use of io.trino.spi.connector.RetryMode.NO_RETRIES in project trino by trinodb.
the class HiveMetadata method beginInsert.
@Override
public HiveInsertTableHandle beginInsert(ConnectorSession session, ConnectorTableHandle tableHandle, List<ColumnHandle> columns, RetryMode retryMode) {
SchemaTableName tableName = ((HiveTableHandle) tableHandle).getSchemaTableName();
Table table = metastore.getTable(tableName.getSchemaName(), tableName.getTableName()).orElseThrow(() -> new TableNotFoundException(tableName));
checkTableIsWritable(table, writesToNonManagedTablesEnabled);
for (Column column : table.getDataColumns()) {
if (!isWritableType(column.getType())) {
throw new TrinoException(NOT_SUPPORTED, format("Inserting into Hive table %s with column type %s not supported", tableName, column.getType()));
}
}
boolean isTransactional = isTransactionalTable(table.getParameters());
if (isTransactional && retryMode != NO_RETRIES) {
throw new TrinoException(NOT_SUPPORTED, "Inserting into Hive transactional tables is not supported with query retries enabled");
}
if (isTransactional && !autoCommit) {
throw new TrinoException(NOT_SUPPORTED, "Inserting into Hive transactional tables is not supported in explicit transactions (use autocommit mode)");
}
if (isSparkBucketedTable(table)) {
throw new TrinoException(NOT_SUPPORTED, "Inserting into Spark bucketed tables is not supported");
}
List<HiveColumnHandle> handles = hiveColumnHandles(table, typeManager, getTimestampPrecision(session)).stream().filter(columnHandle -> !columnHandle.isHidden()).collect(toImmutableList());
HiveStorageFormat tableStorageFormat = extractHiveStorageFormat(table);
Optional.ofNullable(table.getParameters().get(SKIP_HEADER_COUNT_KEY)).map(Integer::parseInt).ifPresent(headerSkipCount -> {
if (headerSkipCount > 1) {
throw new TrinoException(NOT_SUPPORTED, format("Inserting into Hive table with value of %s property greater than 1 is not supported", SKIP_HEADER_COUNT_KEY));
}
});
if (table.getParameters().containsKey(SKIP_FOOTER_COUNT_KEY)) {
throw new TrinoException(NOT_SUPPORTED, format("Inserting into Hive table with %s property not supported", SKIP_FOOTER_COUNT_KEY));
}
LocationHandle locationHandle = locationService.forExistingTable(metastore, session, table);
AcidTransaction transaction = isTransactional ? metastore.beginInsert(session, table) : NO_ACID_TRANSACTION;
HiveInsertTableHandle result = new HiveInsertTableHandle(tableName.getSchemaName(), tableName.getTableName(), handles, metastore.generatePageSinkMetadata(tableName), locationHandle, table.getStorage().getBucketProperty(), tableStorageFormat, isRespectTableFormat(session) ? tableStorageFormat : getHiveStorageFormat(session), transaction, retryMode != NO_RETRIES);
WriteInfo writeInfo = locationService.getQueryWriteInfo(locationHandle);
if (getInsertExistingPartitionsBehavior(session) == InsertExistingPartitionsBehavior.OVERWRITE && writeInfo.getWriteMode() == DIRECT_TO_TARGET_EXISTING_DIRECTORY) {
if (isTransactional) {
throw new TrinoException(NOT_SUPPORTED, "Overwriting existing partition in transactional tables doesn't support DIRECT_TO_TARGET_EXISTING_DIRECTORY write mode");
}
// Partition overwrite operation is nonatomic thus can't and shouldn't be used in non autocommit context.
if (!autoCommit) {
throw new TrinoException(NOT_SUPPORTED, "Overwriting existing partition in non auto commit context doesn't support DIRECT_TO_TARGET_EXISTING_DIRECTORY write mode");
}
}
metastore.declareIntentionToWrite(session, writeInfo.getWriteMode(), writeInfo.getWritePath(), tableName);
return result;
}
use of io.trino.spi.connector.RetryMode.NO_RETRIES in project trino by trinodb.
the class IcebergMetadata method finishInsert.
