ClickHouse/dbms/src/Storages/IStorage.h

363 lines
16 KiB
C++
Raw Normal View History

2011-08-15 01:12:57 +00:00
#pragma once
2010-03-04 19:20:28 +00:00
#include <Common/Exception.h>
2018-11-27 16:45:45 +00:00
#include <Common/RWLock.h>
#include <Core/Names.h>
#include <Core/QueryProcessingStage.h>
#include <Databases/IDatabase.h>
#include <Storages/ITableDeclaration.h>
2019-03-05 10:12:20 +00:00
#include <Storages/TableStructureLockHolder.h>
#include <Storages/SelectQueryInfo.h>
#include <Interpreters/CancellationCode.h>
#include <shared_mutex>
2015-02-10 21:10:58 +00:00
#include <memory>
#include <optional>
#include <Common/ActionLock.h>
2010-03-01 16:59:51 +00:00
namespace DB
{
namespace ErrorCodes
{
extern const int TABLE_IS_DROPPED;
extern const int NOT_IMPLEMENTED;
}
class Context;
class IBlockInputStream;
class IBlockOutputStream;
using StorageActionBlockType = size_t;
using BlockOutputStreamPtr = std::shared_ptr<IBlockOutputStream>;
using BlockInputStreamPtr = std::shared_ptr<IBlockInputStream>;
using BlockInputStreams = std::vector<BlockInputStreamPtr>;
class ASTCreateQuery;
2010-03-01 16:59:51 +00:00
class IStorage;
using StoragePtr = std::shared_ptr<IStorage>;
using StorageWeakPtr = std::weak_ptr<IStorage>;
struct Settings;
class AlterCommands;
2018-06-13 20:02:27 +00:00
class MutationCommands;
class PartitionCommands;
2017-04-16 15:00:33 +00:00
/** Storage. Responsible for
* - storage of the table data;
* - the definition in which files (or not in files) the data is stored;
* - data lookups and appends;
2017-04-16 15:00:33 +00:00
* - data storage structure (compression, etc.)
* - concurrent access to data (locks, etc.)
2010-03-01 16:59:51 +00:00
*/
class IStorage : public std::enable_shared_from_this<IStorage>, private boost::noncopyable, public ITableDeclaration
2010-03-01 16:59:51 +00:00
{
public:
2017-04-16 15:00:33 +00:00
/// The main name of the table type (for example, StorageMergeTree).
virtual std::string getName() const = 0;
/** The name of the table.
*/
virtual std::string getTableName() const = 0;
/** Returns true if the storage receives data from a remote server or servers. */
virtual bool isRemote() const { return false; }
2017-04-16 15:00:33 +00:00
/** Returns true if the storage supports queries with the SAMPLE section. */
virtual bool supportsSampling() const { return false; }
2017-04-16 15:00:33 +00:00
/** Returns true if the storage supports queries with the FINAL section. */
virtual bool supportsFinal() const { return false; }
2017-04-16 15:00:33 +00:00
/** Returns true if the storage supports queries with the PREWHERE section. */
virtual bool supportsPrewhere() const { return false; }
2017-04-25 15:21:03 +00:00
/** Returns true if the storage replicates SELECT, INSERT and ALTER commands among replicas. */
virtual bool supportsReplication() const { return false; }
/** Returns true if the storage supports deduplication of inserted data blocks . */
virtual bool supportsDeduplication() const { return false; }
2017-04-25 15:21:03 +00:00
2019-03-05 10:12:20 +00:00
2019-03-07 18:34:46 +00:00
/// Acquire this lock if you need the table structure to remain constant during the execution of
/// the query. If will_add_new_data is true, this means that the query will add new data to the table
/// (INSERT or a parts merge).
TableStructureReadLockHolder lockStructureForShare(bool will_add_new_data, const String & query_id)
{
TableStructureReadLockHolder result;
2019-03-05 10:12:20 +00:00
if (will_add_new_data)
result.new_data_structure_lock = new_data_structure_lock->getLock(RWLockImpl::Read, query_id);
result.structure_lock = structure_lock->getLock(RWLockImpl::Read, query_id);
if (is_dropped)
throw Exception("Table is dropped", ErrorCodes::TABLE_IS_DROPPED);
2019-03-05 10:12:20 +00:00
return result;
}
2019-03-07 18:34:46 +00:00
/// Acquire this lock at the start of ALTER to lock out other ALTERs and make sure that only you
/// can modify the table structure. It can later be upgraded to the exclusive lock.
