/**
* Encode the primary key values from the table as a byte array. The values must be in the same
* order as the primary key constraint. If the connection and table are both tenant-specific, the
* tenant ID column must not be present in the values.
*
* @param conn an open connection
* @param fullTableName the full table name
* @param values the values of the primary key columns ordered in the same order as the primary
* key constraint
* @return the encoded byte array
* @throws SQLException if the table cannot be found or the incorrect number of of values are
* provided
* @see #decodePK(Connection, String, byte[]) to decode the byte[] back to the values
*/
public static byte[] encodePK(Connection conn, String fullTableName, Object[] values)
throws SQLException {
PTable table = getTable(conn, fullTableName);
PhoenixConnection pconn = conn.unwrap(PhoenixConnection.class);
int offset =
(table.getBucketNum() == null ? 0 : 1)
+ (table.isMultiTenant() && pconn.getTenantId() != null ? 1 : 0);
List<PColumn> pkColumns = table.getPKColumns();
if (pkColumns.size() - offset != values.length) {
throw new SQLException(
"Expected " + (pkColumns.size() - offset) + " but got " + values.length);
}
PDataType type = null;
TrustedByteArrayOutputStream output =
new TrustedByteArrayOutputStream(table.getRowKeySchema().getEstimatedValueLength());
try {
for (int i = offset; i < pkColumns.size(); i++) {
if (type != null && !type.isFixedWidth()) {
output.write(QueryConstants.SEPARATOR_BYTE);
}
type = pkColumns.get(i).getDataType();
byte[] value = type.toBytes(values[i - offset]);
output.write(value);
}
return output.toByteArray();
} finally {
try {
output.close();
} catch (IOException e) {
throw new RuntimeException(e); // Impossible
}
}
}
/*
Map all unique pairs <family, name> to index. Table name is part of TableRowkey, so we do
not care about it
*/
private void initColumnIndexes() throws SQLException {
columnIndexes = new TreeMap(Bytes.BYTES_COMPARATOR);
int columnIndex = 0;
for (int index = 0; index < logicalNames.size(); index++) {
PTable table = PhoenixRuntime.getTable(conn, logicalNames.get(index));
List<PColumn> cls = table.getColumns();
for (int i = 0; i < cls.size(); i++) {
PColumn c = cls.get(i);
byte[] family = new byte[0];
if (c.getFamilyName() != null) // Skip PK column
family = c.getFamilyName().getBytes();
byte[] name = c.getName().getBytes();
byte[] cfn = Bytes.add(family, QueryConstants.NAMESPACE_SEPARATOR_BYTES, name);
if (!columnIndexes.containsKey(cfn)) {
columnIndexes.put(cfn, new Integer(columnIndex));
columnIndex++;
}
}
byte[] emptyColumnFamily = SchemaUtil.getEmptyColumnFamily(table);
byte[] cfn =
Bytes.add(
emptyColumnFamily,
QueryConstants.NAMESPACE_SEPARATOR_BYTES,
QueryConstants.EMPTY_COLUMN_BYTES);
columnIndexes.put(cfn, new Integer(columnIndex));
columnIndex++;
}
}
public static String findExistingColumn(PTable table, List<PColumn> columns) {
for (PColumn column : columns) {
PName familyName = column.getFamilyName();
if (familyName == null) {
try {
return table.getPKColumn(column.getName().getString()).getName().getString();
} catch (ColumnNotFoundException e) {
continue;
}
} else {
try {
return table
.getColumnFamily(familyName.getString())
.getColumn(column.getName().getString())
.getName()
.getString();
} catch (ColumnFamilyNotFoundException e) {
continue; // Shouldn't happen
} catch (ColumnNotFoundException e) {
continue;
}
}
}
return null;
}
public OrderPreservingTracker(
StatementContext context,
GroupBy groupBy,
Ordering ordering,
int nNodes,
TupleProjector projector) {
this.context = context;
if (groupBy.isEmpty()) {
PTable table = context.getResolver().getTables().get(0).getTable();
this.isOrderPreserving = table.rowKeyOrderOptimizable();
boolean isSalted = table.getBucketNum() != null;
boolean isMultiTenant =
context.getConnection().getTenantId() != null && table.isMultiTenant();
boolean isSharedViewIndex = table.getViewIndexId() != null;
// TODO: util for this offset, as it's computed in numerous places
this.pkPositionOffset =
(isSalted ? 1 : 0) + (isMultiTenant ? 1 : 0) + (isSharedViewIndex ? 1 : 0);
} else {
this.isOrderPreserving = true;
this.pkPositionOffset = 0;
}
this.groupBy = groupBy;
this.visitor = new TrackOrderPreservingExpressionVisitor(projector);
this.orderPreservingInfos = Lists.newArrayListWithExpectedSize(nNodes);
this.ordering = ordering;
this.projector = projector;
}
private static void setValues(
byte[][] values,
int[] pkSlotIndex,
int[] columnIndexes,
PTable table,
Map<ImmutableBytesPtr, Map<PColumn, byte[]>> mutation) {
Map<PColumn, byte[]> columnValues = Maps.newHashMapWithExpectedSize(columnIndexes.length);
byte[][] pkValues = new byte[table.getPKColumns().size()][];
// If the table uses salting, the first byte is the salting byte, set to an empty array
// here and we will fill in the byte later in PRowImpl.
if (table.getBucketNum() != null) {
pkValues[0] = new byte[] {0};
}
for (int i = 0; i < values.length; i++) {
byte[] value = values[i];
PColumn column = table.getColumns().get(columnIndexes[i]);
if (SchemaUtil.isPKColumn(column)) {
pkValues[pkSlotIndex[i]] = value;
} else {
columnValues.put(column, value);
}
}
ImmutableBytesPtr ptr = new ImmutableBytesPtr();
table.newKey(ptr, pkValues);
mutation.put(ptr, columnValues);
}
/**
* Returns the column info for the given column for the given table.
*
* @param table
* @param columnName User-specified column name. May be family-qualified or bare.
* @return columnInfo associated with the column in the table
* @throws SQLException if parameters are null or if column is not found or if column is
* ambiguous.
*/
public static ColumnInfo getColumnInfo(PTable table, String columnName) throws SQLException {
if (table == null) {
throw new SQLException("Table must not be null.");
}
if (columnName == null) {
throw new SQLException("columnName must not be null.");
}
columnName = columnName.trim().toUpperCase();
PColumn pColumn = null;
if (columnName.contains(QueryConstants.NAME_SEPARATOR)) {
String[] tokens = columnName.split(QueryConstants.NAME_SEPARATOR_REGEX);
if (tokens.length != 2) {
throw new SQLException(
String.format(
"Unable to process column %s, expected family-qualified name.", columnName));
}
String familyName = tokens[0];
String familyColumn = tokens[1];
PColumnFamily family = table.getColumnFamily(familyName);
pColumn = family.getColumn(familyColumn);
} else {
pColumn = table.getColumn(columnName);
}
return getColumnInfo(pColumn);
}
/**
* Check that none of no columns in our updatable VIEW are changing values.
