ClickHouse/dbms/src/Storages/MergeTree/MergeTreeData.h
2017-11-18 05:34:00 +03:00

573 lines
23 KiB
C++

#pragma once
#include <Core/SortDescription.h>
#include <Common/SimpleIncrement.h>
#include <Interpreters/Context.h>
#include <Interpreters/ExpressionActions.h>
#include <Storages/IStorage.h>
#include <Storages/MergeTree/MergeTreePartInfo.h>
#include <Storages/MergeTree/MergeTreeSettings.h>
#include <IO/ReadBufferFromString.h>
#include <IO/WriteBufferFromFile.h>
#include <IO/ReadBufferFromFile.h>
#include <DataTypes/DataTypeString.h>
#include <DataTypes/DataTypesNumber.h>
#include <DataStreams/GraphiteRollupSortedBlockInputStream.h>
#include <Storages/MergeTree/MergeTreeDataPart.h>
#include <common/RangeFiltered.h>
namespace DB
{
namespace ErrorCodes
{
extern const int LOGICAL_ERROR;
extern const int INVALID_PARTITION_NAME;
extern const int TOO_MUCH_PARTS;
extern const int NO_SUCH_DATA_PART;
extern const int DUPLICATE_DATA_PART;
extern const int DIRECTORY_ALREADY_EXISTS;
extern const int TOO_MANY_UNEXPECTED_DATA_PARTS;
extern const int NO_SUCH_COLUMN_IN_TABLE;
extern const int TABLE_DIFFERS_TOO_MUCH;
}
/// Data structure for *MergeTree engines.
/// Merge tree is used for incremental sorting of data.
/// The table consists of several sorted parts.
/// During insertion new data is sorted according to the primary key and is written to the new part.
/// Parts are merged in the background according to a heuristic algorithm.
/// For each part the index file is created containing primary key values for every n-th row.
/// This allows efficient selection by primary key range predicate.
///
/// Additionally:
///
/// The date column is specified. For each part min and max dates are remembered.
/// Essentially it is an index too.
///
/// Data is partitioned by the value of the partitioning expression.
/// Parts belonging to different partitions are not merged - for the ease of administration (data sync and backup).
///
/// File structure of old-style month-partitioned tables (format_version = 0):
/// Part directory - / min-date _ max-date _ min-id _ max-id _ level /
/// Inside the part directory:
/// checksums.txt - contains the list of all files along with their sizes and checksums.
/// columns.txt - contains the list of all columns and their types.
/// primary.idx - contains the primary index.
/// [Column].bin - contains compressed column data.
/// [Column].mrk - marks, pointing to seek positions allowing to skip n * k rows.
///
/// File structure of tables with custom partitioning (format_version >= 1):
/// Part directory - / partiiton-id _ min-id _ max-id _ level /
/// Inside the part directory:
/// The same files as for month-partitioned tables, plus
/// count.txt - contains total number of rows in this part.
/// partition.dat - contains the value of the partitioning expression.
/// minmax_[Column].idx - MinMax indexes (see MergeTreeDataPart::MinMaxIndex class) for the columns required by the partitioning expression.
///
/// Several modes are implemented. Modes determine additional actions during merge:
/// - Ordinary - don't do anything special
/// - Collapsing - collapse pairs of rows with the opposite values of sign_columns for the same values
/// of primary key (cf. CollapsingSortedBlockInputStream.h)
/// - Replacing - for all rows with the same primary key keep only the latest one. Or, if the version
/// column is set, keep the latest row with the maximal version.
/// - Summing - sum all numeric columns not contained in the primary key for all rows with the same primary key.
/// - Aggregating - merge columns containing aggregate function states for all rows with the same primary key.
/// - Unsorted - during the merge the data is not sorted but merely concatenated; this allows reading the data
/// in the same batches as they were written.
/// - Graphite - performs coarsening of historical data for Graphite (a system for quantitative monitoring).
/// The MergeTreeData class contains a list of parts and the data structure parameters.
