/**
* Returns statistics on this table as a tabular result set. Used for the SHOW TABLE STATS
* statement. The schema of the returned TResultSet is set inside this method.
*/
public TResultSet getTableStats() {
TResultSet result = new TResultSet();
TResultSetMetadata resultSchema = new TResultSetMetadata();
result.setSchema(resultSchema);
for (int i = 0; i < numClusteringCols_; ++i) {
// Add the partition-key values as strings for simplicity.
Column partCol = colsByPos_.get(i);
TColumn colDesc = new TColumn(partCol.getName(), partCol.getType().toThrift());
resultSchema.addToColumns(colDesc);
}
resultSchema.addToColumns(new TColumn("#Rows", ColumnType.BIGINT.toThrift()));
resultSchema.addToColumns(new TColumn("#Files", ColumnType.BIGINT.toThrift()));
resultSchema.addToColumns(new TColumn("Size", ColumnType.STRING.toThrift()));
resultSchema.addToColumns(new TColumn("Format", ColumnType.STRING.toThrift()));
// Pretty print partitions and their stats.
ArrayList<HdfsPartition> orderedPartitions = Lists.newArrayList(partitions_);
Collections.sort(orderedPartitions);
for (HdfsPartition p : orderedPartitions) {
// Ignore dummy default partition.
if (p.getId() == ImpalaInternalServiceConstants.DEFAULT_PARTITION_ID) continue;
TResultRowBuilder rowBuilder = new TResultRowBuilder();
// Add the partition-key values (as strings for simplicity).
for (LiteralExpr expr : p.getPartitionValues()) {
rowBuilder.add(expr.getStringValue());
}
// Add number of rows, files, bytes and the file format.
rowBuilder
.add(p.getNumRows())
.add(p.getFileDescriptors().size())
.addBytes(p.getSize())
.add(p.getInputFormatDescriptor().getFileFormat().toString());
result.addToRows(rowBuilder.get());
}
// For partitioned tables add a summary row at the bottom.
if (numClusteringCols_ > 0) {
TResultRowBuilder rowBuilder = new TResultRowBuilder();
int numEmptyCells = numClusteringCols_ - 1;
rowBuilder.add("Total");
for (int i = 0; i < numEmptyCells; ++i) {
rowBuilder.add("");
}
// Total num rows, files, and bytes (leave format empty).
rowBuilder.add(numRows_).add(numHdfsFiles_).addBytes(totalHdfsBytes_).add("");
result.addToRows(rowBuilder.get());
}
return result;
}
/**
* Create HdfsPartition objects corresponding to 'partitions'.
*
* <p>If there are no partitions in the Hive metadata, a single partition is added with no
* partition keys.
*
* <p>For files that have not been changed, reuses file descriptors from oldFileDescMap.
*/
private void loadPartitions(
List<org.apache.hadoop.hive.metastore.api.Partition> msPartitions,
org.apache.hadoop.hive.metastore.api.Table msTbl,
Map<String, FileDescriptor> oldFileDescMap)
throws IOException, CatalogException {
partitions_.clear();
hdfsBaseDir_ = msTbl.getSd().getLocation();
List<FileDescriptor> newFileDescs = Lists.newArrayList();
// INSERT statements need to refer to this if they try to write to new partitions
// Scans don't refer to this because by definition all partitions they refer to
// exist.
addDefaultPartition(msTbl.getSd());
if (msTbl.getPartitionKeysSize() == 0) {
Preconditions.checkArgument(msPartitions == null || msPartitions.isEmpty());
// This table has no partition key, which means it has no declared partitions.
// We model partitions slightly differently to Hive - every file must exist in a
// partition, so add a single partition with no keys which will get all the
// files in the table's root directory.
addPartition(msTbl.getSd(), null, new ArrayList<LiteralExpr>(), oldFileDescMap, newFileDescs);
Path location = new Path(hdfsBaseDir_);
if (DFS.exists(location)) {
accessLevel_ = getAvailableAccessLevel(location);
}
} else {
// keep track of distinct partition key values and how many nulls there are
Set<String>[] uniquePartitionKeys = new HashSet[numClusteringCols_];
long[] numNullKeys = new long[numClusteringCols_];
for (int i = 0; i < numClusteringCols_; ++i) {
uniquePartitionKeys[i] = new HashSet<String>();
numNullKeys[i] = 0;
}
for (org.apache.hadoop.hive.metastore.api.Partition msPartition : msPartitions) {
// load key values
List<LiteralExpr> keyValues = Lists.newArrayList();
int i = 0;
for (String partitionKey : msPartition.getValues()) {
uniquePartitionKeys[i].add(partitionKey);
// Deal with Hive's special NULL partition key.
if (partitionKey.equals(nullPartitionKeyValue_)) {
keyValues.add(new NullLiteral());
++numNullKeys[i];
} else {
ColumnType type = colsByPos_.get(keyValues.size()).getType();
try {
Expr expr = LiteralExpr.create(partitionKey, type);
// Force the literal to be of type declared in the metadata.
expr = expr.castTo(type);
keyValues.add((LiteralExpr) expr);
} catch (AnalysisException ex) {
LOG.warn("Failed to create literal expression of type: " + type, ex);
throw new InvalidStorageDescriptorException(ex);
}
}
++i;
}
HdfsPartition partition =
addPartition(msPartition.getSd(), msPartition, keyValues, oldFileDescMap, newFileDescs);
// If the partition is null, its HDFS path does not exist, and it was not added to
// this table's partition list. Skip the partition.
if (partition == null) continue;
if (msPartition.getParameters() != null) {
partition.setNumRows(getRowCount(msPartition.getParameters()));
}
if (!TAccessLevelUtil.impliesWriteAccess(partition.getAccessLevel())) {
// TODO: READ_ONLY isn't exactly correct because the it's possible the
// partition does not have READ permissions either. When we start checking
// whether we can READ from a table, this should be updated to set the
// table's access level to the "lowest" effective level across all
// partitions. That is, if one partition has READ_ONLY and another has
// WRITE_ONLY the table's access level should be NONE.
accessLevel_ = TAccessLevel.READ_ONLY;
}
}
// update col stats for partition key cols
for (int i = 0; i < numClusteringCols_; ++i) {
ColumnStats stats = colsByPos_.get(i).getStats();
stats.setNumNulls(numNullKeys[i]);
stats.setNumDistinctValues(uniquePartitionKeys[i].size());
LOG.debug("#col=" + Integer.toString(i) + " stats=" + stats.toString());
}
}
if (newFileDescs.size() > 0) {
loadBlockMd(newFileDescs);
}
uniqueHostPortsCount_ = countUniqueDataNetworkLocations(partitions_);
}