*
* @param tableRef
* @param overlapViewColumns
* @param targetColumns
* @param projector
* @throws SQLException
*/
private static void throwIfNotUpdatable(
TableRef tableRef,
Set<PColumn> overlapViewColumns,
List<PColumn> targetColumns,
RowProjector projector,
boolean sameTable)
throws SQLException {
PTable table = tableRef.getTable();
if (table.getViewType() == ViewType.UPDATABLE && !overlapViewColumns.isEmpty()) {
ImmutableBytesWritable ptr = new ImmutableBytesWritable();
for (int i = 0; i < targetColumns.size(); i++) {
PColumn targetColumn = targetColumns.get(i);
if (overlapViewColumns.contains(targetColumn)) {
Expression source = projector.getColumnProjector(i).getExpression();
if (source.isStateless()) {
source.evaluate(null, ptr);
if (Bytes.compareTo(
ptr.get(),
ptr.getOffset(),
ptr.getLength(),
targetColumn.getViewConstant(),
0,
targetColumn.getViewConstant().length - 1)
== 0) {
continue;
}
}
throw new SQLExceptionInfo.Builder(SQLExceptionCode.CANNOT_UPDATE_VIEW_COLUMN)
.setColumnName(targetColumn.getName().getString())
.build()
.buildException();
}
}
}
}
public static ImmutableBytesPtr getEmptyColumnFamilyPtr(PTable table) {
List<PColumnFamily> families = table.getColumnFamilies();
return families.isEmpty()
? table.getDefaultFamilyName() == null
? QueryConstants.DEFAULT_COLUMN_FAMILY_BYTES_PTR
: table.getDefaultFamilyName().getBytesPtr()
: families.get(0).getName().getBytesPtr();
}
public static byte[] getTableKey(PTable dataTable) {
PName tenantId = dataTable.getTenantId();
PName schemaName = dataTable.getSchemaName();
return getTableKey(
tenantId == null ? ByteUtil.EMPTY_BYTE_ARRAY : tenantId.getBytes(),
schemaName == null ? ByteUtil.EMPTY_BYTE_ARRAY : schemaName.getBytes(),
dataTable.getTableName().getBytes());
}
public static String getEmptyColumnFamilyAsString(PTable table) {
List<PColumnFamily> families = table.getColumnFamilies();
return families.isEmpty()
? table.getDefaultFamilyName() == null
? QueryConstants.DEFAULT_COLUMN_FAMILY
: table.getDefaultFamilyName().getString()
: families.get(0).getName().getString();
}
@Test
public void testCreateTableToBeTransactional() throws Exception {
Properties props = PropertiesUtil.deepCopy(TEST_PROPERTIES);
Connection conn = DriverManager.getConnection(getUrl(), props);
String ddl = "CREATE TABLE TEST_TRANSACTIONAL_TABLE (k varchar primary key) transactional=true";
conn.createStatement().execute(ddl);
PhoenixConnection pconn = conn.unwrap(PhoenixConnection.class);
PTable table = pconn.getTable(new PTableKey(null, "TEST_TRANSACTIONAL_TABLE"));
HTableInterface htable =
pconn.getQueryServices().getTable(Bytes.toBytes("TEST_TRANSACTIONAL_TABLE"));
assertTrue(table.isTransactional());
assertTrue(
htable
.getTableDescriptor()
.getCoprocessors()
.contains(TransactionProcessor.class.getName()));
try {
ddl = "ALTER TABLE TEST_TRANSACTIONAL_TABLE SET transactional=false";
conn.createStatement().execute(ddl);
fail();
} catch (SQLException e) {
assertEquals(SQLExceptionCode.TX_MAY_NOT_SWITCH_TO_NON_TX.getErrorCode(), e.getErrorCode());
}
HBaseAdmin admin = pconn.getQueryServices().getAdmin();
HTableDescriptor desc = new HTableDescriptor(TableName.valueOf("TXN_TEST_EXISTING"));
desc.addFamily(new HColumnDescriptor(QueryConstants.DEFAULT_COLUMN_FAMILY_BYTES));
admin.createTable(desc);
ddl = "CREATE TABLE TXN_TEST_EXISTING (k varchar primary key) transactional=true";
conn.createStatement().execute(ddl);
assertEquals(
Boolean.TRUE.toString(),
admin
.getTableDescriptor(TableName.valueOf("TXN_TEST_EXISTING"))
.getValue(TxConstants.READ_NON_TX_DATA));
// Should be ok, as HBase metadata should match existing metadata.
ddl = "CREATE TABLE IF NOT EXISTS TEST_TRANSACTIONAL_TABLE (k varchar primary key)";
try {
conn.createStatement().execute(ddl);
fail();
} catch (SQLException e) {
assertEquals(SQLExceptionCode.TX_MAY_NOT_SWITCH_TO_NON_TX.getErrorCode(), e.getErrorCode());
}
ddl += " transactional=true";
conn.createStatement().execute(ddl);
table = pconn.getTable(new PTableKey(null, "TEST_TRANSACTIONAL_TABLE"));
htable = pconn.getQueryServices().getTable(Bytes.toBytes("TEST_TRANSACTIONAL_TABLE"));
assertTrue(table.isTransactional());
assertTrue(
htable
.getTableDescriptor()
.getCoprocessors()
.contains(TransactionProcessor.class.getName()));
}
public static short getMaxKeySeq(PTable table) {
int offset = 0;
if (table.getBucketNum() != null) {
offset++;
}
// TODO: for tenant-specific table on tenant-specific connection,
// we should subtract one for tenant column and another one for
// index ID
return (short) (table.getPKColumns().size() - offset);
}
@BeforeClass
public static void doSetup() throws Exception {
startServer(getUrl());
ensureTableCreated(getUrl(), ATABLE_NAME);
ensureTableCreated(getUrl(), ENTITY_HISTORY_TABLE_NAME);
ensureTableCreated(getUrl(), FUNKY_NAME);
ensureTableCreated(getUrl(), PTSDB_NAME);
ensureTableCreated(getUrl(), PTSDB2_NAME);
ensureTableCreated(getUrl(), PTSDB3_NAME);
ensureTableCreated(getUrl(), MULTI_CF_NAME);
ensureTableCreated(getUrl(), JOIN_ORDER_TABLE_FULL_NAME);
ensureTableCreated(getUrl(), JOIN_CUSTOMER_TABLE_FULL_NAME);
ensureTableCreated(getUrl(), JOIN_ITEM_TABLE_FULL_NAME);
ensureTableCreated(getUrl(), JOIN_SUPPLIER_TABLE_FULL_NAME);
ensureTableCreated(getUrl(), TABLE_WITH_ARRAY);
Properties props = new Properties();
props.setProperty(
PhoenixRuntime.CURRENT_SCN_ATTRIB, Long.toString(HConstants.LATEST_TIMESTAMP));
PhoenixConnection conn =
DriverManager.getConnection(PHOENIX_CONNECTIONLESS_JDBC_URL, props)
.unwrap(PhoenixConnection.class);
try {
PTable table = conn.getTable(new PTableKey(null, ATABLE_NAME));
ATABLE = table;
ORGANIZATION_ID =
new ColumnRef(new TableRef(table), table.getColumn("ORGANIZATION_ID").getPosition())
.newColumnExpression();
ENTITY_ID =
new ColumnRef(new TableRef(table), table.getColumn("ENTITY_ID").getPosition())
.newColumnExpression();
A_INTEGER =
new ColumnRef(new TableRef(table), table.getColumn("A_INTEGER").getPosition())
.newColumnExpression();
A_STRING =
new ColumnRef(new TableRef(table), table.getColumn("A_STRING").getPosition())
.newColumnExpression();
B_STRING =
new ColumnRef(new TableRef(table), table.getColumn("B_STRING").getPosition())
.newColumnExpression();
A_DATE =
new ColumnRef(new TableRef(table), table.getColumn("A_DATE").getPosition())
.newColumnExpression();
A_TIME =
new ColumnRef(new TableRef(table), table.getColumn("A_TIME").getPosition())
.newColumnExpression();
A_TIMESTAMP =
new ColumnRef(new TableRef(table), table.getColumn("A_TIMESTAMP").getPosition())
.newColumnExpression();
X_DECIMAL =
new ColumnRef(new TableRef(table), table.getColumn("X_DECIMAL").getPosition())
.newColumnExpression();
} finally {
conn.close();
}
}
/**
* Imperfect estimate of row size given a PTable TODO: keep row count in stats table and use total
* size / row count instead
*
* @param table
* @return estimate of size in bytes of a row
*/
public static long estimateRowSize(PTable table) {
int keyLength = estimateKeyLength(table);
long rowSize = 0;
for (PColumn column : table.getColumns()) {
if (!SchemaUtil.isPKColumn(column)) {
PDataType type = column.getDataType();
Integer maxLength = column.getMaxLength();
int valueLength =
!type.isFixedWidth()
? VAR_KV_LENGTH_ESTIMATE
: maxLength == null ? type.getByteSize() : maxLength;
rowSize +=
KeyValue.getKeyValueDataStructureSize(
keyLength,
column.getFamilyName().getBytes().length,
column.getName().getBytes().length,
valueLength);
}
}
// Empty key value
rowSize +=
KeyValue.getKeyValueDataStructureSize(
keyLength,
getEmptyColumnFamily(table).length,
QueryConstants.EMPTY_COLUMN_BYTES.length,
0);
return rowSize;
}
/**
* Get list of ColumnInfos that contain Column Name and its associated PDataType for an import.
* The supplied list of columns can be null -- if it is non-null, it represents a user-supplied
* list of columns to be imported.