/// To read and modify the data use other classes:
/// - MergeTreeDataSelectExecutor
/// - MergeTreeDataWriter
/// - MergeTreeDataMerger
class MergeTreeData : public ITableDeclaration
{
public:
/// Function to call if the part is suspected to contain corrupt data.
using BrokenPartCallback = std::function<void (const String &)>;
/// Callback to delete outdated parts immediately
using PartsCleanCallback = std::function<void ()>;
using DataPart = MergeTreeDataPart;
using MutableDataPartPtr = std::shared_ptr<DataPart>;
/// After the DataPart is added to the working set, it cannot be changed.
using DataPartPtr = std::shared_ptr<const DataPart>;
using DataPartState = MergeTreeDataPart::State;
using DataPartStates = std::initializer_list<DataPartState>;
using DataPartStateVector = std::vector<DataPartState>;
struct DataPartPtrLess
{
using is_transparent = void;
bool operator()(const DataPartPtr & lhs, const MergeTreePartInfo & rhs) const { return lhs->info < rhs; }
bool operator()(const MergeTreePartInfo & lhs, const DataPartPtr & rhs) const { return lhs < rhs->info; }
bool operator()(const DataPartPtr & lhs, const DataPartPtr & rhs) const { return lhs->info < rhs->info; }
};
using DataParts = std::set<DataPartPtr, DataPartPtrLess>;
using DataPartsVector = std::vector<DataPartPtr>;
/// Some operations on the set of parts return a Transaction object.
/// If neither commit() nor rollback() was called, the destructor rollbacks the operation.
class Transaction : private boost::noncopyable
{
public:
Transaction() {}
void commit();
void rollback();
~Transaction()
{
try
{
rollback();
}
catch(...)
{
tryLogCurrentException("~MergeTreeData::Transaction");
}
}
private:
friend class MergeTreeData;
MergeTreeData * data = nullptr;
/// What to do on rollback.
DataPartsVector parts_to_remove_on_rollback;
DataPartsVector parts_to_add_on_rollback;
void clear()
{
data = nullptr;
parts_to_remove_on_rollback.clear();
parts_to_add_on_rollback.clear();
}
void replaceParts(DataPartState move_precommitted_to, DataPartState move_committed_to, bool remove_without_delay);
};
/// An object that stores the names of temporary files created in the part directory during ALTER of its
/// columns.
class AlterDataPartTransaction : private boost::noncopyable
{
public:
/// Renames temporary files, finishing the ALTER of the part.
void commit();
/// If commit() was not called, deletes temporary files, canceling the ALTER.
~AlterDataPartTransaction();
/// Review the changes before the commit.
const NamesAndTypesList & getNewColumns() const { return new_columns; }
const DataPart::Checksums & getNewChecksums() const { return new_checksums; }
private:
friend class MergeTreeData;
AlterDataPartTransaction(DataPartPtr data_part_) : data_part(data_part_), alter_lock(data_part->alter_mutex) {}
void clear()
{
alter_lock.unlock();
data_part = nullptr;
}
DataPartPtr data_part;
std::unique_lock<std::mutex> alter_lock;
DataPart::Checksums new_checksums;
NamesAndTypesList new_columns;
/// If the value is an empty string, the file is not temporary, and it must be deleted.
NameToNameMap rename_map;
};
using AlterDataPartTransactionPtr = std::unique_ptr<AlterDataPartTransaction>;
/// Parameters for various modes.
struct MergingParams
{
/// Merging mode. See above.
enum Mode
{
Ordinary = 0, /// Enum values are saved. Do not change them.
Collapsing = 1,
Summing = 2,
Aggregating = 3,
Unsorted = 4,
Replacing = 5,
Graphite = 6,
};
Mode mode;
/// For collapsing mode.
String sign_column;
/// For Summing mode. If empty - columns_to_sum is determined automatically.
Names columns_to_sum;
/// For Replacing mode. Can be empty.
String version_column;
/// For Graphite mode.
Graphite::Params graphite_params;
/// Check that needed columns are present and have correct types.
void check(const NamesAndTypesList & columns) const;
String getModeName() const;
};
/// Attach the table corresponding to the directory in full_path (must end with /), with the given columns.
/// Correctness of names and paths is not checked.
///
/// primary_expr_ast - expression used for sorting; empty for UnsortedMergeTree.
/// date_column_name - if not empty, the name of the Date column used for partitioning by month.
/// Otherwise, partition_expr_ast is used for partitioning.
/// require_part_metadata - should checksums.txt and columns.txt exist in the part directory.
