Input / output

Interfaces

class arrow::io::FileInterface

Subclassed by arrow::io::InputStream, arrow::io::OutputStream

Public Functions

virtual Status Close() = 0

Close the stream cleanly.

For writable streams, this will attempt to flush any pending data before releasing the underlying resource.

After Close() is called, closed() returns true and the stream is not available for further operations.

virtual Status Abort()

Close the stream abruptly.

This method does not guarantee that any pending data is flushed. It merely releases any underlying resource used by the stream for its operation.

After Abort() is called, closed() returns true and the stream is not available for further operations.

virtual Result<int64_t> Tell() const = 0

Return the position in this stream.

virtual bool closed() const = 0

Return whether the stream is closed.

class arrow::io::Readable

Subclassed by arrow::io::InputStream

Public Functions

virtual Result<int64_t> Read(int64_t nbytes, void *out) = 0

Read data from current file position.

Read at most nbytes from the current file position into out. The number of bytes read is returned.

virtual Result<std::shared_ptr<Buffer>> Read(int64_t nbytes) = 0

Read data from current file position.

Read at most nbytes from the current file position. Less bytes may be read if EOF is reached. This method updates the current file position.

In some cases (e.g. a memory-mapped file), this method may avoid a memory copy.

virtual const IOContext &io_context() const

EXPERIMENTAL: The IOContext associated with this file.

By default, this is the same as default_io_context(), but it may be overriden by subclasses.

class Seekable

Subclassed by arrow::io::RandomAccessFile, arrow::io::WritableFile

class arrow::io::Writable

Subclassed by arrow::io::OutputStream

Public Functions

virtual Status Write(const void *data, int64_t nbytes) = 0

Write the given data to the stream.

This method always processes the bytes in full. Depending on the semantics of the stream, the data may be written out immediately, held in a buffer, or written asynchronously. In the case where the stream buffers the data, it will be copied. To avoid potentially large copies, use the Write variant that takes an owned Buffer.

virtual Status Write(const std::shared_ptr<Buffer> &data)

Write the given data to the stream.

Since the Buffer owns its memory, this method can avoid a copy if buffering is required. See Write(const void*, int64_t) for details.

virtual Status Flush()

Flush buffered bytes, if any.

class arrow::io::InputStream : public virtual arrow::io::FileInterface, public virtual arrow::io::Readable, public std::enable_shared_from_this<InputStream>

Subclassed by arrow::io::internal::InputStreamConcurrencyWrapper< Derived >, arrow::io::RandomAccessFile, arrow::io::StdinStream, arrow::io::TransformInputStream, arrow::io::internal::InputStreamConcurrencyWrapper< BufferedInputStream >, arrow::io::internal::InputStreamConcurrencyWrapper< CompressedInputStream >, arrow::io::SlowInputStreamBase< InputStream >

Public Functions

Status Advance(int64_t nbytes)

Advance or skip stream indicated number of bytes.

Parameters

[in] nbytes – the number to move forward

Returns

Status

virtual Result<util::string_view> Peek(int64_t nbytes)

Return zero-copy string_view to upcoming bytes.

Do not modify the stream position. The view becomes invalid after any operation on the stream. May trigger buffering if the requested size is larger than the number of buffered bytes.

May return NotImplemented on streams that don’t support it.

Parameters

[in] nbytes – the maximum number of bytes to see

virtual bool supports_zero_copy() const

Return true if InputStream is capable of zero copy Buffer reads.

Zero copy reads imply the use of Buffer-returning Read() overloads.

virtual Result<std::shared_ptr<const KeyValueMetadata>> ReadMetadata()

Read and return stream metadata.

If the stream implementation doesn’t support metadata, empty metadata is returned. Note that it is allowed to return a null pointer rather than an allocated empty metadata.

virtual Future<std::shared_ptr<const KeyValueMetadata>> ReadMetadataAsync(const IOContext &io_context)

Read stream metadata asynchronously.

class arrow::io::RandomAccessFile : public arrow::io::InputStream, public arrow::io::Seekable

Subclassed by arrow::io::HdfsReadableFile, arrow::io::internal::RandomAccessFileConcurrencyWrapper< Derived >, arrow::io::ReadWriteFileInterface, arrow::py::PyReadableFile, arrow::io::internal::RandomAccessFileConcurrencyWrapper< BufferReader >, arrow::io::internal::RandomAccessFileConcurrencyWrapper< CudaBufferReader >, arrow::io::internal::RandomAccessFileConcurrencyWrapper< ReadableFile >, arrow::io::SlowInputStreamBase< RandomAccessFile >

Public Functions

~RandomAccessFile() override

Necessary because we hold a std::unique_ptr.

virtual Result<int64_t> GetSize() = 0

Return the total file size in bytes.