*
* @param conn Phoenix connection from which metadata will be read
* @param tableName Phoenix table name whose columns are to be checked. Can include a schema name
* @param columns user-supplied list of import columns, can be null
*/
public static List<ColumnInfo> generateColumnInfo(
Connection conn, String tableName, List<String> columns) throws SQLException {
PTable table = PhoenixRuntime.getTable(conn, tableName);
List<ColumnInfo> columnInfoList = Lists.newArrayList();
Set<String> unresolvedColumnNames = new TreeSet<String>();
if (columns == null) {
// use all columns in the table
for (PColumn pColumn : table.getColumns()) {
int sqlType = pColumn.getDataType().getResultSetSqlType();
columnInfoList.add(new ColumnInfo(pColumn.toString(), sqlType));
}
} else {
// Leave "null" as indication to skip b/c it doesn't exist
for (int i = 0; i < columns.size(); i++) {
String columnName = columns.get(i);
try {
ColumnInfo columnInfo = PhoenixRuntime.getColumnInfo(table, columnName);
columnInfoList.add(columnInfo);
} catch (ColumnNotFoundException cnfe) {
unresolvedColumnNames.add(columnName.trim());
} catch (AmbiguousColumnException ace) {
unresolvedColumnNames.add(columnName.trim());
}
}
}
// if there exists columns that cannot be resolved, error out.
if (unresolvedColumnNames.size() > 0) {
StringBuilder exceptionMessage = new StringBuilder();
boolean first = true;
exceptionMessage.append("Unable to resolve these column names:\n");
for (String col : unresolvedColumnNames) {
if (first) first = false;
else exceptionMessage.append(",");
exceptionMessage.append(col);
}
exceptionMessage.append("\nAvailable columns with column families:\n");
first = true;
for (PColumn pColumn : table.getColumns()) {
if (first) first = false;
else exceptionMessage.append(",");
exceptionMessage.append(pColumn.toString());
}
throw new SQLException(exceptionMessage.toString());
}
return columnInfoList;
}
@Override
public Void visit(NamedTableNode tableNode) throws SQLException {
String alias = tableNode.getAlias();
TableRef tableRef = createTableRef(tableNode, true);
PTable theTable = tableRef.getTable();
if (alias != null) {
tableMap.put(alias, tableRef);
}
String name = theTable.getName().getString();
// avoid having one name mapped to two identical TableRef.
if (alias == null || !alias.equals(name)) {
tableMap.put(name, tableRef);
}
tables.add(tableRef);
return null;
}
/**
* Decode a byte array value back into the Object values of the primary key constraint. If the
* connection and table are both tenant-specific, the tenant ID column is not expected to have
* been encoded and will not appear in the returned values.
*
* @param conn an open connection
* @param name the full table name
* @param value the value that was encoded with {@link #encodePK(Connection, String, Object[])}
* @return the Object values encoded in the byte array value
* @throws SQLException
*/
public static Object[] decodePK(Connection conn, String name, byte[] value) throws SQLException {
PTable table = getTable(conn, name);
PhoenixConnection pconn = conn.unwrap(PhoenixConnection.class);
int offset =
(table.getBucketNum() == null ? 0 : 1)
+ (table.isMultiTenant() && pconn.getTenantId() != null ? 1 : 0);
RowKeySchema schema = table.getRowKeySchema();
int nValues = schema.getMaxFields() - offset;
Object[] values = new Object[nValues];
ImmutableBytesWritable ptr = new ImmutableBytesWritable();
int i = 0;
schema.iterator(value, ptr, offset);
while (i < nValues && schema.next(ptr, i, value.length) != null) {
values[i] = schema.getField(i).getDataType().toObject(ptr);
i++;
}
return values;
}
public static boolean columnExists(PTable table, String columnName) {
try {
table.getColumn(columnName);
return true;
} catch (ColumnNotFoundException e) {
return false;
} catch (AmbiguousColumnException e) {
return true;
}
}
private static KeyValueSchema buildSchema(PTable table) {
KeyValueSchemaBuilder builder = new KeyValueSchemaBuilder(0);
if (table != null) {
for (PColumn column : table.getColumns()) {
if (!SchemaUtil.isPKColumn(column)) {
builder.addField(column);
}
}
}
return builder.build();
}
protected PTable addDynamicColumns(List<ColumnDef> dynColumns, PTable theTable)
throws SQLException {
if (!dynColumns.isEmpty()) {
List<PColumn> allcolumns = new ArrayList<PColumn>();
List<PColumn> existingColumns = theTable.getColumns();
// Need to skip the salting column, as it's added in the makePTable call below
allcolumns.addAll(
theTable.getBucketNum() == null
? existingColumns
: existingColumns.subList(1, existingColumns.size()));
// Position still based on with the salting columns
int position = existingColumns.size();
PName defaultFamilyName = PNameFactory.newName(SchemaUtil.getEmptyColumnFamily(theTable));
for (ColumnDef dynColumn : dynColumns) {
PName familyName = defaultFamilyName;
PName name = PNameFactory.newName(dynColumn.getColumnDefName().getColumnName());
String family = dynColumn.getColumnDefName().getFamilyName();
if (family != null) {
theTable.getColumnFamily(family); // Verifies that column family exists
familyName = PNameFactory.newName(family);
}
allcolumns.add(
new PColumnImpl(
name,
familyName,
dynColumn.getDataType(),
dynColumn.getMaxLength(),
dynColumn.getScale(),
dynColumn.isNull(),
position,
dynColumn.getSortOrder(),
dynColumn.getArraySize(),
null,
false));
position++;
}
theTable = PTableImpl.makePTable(theTable, allcolumns);
}
return theTable;
}
/**
* Validates that the meta data is valid against the server meta data if we haven't yet done so.
* Otherwise, for every UPSERT VALUES call, we'd need to hit the server to see if the meta data
* has changed.
*
* @param connection
* @return the server time to use for the upsert
* @throws SQLException if the table or any columns no longer exist
*/
private long[] validate() throws SQLException {
int i = 0;
Long scn = connection.getSCN();
PName tenantId = connection.getTenantId();
MetaDataClient client = new MetaDataClient(connection);
long[] timeStamps = new long[this.mutations.size()];
for (Map.Entry<TableRef, Map<ImmutableBytesPtr, Map<PColumn, byte[]>>> entry :
mutations.entrySet()) {
TableRef tableRef = entry.getKey();
long serverTimeStamp = tableRef.getTimeStamp();
PTable table = tableRef.getTable();
// If we're auto committing, we've already validated the schema when we got the
// ColumnResolver,
// so no need to do it again here.
if (!connection.getAutoCommit()) {
MetaDataMutationResult result =
client.updateCache(table.getSchemaName().getString(), table.getTableName().getString());
long timestamp = result.getMutationTime();
if (timestamp != QueryConstants.UNSET_TIMESTAMP) {
serverTimeStamp = timestamp;
if (result.wasUpdated()) {
// TODO: use bitset?