@Override
public Optional<ConnectorOutputMetadata> finishInsert(ConnectorSession session, ConnectorInsertTableHandle insertHandle, Collection<Slice> fragments, Collection<ComputedStatistics> computedStatistics) {
IcebergWritableTableHandle table = (IcebergWritableTableHandle) insertHandle;
Table icebergTable = transaction.table();
List<CommitTaskData> commitTasks = fragments.stream().map(slice -> commitTaskCodec.fromJson(slice.getBytes())).collect(toImmutableList());
Type[] partitionColumnTypes = icebergTable.spec().fields().stream().map(field -> field.transform().getResultType(icebergTable.schema().findType(field.sourceId()))).toArray(Type[]::new);
AppendFiles appendFiles = transaction.newAppend();
ImmutableSet.Builder<String> writtenFiles = ImmutableSet.builder();
for (CommitTaskData task : commitTasks) {
DataFiles.Builder builder = DataFiles.builder(icebergTable.spec()).withPath(task.getPath()).withFileSizeInBytes(task.getFileSizeInBytes()).withFormat(table.getFileFormat().toIceberg()).withMetrics(task.getMetrics().metrics());
if (!icebergTable.spec().fields().isEmpty()) {
String partitionDataJson = task.getPartitionDataJson().orElseThrow(() -> new VerifyException("No partition data for partitioned table"));
builder.withPartition(PartitionData.fromJson(partitionDataJson, partitionColumnTypes));
}
appendFiles.appendFile(builder.build());
writtenFiles.add(task.getPath());
}
// try to leave as little garbage as possible behind
if (table.getRetryMode() != NO_RETRIES) {
cleanExtraOutputFiles(session, writtenFiles.build());
}
appendFiles.commit();
transaction.commitTransaction();
transaction = null;
return Optional.of(new HiveWrittenPartitions(commitTasks.stream().map(CommitTaskData::getPath).collect(toImmutableList())));
}
use of io.trino.spi.connector.RetryMode.NO_RETRIES in project trino by trinodb.
the class AbstractTestHive method doTestTransactionDeleteInsert.
private void doTestTransactionDeleteInsert(HiveStorageFormat storageFormat, SchemaTableName tableName, Domain domainToDrop, MaterializedResult insertData, MaterializedResult expectedData, TransactionDeleteInsertTestTag tag, boolean expectQuerySucceed, Optional<ConflictTrigger> conflictTrigger) throws Exception {
Path writePath = null;
Path targetPath = null;
try (Transaction transaction = newTransaction()) {
try {
ConnectorMetadata metadata = transaction.getMetadata();
ConnectorTableHandle tableHandle = getTableHandle(metadata, tableName);
ConnectorSession session;
rollbackIfEquals(tag, ROLLBACK_RIGHT_AWAY);
// Query 1: delete
session = newSession();
HiveColumnHandle dsColumnHandle = (HiveColumnHandle) metadata.getColumnHandles(session, tableHandle).get("pk2");
TupleDomain<ColumnHandle> tupleDomain = TupleDomain.withColumnDomains(ImmutableMap.of(dsColumnHandle, domainToDrop));
Constraint constraint = new Constraint(tupleDomain, tupleDomain.asPredicate(), tupleDomain.getDomains().orElseThrow().keySet());
tableHandle = applyFilter(metadata, tableHandle, constraint);
tableHandle = metadata.applyDelete(session, tableHandle).get();
metadata.executeDelete(session, tableHandle);
rollbackIfEquals(tag, ROLLBACK_AFTER_DELETE);
// Query 2: insert
session = newSession();
ConnectorInsertTableHandle insertTableHandle = metadata.beginInsert(session, tableHandle, ImmutableList.of(), NO_RETRIES);
rollbackIfEquals(tag, ROLLBACK_AFTER_BEGIN_INSERT);
writePath = getStagingPathRoot(insertTableHandle);
targetPath = getTargetPathRoot(insertTableHandle);
ConnectorPageSink sink = pageSinkProvider.createPageSink(transaction.getTransactionHandle(), session, insertTableHandle);
sink.appendPage(insertData.toPage());
rollbackIfEquals(tag, ROLLBACK_AFTER_APPEND_PAGE);
Collection<Slice> fragments = getFutureValue(sink.finish());
rollbackIfEquals(tag, ROLLBACK_AFTER_SINK_FINISH);
metadata.finishInsert(session, insertTableHandle, fragments, ImmutableList.of());
rollbackIfEquals(tag, ROLLBACK_AFTER_FINISH_INSERT);
assertEquals(tag, COMMIT);
if (conflictTrigger.isPresent()) {
JsonCodec<PartitionUpdate> partitionUpdateCodec = JsonCodec.jsonCodec(PartitionUpdate.class);
List<PartitionUpdate> partitionUpdates = fragments.stream().map(Slice::getBytes).map(partitionUpdateCodec::fromJson).collect(toList());
conflictTrigger.get().triggerConflict(session, tableName, insertTableHandle, partitionUpdates);
}
transaction.commit();
