStmt_ = queryStmt_.clone();
queryStmtAnalyzer = new Analyzer(analyzer);
queryStmt_.analyze(queryStmtAnalyzer);
}
selectListExprs = Expr.cloneList(queryStmt_.getBaseTblResultExprs());
} catch (AnalysisException e) {
if (analyzer.getMissingTbls().isEmpty()) throw e;
}
} else {
selectListExprs = Lists.newArrayList();
}
// Set target table and perform table-type specific analysis and auth checking.
// Also checks if the target table is missing.
setTargetTable(analyzer);
// Abort analysis if there are any missing tables beyond this point.
if (!analyzer.getMissingTbls().isEmpty()) {
throw new AnalysisException("Found missing tables. Aborting analysis.");
}
boolean isHBaseTable = (table_ instanceof HBaseTable);
int numClusteringCols = isHBaseTable ? 0 : table_.getNumClusteringCols();
// Analysis of the INSERT statement from this point is basically the act of matching
// the set of output columns (which come from a column permutation, perhaps
// implicitly, and the PARTITION clause) to the set of input columns (which come from
// the select-list and any statically-valued columns in the PARTITION clause).
//
// First, we compute the set of mentioned columns, and reject statements that refer to
// non-existent columns, or duplicates (we must check both the column permutation, and
// the set of partition keys). Next, we check that all partition columns are
// mentioned. During this process we build the map from select-list expr index to
// column in the targeted table.
//
// Then we check that the select-list contains exactly the right number of expressions
// for all mentioned columns which are not statically-valued partition columns (which
// get their expressions from partitionKeyValues).
//
// Finally, prepareExpressions analyzes the expressions themselves, and confirms that
// they are type-compatible with the target columns. Where columns are not mentioned
// (and by this point, we know that missing columns are not partition columns),
// prepareExpressions assigns them a NULL literal expressions.
// An null permutation clause is the same as listing all non-partition columns in
// order.
List<String> analysisColumnPermutation = columnPermutation_;
if (analysisColumnPermutation == null) {
analysisColumnPermutation = Lists.newArrayList();
ArrayList<Column> tableColumns = table_.getColumns();
for (int i = numClusteringCols; i < tableColumns.size(); ++i) {
analysisColumnPermutation.add(tableColumns.get(i).getName());
}
}
// selectExprTargetColumns maps from select expression index to a column in the target
// table. It will eventually include all mentioned columns that aren't static-valued
// partition columns.
ArrayList<Column> selectExprTargetColumns = Lists.newArrayList();
// Tracks the name of all columns encountered in either the permutation clause or the
// partition clause to detect duplicates.
Set<String> mentionedColumnNames = Sets.newHashSet();
for (String columnName : analysisColumnPermutation) {
Column column = table_.getColumn(columnName);
if (column == null) {
throw new AnalysisException("Unknown column '" + columnName + "' in column permutation");
}
if (!mentionedColumnNames.add(columnName)) {
throw new AnalysisException("Duplicate column '" + columnName + "' in column permutation");
}
selectExprTargetColumns.add(column);
}
int numStaticPartitionExprs = 0;
if (partitionKeyValues_ != null) {
for (PartitionKeyValue pkv : partitionKeyValues_) {
Column column = table_.getColumn(pkv.getColName());
if (column == null) {
throw new AnalysisException(
"Unknown column '" + pkv.getColName() + "' in partition clause");
}
if (column.getPosition() >= numClusteringCols) {
throw new AnalysisException(
"Column '" + pkv.getColName() + "' is not a partition column");
}
if (!mentionedColumnNames.add(pkv.getColName())) {
throw new AnalysisException(
"Duplicate column '" + pkv.getColName() + "' in partition clause");
}
if (!pkv.isDynamic()) {
numStaticPartitionExprs++;
} else {
selectExprTargetColumns.add(column);
}
}
}
// Checks that exactly all columns in the target table are assigned an expr.
checkColumnCoverage(
selectExprTargetColumns,
mentionedColumnNames,
selectListExprs.size(),
numStaticPartitionExprs);
// Make sure static partition key values only contain const exprs.
if (partitionKeyValues_ != null) {
for (PartitionKeyValue kv : partitionKeyValues_) {
kv.analyze(analyzer);
}
}
// Populate partitionKeyExprs from partitionKeyValues and selectExprTargetColumns
prepareExpressions(selectExprTargetColumns, selectListExprs, table_, analyzer);
// Analyze plan hints at the end to prefer reporting other error messages first
// (e.g., the PARTITION clause is not applicable to unpartitioned and HBase tables).
analyzePlanHints(analyzer);
}