table =
connection
.getMetaDataCache()
.getTable(new PTableKey(tenantId, table.getName().getString()));
PColumn[] columns = new PColumn[table.getColumns().size()];
for (Map.Entry<ImmutableBytesPtr, Map<PColumn, byte[]>> rowEntry :
entry.getValue().entrySet()) {
Map<PColumn, byte[]> valueEntry = rowEntry.getValue();
if (valueEntry != PRow.DELETE_MARKER) {
for (PColumn column : valueEntry.keySet()) {
columns[column.getPosition()] = column;
}
}
}
for (PColumn column : columns) {
if (column != null) {
table
.getColumnFamily(column.getFamilyName().getString())
.getColumn(column.getName().getString());
}
}
tableRef.setTable(table);
}
}
}
timeStamps[i++] =
scn == null
? serverTimeStamp == QueryConstants.UNSET_TIMESTAMP
? HConstants.LATEST_TIMESTAMP
: serverTimeStamp
: scn;
}
return timeStamps;
}
private static SelectStatement prependTenantAndViewConstants(
PTable table, SelectStatement select, String tenantId, Set<PColumn> addViewColumns) {
if ((!table.isMultiTenant() || tenantId == null)
&& table.getViewIndexId() == null
&& addViewColumns.isEmpty()) {
return select;
}
List<AliasedNode> selectNodes =
newArrayListWithCapacity(select.getSelect().size() + 1 + addViewColumns.size());
if (table.isMultiTenant() && tenantId != null) {
selectNodes.add(new AliasedNode(null, new LiteralParseNode(tenantId)));
}
if (table.getViewIndexId() != null) {
selectNodes.add(new AliasedNode(null, new LiteralParseNode(table.getViewIndexId())));
}
selectNodes.addAll(select.getSelect());
for (PColumn column : addViewColumns) {
byte[] byteValue = column.getViewConstant();
Object value = column.getDataType().toObject(byteValue, 0, byteValue.length - 1);
selectNodes.add(new AliasedNode(null, new LiteralParseNode(value)));
}
return SelectStatement.create(select, selectNodes);
}
/**
* Estimate the max key length in bytes of the PK for a given table
*
* @param table the table
* @return the max PK length
*/
public static int estimateKeyLength(PTable table) {
int maxKeyLength = 0;
// Calculate the max length of a key (each part must currently be of a fixed width)
int i = 0;
List<PColumn> columns = table.getPKColumns();
while (i < columns.size()) {
PColumn keyColumn = columns.get(i++);
PDataType type = keyColumn.getDataType();
Integer maxLength = keyColumn.getMaxLength();
maxKeyLength +=
!type.isFixedWidth()
? VAR_LENGTH_ESTIMATE
: maxLength == null ? type.getByteSize() : maxLength;
}
return maxKeyLength;
}
public static int getPKPosition(PTable table, PColumn column) {
// TODO: when PColumn has getPKPosition, use that instead
return table.getPKColumns().indexOf(column);
}
public MutationPlan compile(UpsertStatement upsert) throws SQLException {
final PhoenixConnection connection = statement.getConnection();
ConnectionQueryServices services = connection.getQueryServices();
final int maxSize =
services
.getProps()
.getInt(
QueryServices.MAX_MUTATION_SIZE_ATTRIB,
QueryServicesOptions.DEFAULT_MAX_MUTATION_SIZE);
final ColumnResolver resolver = FromCompiler.getResolverForMutation(upsert, connection);
final TableRef tableRef = resolver.getTables().get(0);
final PTable table = tableRef.getTable();
if (table.getType() == PTableType.VIEW) {
if (table.getViewType().isReadOnly()) {
throw new ReadOnlyTableException(
table.getSchemaName().getString(), table.getTableName().getString());
}
}
boolean isSalted = table.getBucketNum() != null;
final boolean isTenantSpecific = table.isMultiTenant() && connection.getTenantId() != null;
final boolean isSharedViewIndex = table.getViewIndexId() != null;
String tenantId = isTenantSpecific ? connection.getTenantId().getString() : null;
int posOffset = isSalted ? 1 : 0;
// Setup array of column indexes parallel to values that are going to be set
List<ColumnName> columnNodes = upsert.getColumns();
final List<PColumn> allColumns = table.getColumns();
Set<PColumn> addViewColumnsToBe = Collections.emptySet();
Set<PColumn> overlapViewColumnsToBe = Collections.emptySet();
int[] columnIndexesToBe;
int nColumnsToSet = 0;
int[] pkSlotIndexesToBe;
List<PColumn> targetColumns;
if (table.getViewType() == ViewType.UPDATABLE) {
addViewColumnsToBe = Sets.newLinkedHashSetWithExpectedSize(allColumns.size());
for (PColumn column : allColumns) {
if (column.getViewConstant() != null) {
addViewColumnsToBe.add(column);
}
}
}
ImmutableBytesWritable ptr = new ImmutableBytesWritable();
// Allow full row upsert if no columns or only dynamic ones are specified and values count match
if (columnNodes.isEmpty()
|| columnNodes.size() == upsert.getTable().getDynamicColumns().size()) {
nColumnsToSet = allColumns.size() - posOffset;
columnIndexesToBe = new int[nColumnsToSet];
pkSlotIndexesToBe = new int[columnIndexesToBe.length];
targetColumns = Lists.newArrayListWithExpectedSize(columnIndexesToBe.length);
targetColumns.addAll(Collections.<PColumn>nCopies(columnIndexesToBe.length, null));
int minPKPos = 0;
if (isTenantSpecific) {
PColumn tenantColumn = table.getPKColumns().get(minPKPos);
columnIndexesToBe[minPKPos] = tenantColumn.getPosition();
targetColumns.set(minPKPos, tenantColumn);
minPKPos++;
}
if (isSharedViewIndex) {
PColumn indexIdColumn = table.getPKColumns().get(minPKPos);
columnIndexesToBe[minPKPos] = indexIdColumn.getPosition();
targetColumns.set(minPKPos, indexIdColumn);
minPKPos++;
}
for (int i = posOffset, j = 0; i < allColumns.size(); i++) {
PColumn column = allColumns.get(i);
if (SchemaUtil.isPKColumn(column)) {
pkSlotIndexesToBe[i - posOffset] = j + posOffset;
if (j++ < minPKPos) { // Skip, as it's already been set above
continue;
}
minPKPos = 0;
}
columnIndexesToBe[i - posOffset + minPKPos] = i;
targetColumns.set(i - posOffset + minPKPos, column);
}
if (!addViewColumnsToBe.isEmpty()) {
// All view columns overlap in this case
overlapViewColumnsToBe = addViewColumnsToBe;
addViewColumnsToBe = Collections.emptySet();
}
} else {
// Size for worse case
int numColsInUpsert = columnNodes.size();
nColumnsToSet =
numColsInUpsert
+ addViewColumnsToBe.size()
+ (isTenantSpecific ? 1 : 0)
+ +(isSharedViewIndex ? 1 : 0);
columnIndexesToBe = new int[nColumnsToSet];
pkSlotIndexesToBe = new int[columnIndexesToBe.length];
targetColumns = Lists.newArrayListWithExpectedSize(columnIndexesToBe.length);
targetColumns.addAll(Collections.<PColumn>nCopies(columnIndexesToBe.length, null));
Arrays.fill(
columnIndexesToBe,
-1); // TODO: necessary? So we'll get an AIOB exception if it's not replaced
Arrays.fill(
pkSlotIndexesToBe,
-1); // TODO: necessary? So we'll get an AIOB exception if it's not replaced
BitSet pkColumnsSet = new BitSet(table.getPKColumns().size());
int i = 0;
// Add tenant column directly, as we don't want to resolve it as this will fail
if (isTenantSpecific) {
PColumn tenantColumn = table.getPKColumns().get(i + posOffset);
columnIndexesToBe[i] = tenantColumn.getPosition();
pkColumnsSet.set(pkSlotIndexesToBe[i] = i + posOffset);
targetColumns.set(i, tenantColumn);
i++;
}
if (isSharedViewIndex) {
PColumn indexIdColumn = table.getPKColumns().get(i + posOffset);
columnIndexesToBe[i] = indexIdColumn.getPosition();
pkColumnsSet.set(pkSlotIndexesToBe[i] = i + posOffset);
targetColumns.set(i, indexIdColumn);
i++;
}
for (ColumnName colName : columnNodes) {
ColumnRef ref =
resolver.resolveColumn(null, colName.getFamilyName(), colName.getColumnName());
PColumn column = ref.getColumn();
if (IndexUtil.getViewConstantValue(column, ptr)) {
if (overlapViewColumnsToBe.isEmpty()) {
overlapViewColumnsToBe = Sets.newHashSetWithExpectedSize(addViewColumnsToBe.size());
}
nColumnsToSet--;
overlapViewColumnsToBe.add(column);
addViewColumnsToBe.remove(column);
}
columnIndexesToBe[i] = ref.getColumnPosition();
targetColumns.set(i, column);
if (SchemaUtil.isPKColumn(column)) {
pkColumnsSet.set(pkSlotIndexesToBe[i] = ref.getPKSlotPosition());
}
i++;
}
for (PColumn column : addViewColumnsToBe) {
columnIndexesToBe[i] = column.getPosition();
targetColumns.set(i, column);
if (SchemaUtil.isPKColumn(column)) {
pkColumnsSet.set(pkSlotIndexesToBe[i] = SchemaUtil.getPKPosition(table, column));
}
i++;
}
for (i = posOffset; i < table.getPKColumns().size(); i++) {
PColumn pkCol = table.getPKColumns().get(i);
if (!pkColumnsSet.get(i)) {
if (!pkCol.isNullable()) {
throw new ConstraintViolationException(
table.getName().getString()
+ "."