TableStructureWriteLockHolder lockAlterIntention(const String & query_id)
{
TableStructureWriteLockHolder result;
2019-03-05 10:12:20 +00:00
result.alter_intention_lock = alter_intention_lock->getLock(RWLockImpl::Write, query_id);
2019-03-05 10:12:20 +00:00
if (is_dropped)
throw Exception("Table is dropped", ErrorCodes::TABLE_IS_DROPPED);
return result;
}
2019-03-07 18:34:46 +00:00
/// Upgrade alter intention lock and make sure that no new data is inserted into the table.
/// This is used by the ALTER MODIFY of the MergeTree storage to consistently determine
/// the set of parts that needs to be altered.
void lockNewDataStructureExclusively(TableStructureWriteLockHolder & lock_holder, const String & query_id)
{
2019-03-05 10:12:20 +00:00
if (!lock_holder.alter_intention_lock)
throw Exception("Alter intention lock for table " + getTableName() + " was not taken. This is a bug.",
ErrorCodes::LOGICAL_ERROR);
lock_holder.new_data_structure_lock = new_data_structure_lock->getLock(RWLockImpl::Write, query_id);
}
2019-03-07 18:34:46 +00:00
/// Upgrade alter intention lock to the full exclusive structure lock. This is done by ALTER queries
/// to ensure that no other query uses the table structure and it can be safely changed.
void lockStructureExclusively(TableStructureWriteLockHolder & lock_holder, const String & query_id)
{
2019-03-05 10:12:20 +00:00
if (!lock_holder.alter_intention_lock)
throw Exception("Alter intention lock for table " + getTableName() + " was not taken. This is a bug.",
ErrorCodes::LOGICAL_ERROR);
if (!lock_holder.new_data_structure_lock)
lock_holder.new_data_structure_lock = new_data_structure_lock->getLock(RWLockImpl::Write, query_id);
lock_holder.structure_lock = structure_lock->getLock(RWLockImpl::Write, query_id);
}
2019-03-07 18:34:46 +00:00
/// Acquire the full exclusive lock immediately. No other queries can run concurrently.
TableStructureWriteLockHolder lockExclusively(const String & query_id)
2019-03-05 10:12:20 +00:00
{
TableStructureWriteLockHolder result;
2019-03-05 10:12:20 +00:00
result.alter_intention_lock = alter_intention_lock->getLock(RWLockImpl::Write, query_id);
if (is_dropped)
throw Exception("Table is dropped", ErrorCodes::TABLE_IS_DROPPED);
result.new_data_structure_lock = new_data_structure_lock->getLock(RWLockImpl::Write, query_id);
result.structure_lock = structure_lock->getLock(RWLockImpl::Write, query_id);
2019-03-05 10:12:20 +00:00
return result;
}
/** Returns stage to which query is going to be processed in read() function.
* (Normally, the function only reads the columns from the list, but in other cases,
* for example, the request can be partially processed on a remote server.)
*/
virtual QueryProcessingStage::Enum getQueryProcessingStage(const Context &) const { return QueryProcessingStage::FetchColumns; }
2017-04-16 15:00:33 +00:00
/** Read a set of columns from the table.
* Accepts a list of columns to read, as well as a description of the query,
* from which information can be extracted about how to retrieve data
* (indexes, locks, etc.)
* Returns a stream with which you can read data sequentially
* or multiple streams for parallel data reading.
* The `processed_stage` must be the result of getQueryProcessingStage() function.
*
* context contains settings for one query.
2017-04-16 15:00:33 +00:00
* Usually Storage does not care about these settings, since they are used in the interpreter.
* But, for example, for distributed query processing, the settings are passed to the remote server.
*
2017-06-02 15:54:39 +00:00
* num_streams - a recommendation, how many streams to return,
* if the storage can return a different number of streams.
*
2017-04-16 15:00:33 +00:00
* It is guaranteed that the structure of the table will not change over the lifetime of the returned streams (that is, there will not be ALTER, RENAME and DROP).
*/
virtual BlockInputStreams read(
2017-12-01 19:34:51 +00:00
const Names & /*column_names*/,
const SelectQueryInfo & /*query_info*/,
const Context & /*context*/,
QueryProcessingStage::Enum /*processed_stage*/,
size_t /*max_block_size*/,
2017-12-01 19:34:51 +00:00
unsigned /*num_streams*/)
{
throw Exception("Method read is not supported by storage " + getName(), ErrorCodes::NOT_IMPLEMENTED);
}
2017-04-16 15:00:33 +00:00
/** Writes the data to a table.
* Receives a description of the query, which can contain information about the data write method.
