1// Licensed to the Apache Software Foundation (ASF) under one
2// or more contributor license agreements. See the NOTICE file
3// distributed with this work for additional information
4// regarding copyright ownership. The ASF licenses this file
5// to you under the Apache License, Version 2.0 (the
6// "License"); you may not use this file except in compliance
7// with the License. You may obtain a copy of the License at
8//
9// http://www.apache.org/licenses/LICENSE-2.0
10//
11// Unless required by applicable law or agreed to in writing,
12// software distributed under the License is distributed on an
13// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
14// KIND, either express or implied. See the License for the
15// specific language governing permissions and limitations
16// under the License.
17
18#ifndef PARQUET_FILE_WRITER_H
19#define PARQUET_FILE_WRITER_H
20
21#include <cstdint>
22#include <memory>
23
24#include "parquet/metadata.h"
25#include "parquet/properties.h"
26#include "parquet/schema.h"
27#include "parquet/util/macros.h"
28#include "parquet/util/memory.h"
29#include "parquet/util/visibility.h"
30
31namespace parquet {
32
33class ColumnWriter;
34class PageWriter;
35class OutputStream;
36
37namespace schema {
38
39class GroupNode;
40
41} // namespace schema
42
43class PARQUET_EXPORT RowGroupWriter {
44 public:
45 // Forward declare a virtual class 'Contents' to aid dependency injection and more
46 // easily create test fixtures
47 // An implementation of the Contents class is defined in the .cc file
48 struct Contents {
49 virtual ~Contents() = default;
50 virtual int num_columns() const = 0;
51 virtual int64_t num_rows() const = 0;
52
53 // to be used only with ParquetFileWriter::AppendRowGroup
54 virtual ColumnWriter* NextColumn() = 0;
55 // to be used only with ParquetFileWriter::AppendBufferedRowGroup
56 virtual ColumnWriter* column(int i) = 0;
57
58 virtual int current_column() const = 0;
59 virtual void Close() = 0;
60
61 // total bytes written by the page writer
62 virtual int64_t total_bytes_written() const = 0;
63 // total bytes still compressed but not written
64 virtual int64_t total_compressed_bytes() const = 0;
65 };
66
67 explicit RowGroupWriter(std::unique_ptr<Contents> contents);
68
69 /// Construct a ColumnWriter for the indicated row group-relative column.
70 ///
71 /// To be used only with ParquetFileWriter::AppendRowGroup
72 /// Ownership is solely within the RowGroupWriter. The ColumnWriter is only
73 /// valid until the next call to NextColumn or Close. As the contents are
74 /// directly written to the sink, once a new column is started, the contents
75 /// of the previous one cannot be modified anymore.
76 ColumnWriter* NextColumn();
77 /// Index of currently written column
78 int current_column();
79 void Close();
80
81 int num_columns() const;
82
83 /// Construct a ColumnWriter for the indicated row group column.
84 ///
85 /// To be used only with ParquetFileWriter::AppendBufferedRowGroup
86 /// Ownership is solely within the RowGroupWriter. The ColumnWriter is
87 /// valid until Close. The contents are buffered in memory and written to sink
88 /// on Close
89 ColumnWriter* column(int i);
90
91 /**
92 * Number of rows that shall be written as part of this RowGroup.