This method does not read or move the current file position, so is safe to call concurrently with e.g. ReadAt().

virtual Result<int64_t> ReadAt(int64_t position, int64_t nbytes, void *out)

Read data from given file position.

At most nbytes bytes are read. The number of bytes read is returned (it can be less than nbytes if EOF is reached).

This method can be safely called from multiple threads concurrently. It is unspecified whether this method updates the file position or not.

The default RandomAccessFile-provided implementation uses Seek() and Read(), but subclasses may override it with a more efficient implementation that doesn’t depend on implicit file positioning.

Parameters

  • [in] position – Where to read bytes from

  • [in] nbytes – The number of bytes to read

  • [out] out – The buffer to read bytes into

Returns

The number of bytes read, or an error

virtual Result<std::shared_ptr<Buffer>> ReadAt(int64_t position, int64_t nbytes)

Read data from given file position.

At most nbytes bytes are read, but it can be less if EOF is reached.

Parameters

  • [in] position – Where to read bytes from

  • [in] nbytes – The number of bytes to read

Returns

A buffer containing the bytes read, or an error

virtual Future<std::shared_ptr<Buffer>> ReadAsync(const IOContext&, int64_t position, int64_t nbytes)

EXPERIMENTAL: Read data asynchronously.

Future<std::shared_ptr<Buffer>> ReadAsync(int64_t position, int64_t nbytes)

EXPERIMENTAL: Read data asynchronously, using the file’s IOContext.

virtual Status WillNeed(const std::vector<ReadRange> &ranges)

EXPERIMENTAL: Inform that the given ranges may be read soon.

Some implementations might arrange to prefetch some of the data. However, no guarantee is made and the default implementation does nothing. For robust prefetching, use ReadAt() or ReadAsync().

Public Static Functions

static std::shared_ptr<InputStream> GetStream(std::shared_ptr<RandomAccessFile> file, int64_t file_offset, int64_t nbytes)

Create an isolated InputStream that reads a segment of a RandomAccessFile.

Multiple such stream can be created and used independently without interference

Parameters

  • [in] file – a file instance

  • [in] file_offset – the starting position in the file

  • [in] nbytes – the extent of bytes to read. The file should have sufficient bytes available

class OutputStream : public virtual arrow::io::FileInterface, public arrow::io::Writable

Subclassed by arrow::io::BufferedOutputStream, arrow::io::BufferOutputStream, arrow::io::CompressedOutputStream, arrow::io::FileOutputStream, arrow::io::HdfsOutputStream, arrow::io::MockOutputStream, arrow::io::StderrStream, arrow::io::StdoutStream, arrow::io::WritableFile, arrow::py::PyOutputStream

class ReadWriteFileInterface : public arrow::io::RandomAccessFile, public arrow::io::WritableFile

Subclassed by arrow::io::MemoryMappedFile

Concrete implementations

In-memory streams

class arrow::io::BufferReader : public arrow::io::internal::RandomAccessFileConcurrencyWrapper<BufferReader>

Random access zero-copy reads on an arrow::Buffer.

Public Functions

explicit BufferReader(const util::string_view &data)

Instantiate from std::string or arrow::util::string_view.

Does not own data

virtual bool closed() const override

Return whether the stream is closed.

virtual bool supports_zero_copy() const override

Return true if InputStream is capable of zero copy Buffer reads.

Zero copy reads imply the use of Buffer-returning Read() overloads.

virtual Future<std::shared_ptr<Buffer>> ReadAsync(const IOContext&, int64_t position, int64_t nbytes) override

EXPERIMENTAL: Read data asynchronously.

virtual Status WillNeed(const std::vector<ReadRange> &ranges) override

EXPERIMENTAL: Inform that the given ranges may be read soon.

Some implementations might arrange to prefetch some of the data. However, no guarantee is made and the default implementation does nothing. For robust prefetching, use ReadAt() or ReadAsync().

class arrow::io::MockOutputStream : public arrow::io::OutputStream

A helper class to track the size of allocations.