+ pkCol.getName().getString()
+ " may not be null");
}
}
}
}
List<ParseNode> valueNodes = upsert.getValues();
QueryPlan plan = null;
RowProjector rowProjectorToBe = null;
final int nValuesToSet;
boolean sameTable = false;
boolean runOnServer = false;
UpsertingParallelIteratorFactory upsertParallelIteratorFactoryToBe = null;
final boolean isAutoCommit = connection.getAutoCommit();
if (valueNodes == null) {
SelectStatement select = upsert.getSelect();
assert (select != null);
select = SubselectRewriter.flatten(select, connection);
ColumnResolver selectResolver = FromCompiler.getResolverForQuery(select, connection);
select = StatementNormalizer.normalize(select, selectResolver);
select = prependTenantAndViewConstants(table, select, tenantId, addViewColumnsToBe);
sameTable =
select.getFrom().size() == 1 && tableRef.equals(selectResolver.getTables().get(0));
/* We can run the upsert in a coprocessor if:
* 1) from has only 1 table and the into table matches from table
* 2) the select query isn't doing aggregation
* 3) autoCommit is on
* 4) the table is not immutable, as the client is the one that figures out the additional
* puts for index tables.
* 5) no limit clause
* Otherwise, run the query to pull the data from the server
* and populate the MutationState (upto a limit).
*/
runOnServer =
sameTable
&& isAutoCommit
&& !table.isImmutableRows()
&& !select.isAggregate()
&& !select.isDistinct()
&& select.getLimit() == null
&& table.getBucketNum() == null;
ParallelIteratorFactory parallelIteratorFactory;
if (select.isAggregate() || select.isDistinct() || select.getLimit() != null) {
parallelIteratorFactory = null;
} else {
// We can pipeline the upsert select instead of spooling everything to disk first,
// if we don't have any post processing that's required.
parallelIteratorFactory =
upsertParallelIteratorFactoryToBe =
new UpsertingParallelIteratorFactory(connection, tableRef);
}
// If we may be able to run on the server, add a hint that favors using the data table
// if all else is equal.
// TODO: it'd be nice if we could figure out in advance if the PK is potentially changing,
// as this would disallow running on the server. We currently use the row projector we
// get back to figure this out.
HintNode hint = upsert.getHint();
if (!upsert.getHint().hasHint(Hint.USE_INDEX_OVER_DATA_TABLE)) {
hint = HintNode.create(hint, Hint.USE_DATA_OVER_INDEX_TABLE);
}
select = SelectStatement.create(select, hint);
// Pass scan through if same table in upsert and select so that projection is computed
// correctly
// Use optimizer to choose the best plan
plan =
new QueryOptimizer(services)
.optimize(statement, select, selectResolver, targetColumns, parallelIteratorFactory);
runOnServer &= plan.getTableRef().equals(tableRef);
rowProjectorToBe = plan.getProjector();
nValuesToSet = rowProjectorToBe.getColumnCount();
// Cannot auto commit if doing aggregation or topN or salted
// Salted causes problems because the row may end up living on a different region
} else {
nValuesToSet =
valueNodes.size()
+ addViewColumnsToBe.size()
+ (isTenantSpecific ? 1 : 0)
+ (isSharedViewIndex ? 1 : 0);
}
final RowProjector projector = rowProjectorToBe;
final UpsertingParallelIteratorFactory upsertParallelIteratorFactory =
upsertParallelIteratorFactoryToBe;
final QueryPlan queryPlan = plan;
// Resize down to allow a subset of columns to be specifiable
if (columnNodes.isEmpty() && columnIndexesToBe.length >= nValuesToSet) {
nColumnsToSet = nValuesToSet;
columnIndexesToBe = Arrays.copyOf(columnIndexesToBe, nValuesToSet);
pkSlotIndexesToBe = Arrays.copyOf(pkSlotIndexesToBe, nValuesToSet);
}
if (nValuesToSet != nColumnsToSet) {
throw new SQLExceptionInfo.Builder(SQLExceptionCode.UPSERT_COLUMN_NUMBERS_MISMATCH)
.setMessage(
"Numbers of columns: " + nColumnsToSet + ". Number of values: " + nValuesToSet)
.build()
.buildException();
}
final int[] columnIndexes = columnIndexesToBe;
final int[] pkSlotIndexes = pkSlotIndexesToBe;
final Set<PColumn> addViewColumns = addViewColumnsToBe;
final Set<PColumn> overlapViewColumns = overlapViewColumnsToBe;
// TODO: break this up into multiple functions
////////////////////////////////////////////////////////////////////
// UPSERT SELECT
/////////////////////////////////////////////////////////////////////
if (valueNodes == null) {
// Before we re-order, check that for updatable view columns
// the projected expression either matches the column name or
// is a constant with the same required value.
throwIfNotUpdatable(tableRef, overlapViewColumnsToBe, targetColumns, projector, sameTable);
////////////////////////////////////////////////////////////////////
// UPSERT SELECT run server-side (maybe)
/////////////////////////////////////////////////////////////////////
if (runOnServer) {
// At most this array will grow bigger by the number of PK columns
int[] allColumnsIndexes = Arrays.copyOf(columnIndexes, columnIndexes.length + nValuesToSet);
int[] reverseColumnIndexes = new int[table.getColumns().size()];
List<Expression> projectedExpressions =
Lists.newArrayListWithExpectedSize(reverseColumnIndexes.length);
Arrays.fill(reverseColumnIndexes, -1);
for (int i = 0; i < nValuesToSet; i++) {
projectedExpressions.add(projector.getColumnProjector(i).getExpression());
reverseColumnIndexes[columnIndexes[i]] = i;
}
/*
* Order projected columns and projected expressions with PK columns
* leading order by slot position
*/
int offset = table.getBucketNum() == null ? 0 : 1;
for (int i = 0; i < table.getPKColumns().size() - offset; i++) {
PColumn column = table.getPKColumns().get(i + offset);
int pos = reverseColumnIndexes[column.getPosition()];
if (pos == -1) {
// Last PK column may be fixed width and nullable
// We don't want to insert a null expression b/c
// it's not valid to set a fixed width type to null.
if (column.getDataType().isFixedWidth()) {
continue;
}
// Add literal null for missing PK columns
pos = projectedExpressions.size();
Expression literalNull =
LiteralExpression.newConstant(null, column.getDataType(), true);
projectedExpressions.add(literalNull);
allColumnsIndexes[pos] = column.getPosition();
}
// Swap select expression at pos with i
Collections.swap(projectedExpressions, i, pos);
// Swap column indexes and reverse column indexes too
int tempPos = allColumnsIndexes[i];
allColumnsIndexes[i] = allColumnsIndexes[pos];
allColumnsIndexes[pos] = tempPos;
reverseColumnIndexes[tempPos] = reverseColumnIndexes[i];
reverseColumnIndexes[i] = i;
}
// If any pk slots are changing, be conservative and don't run this server side.
// If the row ends up living in a different region, we'll get an error otherwise.
for (int i = 0; i < table.getPKColumns().size(); i++) {
PColumn column = table.getPKColumns().get(i);
Expression source = projectedExpressions.get(i);
if (source == null
|| !source.equals(
new ColumnRef(tableRef, column.getPosition()).newColumnExpression())) {
// TODO: we could check the region boundaries to see if the pk will still be in it.
runOnServer = false; // bail on running server side, since PK may be changing
break;
}
}
////////////////////////////////////////////////////////////////////
// UPSERT SELECT run server-side
/////////////////////////////////////////////////////////////////////
if (runOnServer) {
// Iterate through columns being projected
List<PColumn> projectedColumns =
Lists.newArrayListWithExpectedSize(projectedExpressions.size());
for (int i = 0; i < projectedExpressions.size(); i++) {
// Must make new column if position has changed
PColumn column = allColumns.get(allColumnsIndexes[i]);
projectedColumns.add(column.getPosition() == i ? column : new PColumnImpl(column, i));
}
// Build table from projectedColumns
PTable projectedTable = PTableImpl.makePTable(table, projectedColumns);
SelectStatement select =
SelectStatement.create(SelectStatement.COUNT_ONE, upsert.getHint());
final RowProjector aggProjector =
ProjectionCompiler.compile(queryPlan.getContext(), select, GroupBy.EMPTY_GROUP_BY);
/*
* Transfer over PTable representing subset of columns selected, but all PK columns.