/// attach - whether the existing table is attached or the new table is created.
MergeTreeData( const String & database_, const String & table_,
const String & full_path_, NamesAndTypesListPtr columns_,
const NamesAndTypesList & materialized_columns_,
const NamesAndTypesList & alias_columns_,
const ColumnDefaults & column_defaults_,
Context & context_,
const ASTPtr & primary_expr_ast_,
const String & date_column_name,
const ASTPtr & partition_expr_ast_,
const ASTPtr & sampling_expression_, /// nullptr, if sampling is not supported.
const MergingParams & merging_params_,
const MergeTreeSettings & settings_,
const String & log_name_,
bool require_part_metadata_,
bool attach,
BrokenPartCallback broken_part_callback_ = [](const String &){},
PartsCleanCallback parts_clean_callback_ = nullptr
);
/// Load the set of data parts from disk. Call once - immediately after the object is created.
void loadDataParts(bool skip_sanity_checks);
bool supportsSampling() const { return !!sampling_expression; }
bool supportsPrewhere() const { return true; }
bool supportsFinal() const
{
return merging_params.mode == MergingParams::Collapsing
|| merging_params.mode == MergingParams::Summing
|| merging_params.mode == MergingParams::Aggregating
|| merging_params.mode == MergingParams::Replacing;
}
Int64 getMaxDataPartIndex();
const NamesAndTypesList & getColumnsListImpl() const override { return *columns; }
NameAndTypePair getColumn(const String & column_name) const override
{
if (column_name == "_part")
return NameAndTypePair("_part", std::make_shared<DataTypeString>());
if (column_name == "_part_index")
return NameAndTypePair("_part_index", std::make_shared<DataTypeUInt64>());
if (column_name == "_sample_factor")
return NameAndTypePair("_sample_factor", std::make_shared<DataTypeFloat64>());
return ITableDeclaration::getColumn(column_name);
}
bool hasColumn(const String & column_name) const override
{
return ITableDeclaration::hasColumn(column_name)
|| column_name == "_part"
|| column_name == "_part_index"
|| column_name == "_sample_factor";
}
String getDatabaseName() const { return database_name; }
String getTableName() const override { return table_name; }
String getFullPath() const { return full_path; }
String getLogName() const { return log_name; }
/// Returns a copy of the list so that the caller shouldn't worry about locks.
DataParts getDataParts(const DataPartStates & affordable_states) const;
DataPartsVector getDataPartsVector(const DataPartStates & affordable_states) const;
DataPartsVector getDataPartsVector(const DataPartStates & affordable_states, DataPartStateVector & out_states_snapshot) const;
/// Returns a virtual container iteration only through parts with specified states
decltype(auto) getDataPartsRange(const DataPartStates & affordable_states) const
{
return createRangeFiltered(DataPart::getStatesFilter(affordable_states), data_parts);
}
/// Returns Committed parts
DataParts getDataParts() const;
DataPartsVector getDataPartsVector() const;
/// Returns all parts except Temporary and Deleting ones
DataParts getAllDataParts() const;
/// Returns an comitted part with the given name or a part containing it. If there is no such part, returns nullptr.
DataPartPtr getActiveContainingPart(const String & part_name);
/// Returns the part with the given name (and state) or nullptr if no such part.
DataPartPtr getPartIfExists(const String & part_name, const DataPartStates & valid_states = {DataPartState::Committed});
/// Total size of active parts in bytes.
size_t getTotalActiveSizeInBytes() const;
size_t getMaxPartsCountForPartition() const;
/// If the table contains too many active parts, sleep for a while to give them time to merge.
/// If until is non-null, wake up from the sleep earlier if the event happened.
void delayInsertIfNeeded(Poco::Event * until = nullptr);
/// Renames temporary part to a permanent part and adds it to the working set.
/// If increment != nullptr, part index is determing using increment. Otherwise part index remains unchanged.
/// It is assumed that the part does not intersect with existing parts.
/// If out_transaction != nullptr, sets it to an object allowing to rollback part addition (but not the renaming).
void renameTempPartAndAdd(MutableDataPartPtr & part, SimpleIncrement * increment = nullptr, Transaction * out_transaction = nullptr);
/// The same as renameTempPartAndAdd but the part can intersect existing parts.