* Returns an object by which you can write data sequentially.
*
2017-04-16 15:00:33 +00:00
* It is guaranteed that the table structure will not change over the lifetime of the returned streams (that is, there will not be ALTER, RENAME and DROP).
*/
virtual BlockOutputStreamPtr write(
2017-12-01 19:34:51 +00:00
const ASTPtr & /*query*/,
const Context & /*context*/)
{
throw Exception("Method write is not supported by storage " + getName(), ErrorCodes::NOT_IMPLEMENTED);
}
2017-04-16 15:00:33 +00:00
/** Delete the table data. Called before deleting the directory with the data.
2018-06-09 15:48:22 +00:00
* The method can be called only after detaching table from Context (when no queries are performed with table).
* The table is not usable during and after call to this method.
2017-04-16 15:00:33 +00:00
* If you do not need any action other than deleting the directory with data, you can leave this method blank.
*/
virtual void drop() {}
2018-06-09 15:48:22 +00:00
/** Clear the table data and leave it empty.
* Must be called under lockForAlter.
2018-04-21 00:35:20 +00:00
*/
virtual void truncate(const ASTPtr & /*query*/, const Context & /* context */)
2018-04-21 00:35:20 +00:00
{
throw Exception("Truncate is not supported by storage " + getName(), ErrorCodes::NOT_IMPLEMENTED);
}
2017-04-16 15:00:33 +00:00
/** Rename the table.
* Renaming a name in a file with metadata, the name in the list of tables in the RAM, is done separately.
* In this function, you need to rename the directory with the data, if any.
* Called when the table structure is locked for write.
*/
2017-12-01 19:34:51 +00:00
virtual void rename(const String & /*new_path_to_db*/, const String & /*new_database_name*/, const String & /*new_table_name*/)
{
throw Exception("Method rename is not supported by storage " + getName(), ErrorCodes::NOT_IMPLEMENTED);
}
2017-04-16 15:00:33 +00:00
/** ALTER tables in the form of column changes that do not affect the change to Storage or its parameters.
* This method must fully execute the ALTER query, taking care of the locks itself.
* To update the table metadata on disk, this method should call InterpreterAlterQuery::updateMetadata.
*/
virtual void alter(const AlterCommands & params, const String & database_name, const String & table_name, const Context & context, TableStructureWriteLockHolder & structure_lock);
/** ALTER tables with regard to its partitions.
* Should handle locks for each command on its own.
*/
virtual void alterPartition(const ASTPtr & /* query */, const PartitionCommands & /* commands */, const Context & /* context */)
{
throw Exception("Partition operations are not supported by storage " + getName(), ErrorCodes::NOT_IMPLEMENTED);
}
2017-04-16 15:00:33 +00:00
/** Perform any background work. For example, combining parts in a MergeTree type table.
* Returns whether any work has been done.
*/
2017-12-01 19:34:51 +00:00
virtual bool optimize(const ASTPtr & /*query*/, const ASTPtr & /*partition*/, bool /*final*/, bool /*deduplicate*/, const Context & /*context*/)
{
throw Exception("Method optimize is not supported by storage " + getName(), ErrorCodes::NOT_IMPLEMENTED);
}
/// Mutate the table contents
virtual void mutate(const MutationCommands &, const Context &)
{
throw Exception("Mutations are not supported by storage " + getName(), ErrorCodes::NOT_IMPLEMENTED);
}
/// Cancel a mutation.
virtual CancellationCode killMutation(const String & /*mutation_id*/)
{
throw Exception("Mutations are not supported by storage " + getName(), ErrorCodes::NOT_IMPLEMENTED);
}
/** If the table have to do some complicated work on startup,
* that must be postponed after creation of table object
* (like launching some background threads),
* do it in this method.
* You should call this method after creation of object.
* By default, does nothing.
* Cannot be called simultaneously by multiple threads.
*/
virtual void startup() {}
/** If the table have to do some complicated work when destroying an object - do it in advance.
2017-04-16 15:00:33 +00:00
* For example, if the table contains any threads for background work - ask them to complete and wait for completion.
* By default, does nothing.
2017-04-16 15:00:33 +00:00
* Can be called simultaneously from different threads, even after a call to drop().