* Move columns setting PK first in pkSlot order, adding LiteralExpression of null for any missing ones.
* Transfer over List<Expression> for projection.
* In region scan, evaluate expressions in order, collecting first n columns for PK and collection non PK in mutation Map
* Create the PRow and get the mutations, adding them to the batch
*/
final StatementContext context = queryPlan.getContext();
final Scan scan = context.getScan();
scan.setAttribute(
BaseScannerRegionObserver.UPSERT_SELECT_TABLE,
UngroupedAggregateRegionObserver.serialize(projectedTable));
scan.setAttribute(
BaseScannerRegionObserver.UPSERT_SELECT_EXPRS,
UngroupedAggregateRegionObserver.serialize(projectedExpressions));
// Ignore order by - it has no impact
final QueryPlan aggPlan =
new AggregatePlan(
context,
select,
tableRef,
aggProjector,
null,
OrderBy.EMPTY_ORDER_BY,
null,
GroupBy.EMPTY_GROUP_BY,
null);
return new MutationPlan() {
@Override
public PhoenixConnection getConnection() {
return connection;
}
@Override
public ParameterMetaData getParameterMetaData() {
return queryPlan.getContext().getBindManager().getParameterMetaData();
}
@Override
public StatementContext getContext() {
return queryPlan.getContext();
}
@Override
public MutationState execute() throws SQLException {
ImmutableBytesWritable ptr = context.getTempPtr();
tableRef.getTable().getIndexMaintainers(ptr);
ServerCache cache = null;
try {
if (ptr.getLength() > 0) {
IndexMetaDataCacheClient client =
new IndexMetaDataCacheClient(connection, tableRef);
cache = client.addIndexMetadataCache(context.getScanRanges(), ptr);
byte[] uuidValue = cache.getId();
scan.setAttribute(PhoenixIndexCodec.INDEX_UUID, uuidValue);
}
ResultIterator iterator = aggPlan.iterator();
try {
Tuple row = iterator.next();
final long mutationCount =
(Long) aggProjector.getColumnProjector(0).getValue(row, PDataType.LONG, ptr);
return new MutationState(maxSize, connection) {
@Override
public long getUpdateCount() {
return mutationCount;
}
};
} finally {
iterator.close();
}
} finally {
if (cache != null) {
cache.close();
}
}
}
@Override
public ExplainPlan getExplainPlan() throws SQLException {
List<String> queryPlanSteps = aggPlan.getExplainPlan().getPlanSteps();
List<String> planSteps =
Lists.newArrayListWithExpectedSize(queryPlanSteps.size() + 1);
planSteps.add("UPSERT ROWS");
planSteps.addAll(queryPlanSteps);
return new ExplainPlan(planSteps);
}
};
}
}
////////////////////////////////////////////////////////////////////
// UPSERT SELECT run client-side
/////////////////////////////////////////////////////////////////////
return new MutationPlan() {
@Override
public PhoenixConnection getConnection() {
return connection;
}
@Override
public ParameterMetaData getParameterMetaData() {
return queryPlan.getContext().getBindManager().getParameterMetaData();
}
@Override
public StatementContext getContext() {
return queryPlan.getContext();
}
@Override
public MutationState execute() throws SQLException {
ResultIterator iterator = queryPlan.iterator();
if (upsertParallelIteratorFactory == null) {
return upsertSelect(
statement, tableRef, projector, iterator, columnIndexes, pkSlotIndexes);
}
upsertParallelIteratorFactory.setRowProjector(projector);
upsertParallelIteratorFactory.setColumnIndexes(columnIndexes);
upsertParallelIteratorFactory.setPkSlotIndexes(pkSlotIndexes);
Tuple tuple;
long totalRowCount = 0;
while ((tuple = iterator.next()) != null) { // Runs query
Cell kv = tuple.getValue(0);
totalRowCount +=
PDataType.LONG
.getCodec()
.decodeLong(kv.getValueArray(), kv.getValueOffset(), SortOrder.getDefault());
}
// Return total number of rows that have been updated. In the case of auto commit being
// off
// the mutations will all be in the mutation state of the current connection.
return new MutationState(maxSize, statement.getConnection(), totalRowCount);
}
@Override
public ExplainPlan getExplainPlan() throws SQLException {
List<String> queryPlanSteps = queryPlan.getExplainPlan().getPlanSteps();
List<String> planSteps = Lists.newArrayListWithExpectedSize(queryPlanSteps.size() + 1);
planSteps.add("UPSERT SELECT");
planSteps.addAll(queryPlanSteps);
return new ExplainPlan(planSteps);
}
};
}
////////////////////////////////////////////////////////////////////
// UPSERT VALUES
/////////////////////////////////////////////////////////////////////
int nodeIndex = 0;
// initialze values with constant byte values first
final byte[][] values = new byte[nValuesToSet][];
if (isTenantSpecific) {
values[nodeIndex++] = connection.getTenantId().getBytes();
}
if (isSharedViewIndex) {
values[nodeIndex++] = MetaDataUtil.getViewIndexIdDataType().toBytes(table.getViewIndexId());
}
final int nodeIndexOffset = nodeIndex;
// Allocate array based on size of all columns in table,
// since some values may not be set (if they're nullable).
final StatementContext context = new StatementContext(statement, resolver, new Scan());
UpsertValuesCompiler expressionBuilder = new UpsertValuesCompiler(context);
final List<Expression> constantExpressions =
Lists.newArrayListWithExpectedSize(valueNodes.size());
// First build all the expressions, as with sequences we want to collect them all first
// and initialize them in one batch
for (ParseNode valueNode : valueNodes) {
if (!valueNode.isStateless()) {
throw new SQLExceptionInfo.Builder(SQLExceptionCode.VALUE_IN_UPSERT_NOT_CONSTANT)
.build()
.buildException();
}
PColumn column = allColumns.get(columnIndexes[nodeIndex]);
expressionBuilder.setColumn(column);
Expression expression = valueNode.accept(expressionBuilder);
if (expression.getDataType() != null
&& !expression.getDataType().isCastableTo(column.getDataType())) {
throw TypeMismatchException.newException(
expression.getDataType(),
column.getDataType(),
"expression: " + expression.toString() + " in column " + column);
}
constantExpressions.add(expression);
nodeIndex++;
}
return new MutationPlan() {
@Override
public PhoenixConnection getConnection() {
return connection;
}
@Override
public ParameterMetaData getParameterMetaData() {
return context.getBindManager().getParameterMetaData();
}
@Override
public StatementContext getContext() {
return context;
}
@Override
public MutationState execute() throws SQLException {
ImmutableBytesWritable ptr = context.getTempPtr();
final SequenceManager sequenceManager = context.getSequenceManager();
// Next evaluate all the expressions
int nodeIndex = nodeIndexOffset;
Tuple tuple =
sequenceManager.getSequenceCount() == 0 ? null : sequenceManager.newSequenceTuple(null);
for (Expression constantExpression : constantExpressions) {
PColumn column = allColumns.get(columnIndexes[nodeIndex]);
constantExpression.evaluate(tuple, ptr);
Object value = null;
if (constantExpression.getDataType() != null) {
value =
constantExpression
.getDataType()
.toObject(
ptr,
constantExpression.getSortOrder(),
constantExpression.getMaxLength(),
constantExpression.getScale());
if (!constantExpression.getDataType().isCoercibleTo(column.getDataType(), value)) {
throw TypeMismatchException.newException(
constantExpression.getDataType(),
column.getDataType(),
"expression: " + constantExpression.toString() + " in column " + column);
}
if (!column
.getDataType()
.isSizeCompatible(
ptr,
value,
constantExpression.getDataType(),
constantExpression.getMaxLength(),
constantExpression.getScale(),
column.getMaxLength(),
column.getScale())) {
throw new SQLExceptionInfo.Builder(SQLExceptionCode.DATA_EXCEEDS_MAX_CAPACITY)
.setColumnName(column.getName().getString())
.setMessage("value=" + constantExpression.toString())
.build()
.buildException();
}
}
column
.getDataType()
.coerceBytes(
ptr,
value,
constantExpression.getDataType(),
constantExpression.getMaxLength(),
constantExpression.getScale(),
constantExpression.getSortOrder(),
column.getMaxLength(),
column.getScale(),