Writes to this stream do not copy or retain any data, they just bump a size counter that can be later used to know exactly which data size needs to be allocated for actual writing.

Public Functions

virtual Status Close() override

Close the stream cleanly.

For writable streams, this will attempt to flush any pending data before releasing the underlying resource.

After Close() is called, closed() returns true and the stream is not available for further operations.

virtual bool closed() const override

Return whether the stream is closed.

virtual Result<int64_t> Tell() const override

Return the position in this stream.

virtual Status Write(const void *data, int64_t nbytes) override

Write the given data to the stream.

This method always processes the bytes in full. Depending on the semantics of the stream, the data may be written out immediately, held in a buffer, or written asynchronously. In the case where the stream buffers the data, it will be copied. To avoid potentially large copies, use the Write variant that takes an owned Buffer.

class arrow::io::BufferOutputStream : public arrow::io::OutputStream

An output stream that writes to a resizable buffer.

Public Functions

virtual Status Close() override

Close the stream, preserving the buffer (retrieve it with Finish()).

virtual bool closed() const override

Return whether the stream is closed.

virtual Result<int64_t> Tell() const override

Return the position in this stream.

virtual Status Write(const void *data, int64_t nbytes) override

Write the given data to the stream.

This method always processes the bytes in full. Depending on the semantics of the stream, the data may be written out immediately, held in a buffer, or written asynchronously. In the case where the stream buffers the data, it will be copied. To avoid potentially large copies, use the Write variant that takes an owned Buffer.

Result<std::shared_ptr<Buffer>> Finish()

Close the stream and return the buffer.

Status Reset(int64_t initial_capacity = 1024, MemoryPool *pool = default_memory_pool())

Initialize state of OutputStream with newly allocated memory and set position to 0.

Parameters

  • [in] initial_capacity – the starting allocated capacity

  • [inout] pool – the memory pool to use for allocations

Returns

Status

Public Static Functions

static Result<std::shared_ptr<BufferOutputStream>> Create(int64_t initial_capacity = 4096, MemoryPool *pool = default_memory_pool())

Create in-memory output stream with indicated capacity using a memory pool.

Parameters

  • [in] initial_capacity – the initial allocated internal capacity of the OutputStream

  • [inout] pool – a MemoryPool to use for allocations

Returns

the created stream

class arrow::io::FixedSizeBufferWriter : public arrow::io::WritableFile

An output stream that writes into a fixed-size mutable buffer.

Public Functions

explicit FixedSizeBufferWriter(const std::shared_ptr<Buffer> &buffer)

Input buffer must be mutable, will abort if not.

virtual Status Close() override

Close the stream cleanly.

For writable streams, this will attempt to flush any pending data before releasing the underlying resource.

After Close() is called, closed() returns true and the stream is not available for further operations.

virtual bool closed() const override

Return whether the stream is closed.

virtual Result<int64_t> Tell() const override

Return the position in this stream.

virtual Status Write(const void *data, int64_t nbytes) override

Write the given data to the stream.

This method always processes the bytes in full. Depending on the semantics of the stream, the data may be written out immediately, held in a buffer, or written asynchronously. In the case where the stream buffers the data, it will be copied. To avoid potentially large copies, use the Write variant that takes an owned Buffer.

Local files

class arrow::io::ReadableFile : public arrow::io::internal::RandomAccessFileConcurrencyWrapper<ReadableFile>

An operating system file open in read-only mode.

Reads through this implementation are unbuffered. If many small reads need to be issued, it is recommended to use a buffering layer for good performance.

Public Functions

virtual bool closed() const override

Return whether the stream is closed.

virtual Status WillNeed(const std::vector<ReadRange> &ranges) override

EXPERIMENTAL: Inform that the given ranges may be read soon.

Some implementations might arrange to prefetch some of the data. However, no guarantee is made and the default implementation does nothing. For robust prefetching, use ReadAt() or ReadAsync().

Public Static Functions

static Result<std::shared_ptr<ReadableFile>> Open(const std::string &path, MemoryPool *pool = default_memory_pool())

Open a local file for reading.

Parameters

  • [in] path – with UTF8 encoding

  • [in] pool – a MemoryPool for memory allocations

Returns

ReadableFile instance

static Result<std::shared_ptr<ReadableFile>> Open(int fd, MemoryPool *pool = default_memory_pool())

Open a local file for reading.