*/
virtual void shutdown() {}
/// Asks table to stop executing some action identified by action_type
/// If table does not support such type of lock, and empty lock is returned
virtual ActionLock getActionLock(StorageActionBlockType /* action_type */)
{
return {};
}
bool is_dropped{false};
/// Does table support index for IN sections
virtual bool supportsIndexForIn() const { return false; }
2018-01-23 08:18:12 +00:00
/// Provides a hint that the storage engine may evaluate the IN-condition by using an index.
virtual bool mayBenefitFromIndexForIn(const ASTPtr & /* left_in_operand */, const Context & /* query_context */) const { return false; }
/// Checks validity of the data
virtual bool checkData() const { throw Exception("Check query is not supported for " + getName() + " storage", ErrorCodes::NOT_IMPLEMENTED); }
/// Checks that table could be dropped right now
2018-08-03 09:54:46 +00:00
/// Otherwise - throws an exception with detailed information.
2018-08-03 13:19:53 +00:00
/// We do not use mutex because it is not very important that the size could change during the operation.
virtual void checkTableCanBeDropped() const {}
/// Checks that Partition could be dropped right now
2018-08-03 09:54:46 +00:00
/// Otherwise - throws an exception with detailed information.
2018-08-03 13:19:53 +00:00
/// We do not use mutex because it is not very important that the size could change during the operation.
virtual void checkPartitionCanBeDropped(const ASTPtr & /*partition*/) {}
/** Notify engine about updated dependencies for this storage. */
virtual void updateDependencies() {}
/// Returns data path if storage supports it, empty string otherwise.
virtual String getDataPath() const { return {}; }
2018-11-27 17:07:10 +00:00
/// Returns ASTExpressionList of partition key expression for storage or nullptr if there is none.
virtual ASTPtr getPartitionKeyAST() const { return nullptr; }
2018-11-27 17:07:10 +00:00
/// Returns ASTExpressionList of sorting key expression for storage or nullptr if there is none.
virtual ASTPtr getSortingKeyAST() const { return nullptr; }
2018-10-04 08:58:19 +00:00
2018-11-27 17:07:10 +00:00
/// Returns ASTExpressionList of primary key expression for storage or nullptr if there is none.
virtual ASTPtr getPrimaryKeyAST() const { return nullptr; }
2018-11-09 09:39:07 +00:00
2018-11-27 17:07:10 +00:00
/// Returns sampling expression AST for storage or nullptr if there is none.
virtual ASTPtr getSamplingKeyAST() const { return nullptr; }
2018-11-27 17:07:10 +00:00
/// Returns additional columns that need to be read to calculate partition key.
virtual Names getColumnsRequiredForPartitionKey() const { return {}; }
2018-11-27 17:07:10 +00:00
/// Returns additional columns that need to be read to calculate sorting key.
virtual Names getColumnsRequiredForSortingKey() const { return {}; }
2018-11-27 17:07:10 +00:00
/// Returns additional columns that need to be read to calculate primary key.
virtual Names getColumnsRequiredForPrimaryKey() const { return {}; }
2018-11-27 17:07:10 +00:00
/// Returns additional columns that need to be read to calculate sampling key.
virtual Names getColumnsRequiredForSampling() const { return {}; }
/// Returns additional columns that need to be read for FINAL to work.
virtual Names getColumnsRequiredForFinal() const { return {}; }
using ITableDeclaration::ITableDeclaration;
using std::enable_shared_from_this<IStorage>::shared_from_this;
2014-03-09 17:36:01 +00:00
private:
2019-03-05 10:12:20 +00:00
/// You always need to take the next three locks in this order.
2019-03-07 18:34:46 +00:00
/// If you hold this lock exclusively, you can be sure that no other structure modifying queries
/// (e.g. ALTER, DROP) are concurrently executing. But queries that only read table structure
/// (e.g. SELECT, INSERT) can continue to execute.
2019-03-05 10:12:20 +00:00
mutable RWLock alter_intention_lock = RWLockImpl::create();
2019-03-07 18:34:46 +00:00
/// It is taken for share for the entire INSERT query and the entire merge of the parts (for MergeTree).
/// ALTER queries acquire an exclusive lock to ensure
2019-03-05 10:12:20 +00:00
mutable RWLock new_data_structure_lock = RWLockImpl::create();
2019-03-07 18:34:46 +00:00
/// Lock for the table column structure (names, types, etc.) and data path.
/// It is taken in exclusive mode by queries that modify them (e.g. RENAME, ALTER and DROP)
/// and in share mode by other queries.
2018-11-27 16:45:45 +00:00
mutable RWLock structure_lock = RWLockImpl::create();
};
2010-03-01 16:59:51 +00:00
2017-04-16 15:00:33 +00:00
/// table name -> table
using Tables = std::map<String, StoragePtr>;
2011-08-15 01:12:57 +00:00
}