/// Deletes and returns all parts covered by the added part (in ascending order).
DataPartsVector renameTempPartAndReplace(
MutableDataPartPtr & part, SimpleIncrement * increment = nullptr, Transaction * out_transaction = nullptr);
/// Removes parts from the working set parts.
/// Parts in add must already be in data_parts with PreCommitted, Committed, or Outdated states.
/// If clear_without_timeout is true, the parts will be deleted at once, or during the next call to
/// clearOldParts (ignoring old_parts_lifetime).
void removePartsFromWorkingSet(const DataPartsVector & remove, bool clear_without_timeout);
/// Renames the part to detached/<prefix>_<part> and forgets about it. The data won't be deleted in
/// clearOldParts.
/// If restore_covered is true, adds to the working set inactive parts, which were merged into the deleted part.
void renameAndDetachPart(const DataPartPtr & part, const String & prefix = "", bool restore_covered = false, bool move_to_detached = true);
/// Returns old inactive parts that can be deleted. At the same time removes them from the list of parts
/// but not from the disk.
DataPartsVector grabOldParts();
/// Reverts the changes made by grabOldParts(), parts should be in Deleting state.
void rollbackDeletingParts(const DataPartsVector & parts);
/// Removes parts from data_parts, they should be in Deleting state
void removePartsFinally(const DataPartsVector & parts);
/// Delete irrelevant parts.
void clearOldParts();
/// Deleate all directories which names begin with "tmp"
/// Set non-negative parameter value to override MergeTreeSettings temporary_directories_lifetime
void clearOldTemporaryDirectories(ssize_t custom_directories_lifetime_seconds = -1);
/// After the call to dropAllData() no method can be called.
/// Deletes the data directory and flushes the uncompressed blocks cache and the marks cache.
void dropAllData();
/// Moves the entire data directory.
/// Flushes the uncompressed blocks cache and the marks cache.
/// Must be called with locked lockStructureForAlter().
void setPath(const String & full_path, bool move_data);
/// Check if the ALTER can be performed:
/// - all needed columns are present.
/// - all type conversions can be done.
/// - columns corresponding to primary key, sign, sampling expression and date are not affected.
/// If something is wrong, throws an exception.
void checkAlter(const AlterCommands & commands);
/// Performs ALTER of the data part, writes the result to temporary files.
/// Returns an object allowing to rename temporary files to permanent files.
/// If the number of affected columns is suspiciously high and skip_sanity_checks is false, throws an exception.
/// If no data transformations are necessary, returns nullptr.
AlterDataPartTransactionPtr alterDataPart(
const DataPartPtr & part,
const NamesAndTypesList & new_columns,
const ASTPtr & new_primary_key,
bool skip_sanity_checks);
/// Must be called with locked lockStructureForAlter().
void setColumnsList(const NamesAndTypesList & new_columns) { columns = std::make_shared<NamesAndTypesList>(new_columns); }
/// Should be called if part data is suspected to be corrupted.
void reportBrokenPart(const String & name)
{
broken_part_callback(name);
}
ExpressionActionsPtr getPrimaryExpression() const { return primary_expr; }
SortDescription getSortDescription() const { return sort_descr; }
/// Check that the part is not broken and calculate the checksums for it if they are not present.
MutableDataPartPtr loadPartAndFixMetadata(const String & relative_path);
/** Create local backup (snapshot) for parts with specified prefix.
* Backup is created in directory clickhouse_dir/shadow/i/, where i - incremental number,
* or if 'with_name' is specified - backup is created in directory with specified name.
*/
void freezePartition(const ASTPtr & partition, const String & with_name, const Context & context);
/// Returns the size of partition in bytes.
size_t getPartitionSize(const std::string & partition_id) const;
struct ColumnSize
{
size_t marks = 0;
size_t data_compressed = 0;
size_t data_uncompressed = 0;
size_t getTotalCompressedSize() const
{
return marks + data_compressed;
}
};
size_t getColumnCompressedSize(const std::string & name) const
{
std::lock_guard<std::mutex> lock{data_parts_mutex};
const auto it = column_sizes.find(name);
return it == std::end(column_sizes) ? 0 : it->second.data_compressed;
}
using ColumnSizes = std::unordered_map<std::string, ColumnSize>;
ColumnSizes getColumnSizes() const
{
std::lock_guard<std::mutex> lock{data_parts_mutex};
return column_sizes;
}
/// NOTE Could be off after DROPped and MODIFYed columns in ALTER. Doesn't include primary.idx.
