| 1 | #include <Storages/MergeTree/MergeTreePartition.h> |
| 2 | #include <Storages/MergeTree/MergeTreeData.h> |
| 3 | #include <Storages/MergeTree/MergeTreeDataPart.h> |
| 4 | #include <IO/ReadBufferFromFile.h> |
| 5 | #include <IO/HashingWriteBuffer.h> |
| 6 | #include <Common/FieldVisitors.h> |
| 7 | #include <DataTypes/DataTypeDate.h> |
| 8 | #include <DataTypes/DataTypeTuple.h> |
| 9 | #include <Columns/ColumnTuple.h> |
| 10 | #include <Common/SipHash.h> |
| 11 | #include <Common/typeid_cast.h> |
| 12 | #include <Common/hex.h> |
| 13 | #include <Core/Block.h> |
| 14 | |
| 15 | #include <Poco/File.h> |
| 16 | |
| 17 | namespace DB |
| 18 | { |
| 19 | |
| 20 | static ReadBufferFromFile openForReading(const String & path) |
| 21 | { |
| 22 | return ReadBufferFromFile(path, std::min(static_cast<Poco::File::FileSize>(DBMS_DEFAULT_BUFFER_SIZE), Poco::File(path).getSize())); |
| 23 | } |
| 24 | |
| 25 | String MergeTreePartition::getID(const MergeTreeData & storage) const |
| 26 | { |
| 27 | return getID(storage.partition_key_sample); |
| 28 | } |
| 29 | |
| 30 | /// NOTE: This ID is used to create part names which are then persisted in ZK and as directory names on the file system. |
| 31 | /// So if you want to change this method, be sure to guarantee compatibility with existing table data. |
| 32 | String MergeTreePartition::getID(const Block & partition_key_sample) const |
| 33 | { |
| 34 | if (value.size() != partition_key_sample.columns()) |
| 35 | throw Exception("Invalid partition key size: " + toString(value.size()), ErrorCodes::LOGICAL_ERROR); |
| 36 | |
| 37 | if (value.empty()) |
| 38 | return "all" ; /// It is tempting to use an empty string here. But that would break directory structure in ZK. |
| 39 | |
| 40 | /// In case all partition fields are represented by integral types, try to produce a human-readable ID. |
| 41 | /// Otherwise use a hex-encoded hash. |
| 42 | bool are_all_integral = true; |
| 43 | for (const Field & field : value) |
| 44 | { |
| 45 | if (field.getType() != Field::Types::UInt64 && field.getType() != Field::Types::Int64) |
| 46 | { |
| 47 | are_all_integral = false; |
| 48 | break; |
| 49 | } |
| 50 | } |
| 51 | |
| 52 | String result; |
| 53 | |
| 54 | if (are_all_integral) |
| 55 | { |
| 56 | FieldVisitorToString to_string_visitor; |
| 57 | for (size_t i = 0; i < value.size(); ++i) |
| 58 | { |
| 59 | if (i > 0) |
| 60 | result += '-'; |
| 61 | |
| 62 | if (typeid_cast<const DataTypeDate *>(partition_key_sample.getByPosition(i).type.get())) |
| 63 | result += toString(DateLUT::instance().toNumYYYYMMDD(DayNum(value[i].safeGet<UInt64>()))); |
| 64 | else |
| 65 | result += applyVisitor(to_string_visitor, value[i]); |
| 66 | |
| 67 | /// It is tempting to output DateTime as YYYYMMDDhhmmss, but that would make partition ID |
| 68 | /// timezone-dependent. |
| 69 | } |
| 70 | |
| 71 | return result; |
| 72 | } |
| 73 | |
| 74 | SipHash hash; |
| 75 | FieldVisitorHash hashing_visitor(hash); |
| 76 | for (const Field & field : value) |
| 77 | applyVisitor(hashing_visitor, field); |
| 78 | |
| 79 | char hash_data[16]; |
| 80 | hash.get128(hash_data); |
| 81 | result.resize(32); |
| 82 | for (size_t i = 0; i < 16; ++i) |
| 83 | writeHexByteLowercase(hash_data[i], &result[2 * i]); |
| 84 | |
| 85 | return result; |
| 86 | } |
| 87 | |
| 88 | void MergeTreePartition::serializeText(const MergeTreeData & storage, WriteBuffer & out, const FormatSettings & format_settings) const |
| 89 | { |
| 90 | size_t key_size = storage.partition_key_sample.columns(); |
| 91 | |
| 92 | if (key_size == 0) |
| 93 | { |
| 94 | writeCString("tuple()" , out); |
| 95 | } |
| 96 | else if (key_size == 1) |
| 97 | { |
| 98 | const DataTypePtr & type = storage.partition_key_sample.getByPosition(0).type; |
| 99 | auto column = type->createColumn(); |
| 100 | column->insert(value[0]); |
| 101 | type->serializeAsText(*column, 0, out, format_settings); |
| 102 | } |
| 103 | else |
| 104 | { |
| 105 | DataTypes types; |
| 106 | Columns columns; |
| 107 | for (size_t i = 0; i < key_size; ++i) |
| 108 | { |
| 109 | const auto & type = storage.partition_key_sample.getByPosition(i).type; |
| 110 | types.push_back(type); |
| 111 | auto column = type->createColumn(); |
| 112 | column->insert(value[i]); |
| 113 | columns.push_back(std::move(column)); |
| 114 | } |
| 115 | |
| 116 | DataTypeTuple tuple_type(types); |
| 117 | auto tuple_column = ColumnTuple::create(columns); |
| 118 | tuple_type.serializeText(*tuple_column, 0, out, format_settings); |
| 119 | } |
| 120 | } |
| 121 | |
| 122 | void MergeTreePartition::load(const MergeTreeData & storage, const String & part_path) |
| 123 | { |
| 124 | if (!storage.partition_key_expr) |
| 125 | return; |
| 126 | |
| 127 | ReadBufferFromFile file = openForReading(part_path + "partition.dat" ); |
| 128 | value.resize(storage.partition_key_sample.columns()); |
| 129 | for (size_t i = 0; i < storage.partition_key_sample.columns(); ++i) |
| 130 | storage.partition_key_sample.getByPosition(i).type->deserializeBinary(value[i], file); |
| 131 | } |
| 132 | |
| 133 | void MergeTreePartition::store(const MergeTreeData & storage, const String & part_path, MergeTreeDataPartChecksums & checksums) const |
| 134 | { |
| 135 | store(storage.partition_key_sample, part_path, checksums); |
| 136 | } |
| 137 | |
| 138 | void MergeTreePartition::store(const Block & partition_key_sample, const String & part_path, MergeTreeDataPartChecksums & checksums) const |
| 139 | { |
| 140 | if (!partition_key_sample) |
| 141 | return; |
| 142 | |
| 143 | WriteBufferFromFile out(part_path + "partition.dat" ); |
| 144 | HashingWriteBuffer out_hashing(out); |
| 145 | for (size_t i = 0; i < value.size(); ++i) |
| 146 | partition_key_sample.getByPosition(i).type->serializeBinary(value[i], out_hashing); |
| 147 | out_hashing.next(); |
| 148 | checksums.files["partition.dat" ].file_size = out_hashing.count(); |
| 149 | checksums.files["partition.dat" ].file_hash = out_hashing.getHash(); |
| 150 | } |
| 151 | |
| 152 | } |
| 153 | |