column.getSortOrder());
if (overlapViewColumns.contains(column)
&& Bytes.compareTo(
ptr.get(),
ptr.getOffset(),
ptr.getLength(),
column.getViewConstant(),
0,
column.getViewConstant().length - 1)
!= 0) {
throw new SQLExceptionInfo.Builder(SQLExceptionCode.CANNOT_UPDATE_VIEW_COLUMN)
.setColumnName(column.getName().getString())
.setMessage("value=" + constantExpression.toString())
.build()
.buildException();
}
values[nodeIndex] = ByteUtil.copyKeyBytesIfNecessary(ptr);
nodeIndex++;
}
// Add columns based on view
for (PColumn column : addViewColumns) {
if (IndexUtil.getViewConstantValue(column, ptr)) {
values[nodeIndex++] = ByteUtil.copyKeyBytesIfNecessary(ptr);
} else {
throw new IllegalStateException();
}
}
Map<ImmutableBytesPtr, Map<PColumn, byte[]>> mutation = Maps.newHashMapWithExpectedSize(1);
setValues(values, pkSlotIndexes, columnIndexes, tableRef.getTable(), mutation);
return new MutationState(tableRef, mutation, 0, maxSize, connection);
}
@Override
public ExplainPlan getExplainPlan() throws SQLException {
List<String> planSteps = Lists.newArrayListWithExpectedSize(2);
if (context.getSequenceManager().getSequenceCount() > 0) {
planSteps.add(
"CLIENT RESERVE " + context.getSequenceManager().getSequenceCount() + " SEQUENCES");
}
planSteps.add("PUT SINGLE ROW");
return new ExplainPlan(planSteps);
}
};
}
private static MutationState upsertSelect(
PhoenixStatement statement,
TableRef tableRef,
RowProjector projector,
ResultIterator iterator,
int[] columnIndexes,
int[] pkSlotIndexes)
throws SQLException {
try {
PhoenixConnection connection = statement.getConnection();
ConnectionQueryServices services = connection.getQueryServices();
int maxSize =
services
.getProps()
.getInt(
QueryServices.MAX_MUTATION_SIZE_ATTRIB,
QueryServicesOptions.DEFAULT_MAX_MUTATION_SIZE);
int batchSize = Math.min(connection.getMutateBatchSize(), maxSize);
boolean isAutoCommit = connection.getAutoCommit();
byte[][] values = new byte[columnIndexes.length][];
int rowCount = 0;
Map<ImmutableBytesPtr, Map<PColumn, byte[]>> mutation =
Maps.newHashMapWithExpectedSize(batchSize);
PTable table = tableRef.getTable();
ResultSet rs = new PhoenixResultSet(iterator, projector, statement);
ImmutableBytesWritable ptr = new ImmutableBytesWritable();
while (rs.next()) {
for (int i = 0; i < values.length; i++) {
PColumn column = table.getColumns().get(columnIndexes[i]);
byte[] bytes = rs.getBytes(i + 1);
ptr.set(bytes == null ? ByteUtil.EMPTY_BYTE_ARRAY : bytes);
Object value = rs.getObject(i + 1);
int rsPrecision = rs.getMetaData().getPrecision(i + 1);
Integer precision = rsPrecision == 0 ? null : rsPrecision;
int rsScale = rs.getMetaData().getScale(i + 1);
Integer scale = rsScale == 0 ? null : rsScale;
// We are guaranteed that the two column will have compatible types,
// as we checked that before.
if (!column
.getDataType()
.isSizeCompatible(
ptr,
value,
column.getDataType(),
precision,
scale,
column.getMaxLength(),
column.getScale())) {
throw new SQLExceptionInfo.Builder(SQLExceptionCode.DATA_EXCEEDS_MAX_CAPACITY)
.setColumnName(column.getName().getString())
.setMessage("value=" + column.getDataType().toStringLiteral(ptr, null))
.build()
.buildException();
}
column
.getDataType()
.coerceBytes(
ptr,
value,
column.getDataType(),
precision,
scale,
SortOrder.getDefault(),
column.getMaxLength(),
column.getScale(),
column.getSortOrder());
values[i] = ByteUtil.copyKeyBytesIfNecessary(ptr);
}
setValues(values, pkSlotIndexes, columnIndexes, table, mutation);
rowCount++;
// Commit a batch if auto commit is true and we're at our batch size
if (isAutoCommit && rowCount % batchSize == 0) {
MutationState state = new MutationState(tableRef, mutation, 0, maxSize, connection);
connection.getMutationState().join(state);
connection.commit();
mutation.clear();
}
}
// If auto commit is true, this last batch will be committed upon return
return new MutationState(
tableRef, mutation, rowCount / batchSize * batchSize, maxSize, connection);
} finally {
iterator.close();
}
}
@SuppressWarnings("deprecation")
public void commit() throws SQLException {
int i = 0;
byte[] tenantId = connection.getTenantId() == null ? null : connection.getTenantId().getBytes();
long[] serverTimeStamps = validate();
Iterator<Map.Entry<TableRef, Map<ImmutableBytesPtr, Map<PColumn, byte[]>>>> iterator =
this.mutations.entrySet().iterator();
List<Map.Entry<TableRef, Map<ImmutableBytesPtr, Map<PColumn, byte[]>>>> committedList =
Lists.newArrayListWithCapacity(this.mutations.size());
// add tracing for this operation
TraceScope trace = Tracing.startNewSpan(connection, "Committing mutations to tables");
Span span = trace.getSpan();
while (iterator.hasNext()) {
Map.Entry<TableRef, Map<ImmutableBytesPtr, Map<PColumn, byte[]>>> entry = iterator.next();
Map<ImmutableBytesPtr, Map<PColumn, byte[]>> valuesMap = entry.getValue();
TableRef tableRef = entry.getKey();
PTable table = tableRef.getTable();
table.getIndexMaintainers(tempPtr);
boolean hasIndexMaintainers = tempPtr.getLength() > 0;
boolean isDataTable = true;
long serverTimestamp = serverTimeStamps[i++];
Iterator<Pair<byte[], List<Mutation>>> mutationsIterator =
addRowMutations(tableRef, valuesMap, serverTimestamp, false);
while (mutationsIterator.hasNext()) {
Pair<byte[], List<Mutation>> pair = mutationsIterator.next();
byte[] htableName = pair.getFirst();
List<Mutation> mutations = pair.getSecond();
// create a span per target table
// TODO maybe we can be smarter about the table name to string here?
Span child =
Tracing.child(span, "Writing mutation batch for table: " + Bytes.toString(htableName));
int retryCount = 0;
boolean shouldRetry = false;
do {
ServerCache cache = null;
if (hasIndexMaintainers && isDataTable) {
byte[] attribValue = null;
byte[] uuidValue;
if (IndexMetaDataCacheClient.useIndexMetadataCache(
connection, mutations, tempPtr.getLength())) {
IndexMetaDataCacheClient client = new IndexMetaDataCacheClient(connection, tableRef);
cache = client.addIndexMetadataCache(mutations, tempPtr);
child.addTimelineAnnotation("Updated index metadata cache");
uuidValue = cache.getId();
// If we haven't retried yet, retry for this case only, as it's possible that
// a split will occur after we send the index metadata cache to all known
// region servers.
shouldRetry = true;
} else {
attribValue = ByteUtil.copyKeyBytesIfNecessary(tempPtr);
uuidValue = ServerCacheClient.generateId();
}
// Either set the UUID to be able to access the index metadata from the cache
// or set the index metadata directly on the Mutation
for (Mutation mutation : mutations) {
if (tenantId != null) {
mutation.setAttribute(PhoenixRuntime.TENANT_ID_ATTRIB, tenantId);
}
mutation.setAttribute(PhoenixIndexCodec.INDEX_UUID, uuidValue);
if (attribValue != null) {
mutation.setAttribute(PhoenixIndexCodec.INDEX_MD, attribValue);
}
}
}
SQLException sqlE = null;
HTableInterface hTable = connection.getQueryServices().getTable(htableName);
try {
if (logger.isDebugEnabled()) logMutationSize(hTable, mutations);
long startTime = System.currentTimeMillis();
child.addTimelineAnnotation("Attempt " + retryCount);
hTable.batch(mutations);
child.stop();
shouldRetry = false;
if (logger.isDebugEnabled())
logger.debug(
"Total time for batch call of "
+ mutations.size()
+ " mutations into "
+ table.getName().getString()
+ ": "
+ (System.currentTimeMillis() - startTime)
+ " ms");
committedList.add(entry);
} catch (Exception e) {
SQLException inferredE = ServerUtil.parseServerExceptionOrNull(e);
if (inferredE != null) {
if (shouldRetry
&& retryCount == 0
&& inferredE.getErrorCode()
== SQLExceptionCode.INDEX_METADATA_NOT_FOUND.getErrorCode()) {
// Swallow this exception once, as it's possible that we split after sending the
// index metadata
// and one of the region servers doesn't have it. This will cause it to have it the
// next go around.