if (conflictTrigger.isPresent()) {
assertTrue(expectQuerySucceed);
conflictTrigger.get().verifyAndCleanup(session, tableName);
}
} catch (TestingRollbackException e) {
transaction.rollback();
} catch (TrinoException e) {
assertFalse(expectQuerySucceed);
if (conflictTrigger.isPresent()) {
conflictTrigger.get().verifyAndCleanup(newSession(), tableName);
}
}
}
// check that temporary files are removed
if (writePath != null && !writePath.equals(targetPath)) {
HdfsContext context = new HdfsContext(newSession());
FileSystem fileSystem = hdfsEnvironment.getFileSystem(context, writePath);
assertFalse(fileSystem.exists(writePath));
}
try (Transaction transaction = newTransaction()) {
// verify partitions
List<String> partitionNames = transaction.getMetastore().getPartitionNames(tableName.getSchemaName(), tableName.getTableName()).orElseThrow(() -> new AssertionError("Table does not exist: " + tableName));
assertEqualsIgnoreOrder(partitionNames, expectedData.getMaterializedRows().stream().map(row -> format("pk1=%s/pk2=%s", row.getField(1), row.getField(2))).distinct().collect(toImmutableList()));
// load the new table
ConnectorSession session = newSession();
ConnectorMetadata metadata = transaction.getMetadata();
metadata.beginQuery(session);
ConnectorTableHandle tableHandle = getTableHandle(metadata, tableName);
List<ColumnHandle> columnHandles = filterNonHiddenColumnHandles(metadata.getColumnHandles(session, tableHandle).values());
// verify the data
MaterializedResult result = readTable(transaction, tableHandle, columnHandles, session, TupleDomain.all(), OptionalInt.empty(), Optional.of(storageFormat));
assertEqualsIgnoreOrder(result.getMaterializedRows(), expectedData.getMaterializedRows());
}
}
use of io.trino.spi.connector.RetryMode.NO_RETRIES in project trino by trinodb.
the class AbstractTestHive method doTestMismatchSchemaTable.
protected void doTestMismatchSchemaTable(SchemaTableName schemaTableName, HiveStorageFormat storageFormat, List<ColumnMetadata> tableBefore, MaterializedResult dataBefore, List<ColumnMetadata> tableAfter, MaterializedResult dataAfter) throws Exception {
String schemaName = schemaTableName.getSchemaName();
String tableName = schemaTableName.getTableName();
doCreateEmptyTable(schemaTableName, storageFormat, tableBefore);
// insert the data
try (Transaction transaction = newTransaction()) {
ConnectorSession session = newSession();
ConnectorMetadata metadata = transaction.getMetadata();
ConnectorTableHandle tableHandle = getTableHandle(metadata, schemaTableName);
ConnectorInsertTableHandle insertTableHandle = metadata.beginInsert(session, tableHandle, ImmutableList.of(), NO_RETRIES);
ConnectorPageSink sink = pageSinkProvider.createPageSink(transaction.getTransactionHandle(), session, insertTableHandle);
sink.appendPage(dataBefore.toPage());
Collection<Slice> fragments = getFutureValue(sink.finish());
metadata.finishInsert(session, insertTableHandle, fragments, ImmutableList.of());
transaction.commit();
}
// load the table and verify the data
try (Transaction transaction = newTransaction()) {
ConnectorSession session = newSession();
ConnectorMetadata metadata = transaction.getMetadata();
metadata.beginQuery(session);
ConnectorTableHandle tableHandle = getTableHandle(metadata, schemaTableName);
List<ColumnHandle> columnHandles = metadata.getColumnHandles(session, tableHandle).values().stream().filter(columnHandle -> !((HiveColumnHandle) columnHandle).isHidden()).collect(toList());
MaterializedResult result = readTable(transaction, tableHandle, columnHandles, session, TupleDomain.all(), OptionalInt.empty(), Optional.empty());
assertEqualsIgnoreOrder(result.getMaterializedRows(), dataBefore.getMaterializedRows());
transaction.commit();
}
// alter the table schema
try (Transaction transaction = newTransaction()) {
ConnectorSession session = newSession();
PrincipalPrivileges principalPrivileges = testingPrincipalPrivilege(session);
Table oldTable = transaction.getMetastore().getTable(schemaName, tableName).get();
List<Column> dataColumns = tableAfter.stream().filter(columnMetadata -> !columnMetadata.getName().equals("ds")).map(columnMetadata -> new Column(columnMetadata.getName(), toHiveType(columnMetadata.getType()), Optional.empty())).collect(toList());
Table.Builder newTable = Table.builder(oldTable).setDataColumns(dataColumns);