size_t getTotalCompressedSize() const
{
std::lock_guard<std::mutex> lock{data_parts_mutex};
size_t total_size = 0;
for (const auto & col : column_sizes)
total_size += col.second.getTotalCompressedSize();
return total_size;
}
void recalculateColumnSizes()
{
std::lock_guard<std::mutex> lock{data_parts_mutex};
calculateColumnSizesImpl();
}
/// For ATTACH/DETACH/DROP PARTITION.
String getPartitionIDFromQuery(const ASTPtr & partition, const Context & context);
MergeTreeDataFormatVersion format_version;
Context & context;
const ASTPtr sampling_expression;
const size_t index_granularity;
/// Merging params - what additional actions to perform during merge.
const MergingParams merging_params;
const MergeTreeSettings settings;
ASTPtr primary_expr_ast;
Block primary_key_sample;
DataTypes primary_key_data_types;
ASTPtr partition_expr_ast;
ExpressionActionsPtr partition_expr;
Names partition_expr_columns;
DataTypes partition_expr_column_types;
ExpressionActionsPtr minmax_idx_expr;
Names minmax_idx_columns;
DataTypes minmax_idx_column_types;
Int64 minmax_idx_date_column_pos = -1; /// In a common case minmax index includes a date column.
SortDescription minmax_idx_sort_descr; /// For use with PKCondition.
/// Limiting parallel sends per one table, used in DataPartsExchange
std::atomic_uint current_table_sends {0};
/// For generating names of temporary parts during insertion.
SimpleIncrement insert_increment;
private:
friend struct MergeTreeDataPart;
friend class StorageMergeTree;
friend class ReplicatedMergeTreeAlterThread;
friend class MergeTreeDataMerger;
bool require_part_metadata;
ExpressionActionsPtr primary_expr;
SortDescription sort_descr;
String database_name;
String table_name;
String full_path;
NamesAndTypesListPtr columns;
/// Current column sizes in compressed and uncompressed form.
ColumnSizes column_sizes;
/// Engine-specific methods
BrokenPartCallback broken_part_callback;
String log_name;
Logger * log;
/// Current set of data parts.
DataParts data_parts;
mutable std::mutex data_parts_mutex;
// TODO: this mutex could be a bottleneck. If so, make it shared, and split parts onto partitions
/// Used to serialize calls to grabOldParts.
std::mutex grab_old_parts_mutex;
/// The same for clearOldTemporaryDirectories.
std::mutex clear_old_temporary_directories_mutex;
/// Check that columns list doesn't contain multidimensional arrays.
/// If attach is true (attaching an existing table), writes an error message to log.
/// Otherwise (new table or alter) throws an exception.
void checkNoMultidimensionalArrays(const NamesAndTypesList & columns, bool attach) const;
void initPrimaryKey();
void initPartitionKey();
/// Expression for column type conversion.
/// If no conversions are needed, out_expression=nullptr.
/// out_rename_map maps column files for the out_expression onto new table files.
/// out_force_update_metadata denotes if metadata must be changed even if out_rename_map is empty (used
/// for transformation-free changing of Enum values list).
/// Files to be deleted are mapped to an empty string in out_rename_map.
/// If part == nullptr, just checks that all type conversions are possible.
void createConvertExpression(const DataPartPtr & part, const NamesAndTypesList & old_columns, const NamesAndTypesList & new_columns,
ExpressionActionsPtr & out_expression, NameToNameMap & out_rename_map, bool & out_force_update_metadata) const;
/// Calculates column sizes in compressed form for the current state of data_parts. Call with data_parts mutex locked.
void calculateColumnSizesImpl();
/// Adds or subtracts the contribution of the part to compressed column sizes.
void addPartContributionToColumnSizes(const DataPartPtr & part);
void removePartContributionToColumnSizes(const DataPartPtr & part);
/// If there is no part in the partition with ID `partition_id`, returns empty ptr. Should be called under the lock.
DataPartPtr getAnyPartInPartition(const String & partition_id, std::lock_guard<std::mutex> & data_parts_lock);
};
}