93 */
94 int64_t num_rows() const;
95
96 int64_t total_bytes_written() const;
97 int64_t total_compressed_bytes() const;
98
99 private:
100 // Holds a pointer to an instance of Contents implementation
101 std::unique_ptr<Contents> contents_;
102};
103
104PARQUET_EXPORT
105void WriteFileMetaData(const FileMetaData& file_metadata, OutputStream* sink);
106
107class PARQUET_EXPORT ParquetFileWriter {
108 public:
109 // Forward declare a virtual class 'Contents' to aid dependency injection and more
110 // easily create test fixtures
111 // An implementation of the Contents class is defined in the .cc file
112 struct Contents {
113 Contents(const std::shared_ptr<::parquet::schema::GroupNode>& schema,
114 const std::shared_ptr<const KeyValueMetadata>& key_value_metadata)
115 : schema_(), key_value_metadata_(key_value_metadata) {
116 schema_.Init(schema);
117 }
118 virtual ~Contents() {}
119 // Perform any cleanup associated with the file contents
120 virtual void Close() = 0;
121
122 /// \note Deprecated since 1.3.0
123 RowGroupWriter* AppendRowGroup(int64_t num_rows);
124
125 virtual RowGroupWriter* AppendRowGroup() = 0;
126 virtual RowGroupWriter* AppendBufferedRowGroup() = 0;
127
128 virtual int64_t num_rows() const = 0;
129 virtual int num_columns() const = 0;
130 virtual int num_row_groups() const = 0;
131
132 virtual const std::shared_ptr<WriterProperties>& properties() const = 0;
133
134 const std::shared_ptr<const KeyValueMetadata>& key_value_metadata() const {
135 return key_value_metadata_;
136 }
137
138 // Return const-pointer to make it clear that this object is not to be copied
139 const SchemaDescriptor* schema() const { return &schema_; }
140
141 SchemaDescriptor schema_;
142
143 /// This should be the only place this is stored. Everything else is a const reference
144 std::shared_ptr<const KeyValueMetadata> key_value_metadata_;
145 };
146
147 ParquetFileWriter();
148 ~ParquetFileWriter();
149
150 static std::unique_ptr<ParquetFileWriter> Open(
151 const std::shared_ptr<::arrow::io::OutputStream>& sink,
152 const std::shared_ptr<schema::GroupNode>& schema,
153 const std::shared_ptr<WriterProperties>& properties = default_writer_properties(),
154 const std::shared_ptr<const KeyValueMetadata>& key_value_metadata = NULLPTR);
155
156 static std::unique_ptr<ParquetFileWriter> Open(
157 const std::shared_ptr<OutputStream>& sink,
158 const std::shared_ptr<schema::GroupNode>& schema,
159 const std::shared_ptr<WriterProperties>& properties = default_writer_properties(),
160 const std::shared_ptr<const KeyValueMetadata>& key_value_metadata = NULLPTR);
161
162 void Open(std::unique_ptr<Contents> contents);
163 void Close();
164
165 // Construct a RowGroupWriter for the indicated number of rows.
166 //
167 // Ownership is solely within the ParquetFileWriter. The RowGroupWriter is only valid
168 // until the next call to AppendRowGroup or AppendBufferedRowGroup or Close.
169 // @param num_rows The number of rows that are stored in the new RowGroup
170 //
171 // \deprecated Since 1.3.0
172 RowGroupWriter* AppendRowGroup(int64_t num_rows);
173
174 /// Construct a RowGroupWriter with an arbitrary number of rows.
175 ///
176 /// Ownership is solely within the ParquetFileWriter. The RowGroupWriter is only valid
177 /// until the next call to AppendRowGroup or AppendBufferedRowGroup or Close.
178 RowGroupWriter* AppendRowGroup();
179
180 /// Construct a RowGroupWriter that buffers all the values until the RowGroup is ready.
181 /// Use this if you want to write a RowGroup based on a certain size
182 ///
183 /// Ownership is solely within the ParquetFileWriter. The RowGroupWriter is only valid
184 /// until the next call to AppendRowGroup or AppendBufferedRowGroup or Close.
185 RowGroupWriter* AppendBufferedRowGroup();
186
187 /// Number of columns.
188 ///
189 /// This number is fixed during the lifetime of the writer as it is determined via
190 /// the schema.
191 int num_columns() const;
192
193 /// Number of rows in the yet started RowGroups.
194 ///
195 /// Changes on the addition of a new RowGroup.
196 int64_t num_rows() const;
197
198 /// Number of started RowGroups.
199 int num_row_groups() const;
200
201 /// Configuration passed to the writer, e.g. the used Parquet format version.
202 const std::shared_ptr<WriterProperties>& properties() const;
203
204 /// Returns the file schema descriptor
205 const SchemaDescriptor* schema() const;
206
207 /// Returns a column descriptor in schema
208 const ColumnDescriptor* descr(int i) const;
209
210 /// Returns the file custom metadata
211 const std::shared_ptr<const KeyValueMetadata>& key_value_metadata() const;
212
213 private:
214 // Holds a pointer to an instance of Contents implementation
215 std::unique_ptr<Contents> contents_;
216};
217
218} // namespace parquet
219
220#endif // PARQUET_FILE_WRITER_H
221