transaction.getMetastore().replaceTable(schemaName, tableName, newTable.build(), principalPrivileges);
transaction.commit();
}
// load the altered table and verify the data
try (Transaction transaction = newTransaction()) {
ConnectorSession session = newSession();
ConnectorMetadata metadata = transaction.getMetadata();
metadata.beginQuery(session);
ConnectorTableHandle tableHandle = getTableHandle(metadata, schemaTableName);
List<ColumnHandle> columnHandles = metadata.getColumnHandles(session, tableHandle).values().stream().filter(columnHandle -> !((HiveColumnHandle) columnHandle).isHidden()).collect(toList());
MaterializedResult result = readTable(transaction, tableHandle, columnHandles, session, TupleDomain.all(), OptionalInt.empty(), Optional.empty());
assertEqualsIgnoreOrder(result.getMaterializedRows(), dataAfter.getMaterializedRows());
transaction.commit();
}
// insertions to the partitions with type mismatches should fail
try (Transaction transaction = newTransaction()) {
ConnectorSession session = newSession();
ConnectorMetadata metadata = transaction.getMetadata();
ConnectorTableHandle tableHandle = getTableHandle(metadata, schemaTableName);
ConnectorInsertTableHandle insertTableHandle = metadata.beginInsert(session, tableHandle, ImmutableList.of(), NO_RETRIES);
ConnectorPageSink sink = pageSinkProvider.createPageSink(transaction.getTransactionHandle(), session, insertTableHandle);
sink.appendPage(dataAfter.toPage());
Collection<Slice> fragments = getFutureValue(sink.finish());
metadata.finishInsert(session, insertTableHandle, fragments, ImmutableList.of());
transaction.commit();
fail("expected exception");
} catch (TrinoException e) {
// expected
assertEquals(e.getErrorCode(), HIVE_PARTITION_SCHEMA_MISMATCH.toErrorCode());
}
}
use of io.trino.spi.connector.RetryMode.NO_RETRIES in project trino by trinodb.
the class IcebergMetadata method finishRefreshMaterializedView.
@Override
public Optional<ConnectorOutputMetadata> finishRefreshMaterializedView(ConnectorSession session, ConnectorTableHandle tableHandle, ConnectorInsertTableHandle insertHandle, Collection<Slice> fragments, Collection<ComputedStatistics> computedStatistics, List<ConnectorTableHandle> sourceTableHandles) {
// delete before insert .. simulating overwrite
executeDelete(session, tableHandle);
IcebergWritableTableHandle table = (IcebergWritableTableHandle) insertHandle;
Table icebergTable = transaction.table();
List<CommitTaskData> commitTasks = fragments.stream().map(slice -> commitTaskCodec.fromJson(slice.getBytes())).collect(toImmutableList());
Type[] partitionColumnTypes = icebergTable.spec().fields().stream().map(field -> field.transform().getResultType(icebergTable.schema().findType(field.sourceId()))).toArray(Type[]::new);
AppendFiles appendFiles = transaction.newFastAppend();
ImmutableSet.Builder<String> writtenFiles = ImmutableSet.builder();
for (CommitTaskData task : commitTasks) {
DataFiles.Builder builder = DataFiles.builder(icebergTable.spec()).withPath(task.getPath()).withFileSizeInBytes(task.getFileSizeInBytes()).withFormat(table.getFileFormat().toIceberg()).withMetrics(task.getMetrics().metrics());
if (!icebergTable.spec().fields().isEmpty()) {
String partitionDataJson = task.getPartitionDataJson().orElseThrow(() -> new VerifyException("No partition data for partitioned table"));
builder.withPartition(PartitionData.fromJson(partitionDataJson, partitionColumnTypes));
}
appendFiles.appendFile(builder.build());
writtenFiles.add(task.getPath());
}
String dependencies = sourceTableHandles.stream().map(handle -> (IcebergTableHandle) handle).filter(handle -> handle.getSnapshotId().isPresent()).map(handle -> handle.getSchemaTableName() + "=" + handle.getSnapshotId().get()).distinct().collect(joining(","));
// try to leave as little garbage as possible behind
if (table.getRetryMode() != NO_RETRIES) {
cleanExtraOutputFiles(session, writtenFiles.build());
}
// Update the 'dependsOnTables' property that tracks tables on which the materialized view depends and the corresponding snapshot ids of the tables
appendFiles.set(DEPENDS_ON_TABLES, dependencies);
appendFiles.commit();
transaction.commitTransaction();
transaction = null;
return Optional.of(new HiveWrittenPartitions(commitTasks.stream().map(CommitTaskData::getPath).collect(toImmutableList())));
}
Aggregations