// If it fails again, we don't retry.
String msg =
"Swallowing exception and retrying after clearing meta cache on connection. "
+ inferredE;
logger.warn(msg);
connection.getQueryServices().clearTableRegionCache(htableName);
// add a new child span as this one failed
child.addTimelineAnnotation(msg);
child.stop();
child = Tracing.child(span, "Failed batch, attempting retry");
continue;
}
e = inferredE;
}
// Throw to client with both what was committed so far and what is left to be committed.
// That way, client can either undo what was done or try again with what was not done.
sqlE =
new CommitException(
e,
this,
new MutationState(
committedList, this.sizeOffset, this.maxSize, this.connection));
} finally {
try {
hTable.close();
} catch (IOException e) {
if (sqlE != null) {
sqlE.setNextException(ServerUtil.parseServerException(e));
} else {
sqlE = ServerUtil.parseServerException(e);
}
} finally {
try {
if (cache != null) {
cache.close();
}
} finally {
if (sqlE != null) {
throw sqlE;
}
}
}
}
} while (shouldRetry && retryCount++ < 1);
isDataTable = false;
}
numRows -= entry.getValue().size();
iterator.remove(); // Remove batches as we process them
}
trace.close();
assert (numRows == 0);
assert (this.mutations.isEmpty());
}
public ServerCache addServerCache(
ScanRanges keyRanges,
final ImmutableBytesWritable cachePtr,
final byte[] txState,
final ServerCacheFactory cacheFactory,
final TableRef cacheUsingTableRef)
throws SQLException {
ConnectionQueryServices services = connection.getQueryServices();
MemoryChunk chunk = services.getMemoryManager().allocate(cachePtr.getLength());
List<Closeable> closeables = new ArrayList<Closeable>();
closeables.add(chunk);
ServerCache hashCacheSpec = null;
SQLException firstException = null;
final byte[] cacheId = generateId();
/** Execute EndPoint in parallel on each server to send compressed hash cache */
// TODO: generalize and package as a per region server EndPoint caller
// (ideally this would be functionality provided by the coprocessor framework)
boolean success = false;
ExecutorService executor = services.getExecutor();
List<Future<Boolean>> futures = Collections.emptyList();
try {
final PTable cacheUsingTable = cacheUsingTableRef.getTable();
List<HRegionLocation> locations =
services.getAllTableRegions(cacheUsingTable.getPhysicalName().getBytes());
int nRegions = locations.size();
// Size these based on worst case
futures = new ArrayList<Future<Boolean>>(nRegions);
Set<HRegionLocation> servers = new HashSet<HRegionLocation>(nRegions);
for (HRegionLocation entry : locations) {
// Keep track of servers we've sent to and only send once
byte[] regionStartKey = entry.getRegionInfo().getStartKey();
byte[] regionEndKey = entry.getRegionInfo().getEndKey();
if (!servers.contains(entry)
&& keyRanges.intersects(
regionStartKey,
regionEndKey,
cacheUsingTable.getIndexType() == IndexType.LOCAL
? ScanUtil.getRowKeyOffset(regionStartKey, regionEndKey)
: 0,
true)) {
// Call RPC once per server
servers.add(entry);
if (LOG.isDebugEnabled()) {
LOG.debug(
addCustomAnnotations("Adding cache entry to be sent for " + entry, connection));
}
final byte[] key = entry.getRegionInfo().getStartKey();
final HTableInterface htable =
services.getTable(cacheUsingTableRef.getTable().getPhysicalName().getBytes());
closeables.add(htable);
futures.add(
executor.submit(
new JobCallable<Boolean>() {
@Override
public Boolean call() throws Exception {
final Map<byte[], AddServerCacheResponse> results;
try {
results =
htable.coprocessorService(
ServerCachingService.class,
key,
key,
new Batch.Call<ServerCachingService, AddServerCacheResponse>() {
@Override
public AddServerCacheResponse call(ServerCachingService instance)
throws IOException {
ServerRpcController controller = new ServerRpcController();
BlockingRpcCallback<AddServerCacheResponse> rpcCallback =
new BlockingRpcCallback<AddServerCacheResponse>();
AddServerCacheRequest.Builder builder =
AddServerCacheRequest.newBuilder();
if (connection.getTenantId() != null) {
try {
byte[] tenantIdBytes =
ScanUtil.getTenantIdBytes(
cacheUsingTable.getRowKeySchema(),
cacheUsingTable.getBucketNum() != null,
connection.getTenantId(),
cacheUsingTable.isMultiTenant());
builder.setTenantId(ByteStringer.wrap(tenantIdBytes));
} catch (SQLException e) {
new IOException(e);
}
}
builder.setCacheId(ByteStringer.wrap(cacheId));
builder.setCachePtr(
org.apache.phoenix.protobuf.ProtobufUtil.toProto(cachePtr));
ServerCacheFactoryProtos.ServerCacheFactory.Builder
svrCacheFactoryBuider =
ServerCacheFactoryProtos.ServerCacheFactory
.newBuilder();
svrCacheFactoryBuider.setClassName(
cacheFactory.getClass().getName());
builder.setCacheFactory(svrCacheFactoryBuider.build());
builder.setTxState(HBaseZeroCopyByteString.wrap(txState));
instance.addServerCache(
controller, builder.build(), rpcCallback);
if (controller.getFailedOn() != null) {
throw controller.getFailedOn();
}
return rpcCallback.get();
}
});
} catch (Throwable t) {
throw new Exception(t);
}
if (results != null && results.size() == 1) {
return results.values().iterator().next().getReturn();
}
return false;
}
/**
* Defines the grouping for round robin behavior. All threads spawned to process
* this scan will be grouped together and time sliced with other simultaneously
* executing parallel scans.
*/
@Override
public Object getJobId() {
return ServerCacheClient.this;
}
@Override
public TaskExecutionMetricsHolder getTaskExecutionMetric() {
return NO_OP_INSTANCE;
}
}));
} else {
if (LOG.isDebugEnabled()) {
LOG.debug(
addCustomAnnotations(
"NOT adding cache entry to be sent for "
+ entry
+ " since one already exists for that entry",
connection));
}
}
}
hashCacheSpec = new ServerCache(cacheId, servers, cachePtr.getLength());
// Execute in parallel
int timeoutMs =
services
.getProps()
.getInt(
QueryServices.THREAD_TIMEOUT_MS_ATTRIB,
QueryServicesOptions.DEFAULT_THREAD_TIMEOUT_MS);
for (Future<Boolean> future : futures) {
future.get(timeoutMs, TimeUnit.MILLISECONDS);
}
cacheUsingTableRefMap.put(Bytes.mapKey(cacheId), cacheUsingTableRef);
success = true;
} catch (SQLException e) {
firstException = e;
} catch (Exception e) {
firstException = new SQLException(e);
} finally {
try {
if (!success) {
SQLCloseables.closeAllQuietly(Collections.singletonList(hashCacheSpec));
for (Future<Boolean> future : futures) {
future.cancel(true);
}
}
} finally {
try {
Closeables.closeAll(closeables);
} catch (IOException e) {
if (firstException == null) {
firstException = new SQLException(e);
}
} finally {
if (firstException != null) {
throw firstException;
}
}
}
}
if (LOG.isDebugEnabled()) {
LOG.debug(
addCustomAnnotations("Cache " + cacheId + " successfully added to servers.", connection));
}
return hashCacheSpec;
}
public static boolean hasRowTimestampColumn(PTable table) {
return table.getRowTimestampColPos() > 0;
}
public static boolean isMetaTable(PTable table) {
return PhoenixDatabaseMetaData.SYSTEM_CATALOG_SCHEMA.equals(table.getSchemaName().getString())
&& PhoenixDatabaseMetaData.SYSTEM_CATALOG_TABLE.equals(table.getTableName().getString());
}