The file descriptor becomes owned by the

ReadableFile, and will be closed on Close() or destruction.

Parameters

  • [in] fd – file descriptor

  • [in] pool – a MemoryPool for memory allocations

Returns

ReadableFile instance

class arrow::io::FileOutputStream : public arrow::io::OutputStream

An operating system file open in write-only mode.

Public Functions

virtual Status Close() override

Close the stream cleanly.

For writable streams, this will attempt to flush any pending data before releasing the underlying resource.

After Close() is called, closed() returns true and the stream is not available for further operations.

virtual bool closed() const override

Return whether the stream is closed.

virtual Result<int64_t> Tell() const override

Return the position in this stream.

virtual Status Write(const void *data, int64_t nbytes) override

Write the given data to the stream.

This method always processes the bytes in full. Depending on the semantics of the stream, the data may be written out immediately, held in a buffer, or written asynchronously. In the case where the stream buffers the data, it will be copied. To avoid potentially large copies, use the Write variant that takes an owned Buffer.

Public Static Functions

static Result<std::shared_ptr<FileOutputStream>> Open(const std::string &path, bool append = false)

Open a local file for writing, truncating any existing file.

When opening a new file, any existing file with the indicated path is truncated to 0 bytes, deleting any existing data

Parameters

  • [in] path – with UTF8 encoding

  • [in] append – append to existing file, otherwise truncate to 0 bytes

Returns

an open FileOutputStream

static Result<std::shared_ptr<FileOutputStream>> Open(int fd)

Open a file descriptor for writing.

The underlying file isn’t truncated.

The file descriptor becomes owned by the

OutputStream, and will be closed on Close() or destruction.

Parameters

[in] fd – file descriptor

Returns

an open FileOutputStream

class arrow::io::MemoryMappedFile : public arrow::io::ReadWriteFileInterface

A file interface that uses memory-mapped files for memory interactions.

This implementation supports zero-copy reads. The same class is used for both reading and writing.

If opening a file in a writable mode, it is not truncated first as with FileOutputStream.

Public Functions

virtual Status Close() override

Close the stream cleanly.

For writable streams, this will attempt to flush any pending data before releasing the underlying resource.

After Close() is called, closed() returns true and the stream is not available for further operations.

virtual bool closed() const override

Return whether the stream is closed.

virtual Result<int64_t> Tell() const override

Return the position in this stream.

virtual Result<int64_t> Read(int64_t nbytes, void *out) override

Read data from current file position.

Read at most nbytes from the current file position into out. The number of bytes read is returned.

virtual Result<std::shared_ptr<Buffer>> Read(int64_t nbytes) override

Read data from current file position.

Read at most nbytes from the current file position. Less bytes may be read if EOF is reached. This method updates the current file position.

In some cases (e.g. a memory-mapped file), this method may avoid a memory copy.

virtual Result<std::shared_ptr<Buffer>> ReadAt(int64_t position, int64_t nbytes) override

Read data from given file position.

At most nbytes bytes are read, but it can be less if EOF is reached.

Parameters

  • [in] position – Where to read bytes from

  • [in] nbytes – The number of bytes to read

Returns

A buffer containing the bytes read, or an error

virtual Result<int64_t> ReadAt(int64_t position, int64_t nbytes, void *out) override

Read data from given file position.

At most nbytes bytes are read. The number of bytes read is returned (it can be less than nbytes if EOF is reached).

This method can be safely called from multiple threads concurrently. It is unspecified whether this method updates the file position or not.

The default RandomAccessFile-provided implementation uses Seek() and Read(), but subclasses may override it with a more efficient implementation that doesn’t depend on implicit file positioning.

Parameters

  • [in] position – Where to read bytes from

  • [in] nbytes – The number of bytes to read

  • [out] out – The buffer to read bytes into

Returns

The number of bytes read, or an error

virtual Future<std::shared_ptr<Buffer>> ReadAsync(const IOContext&, int64_t position, int64_t nbytes) override

EXPERIMENTAL: Read data asynchronously.

virtual Status WillNeed(const std::vector<ReadRange> &ranges) override

EXPERIMENTAL: Inform that the given ranges may be read soon.

Some implementations might arrange to prefetch some of the data. However, no guarantee is made and the default implementation does nothing. For robust prefetching, use ReadAt() or ReadAsync().

virtual bool supports_zero_copy() const override

Return true if InputStream is capable of zero copy Buffer reads.

Zero copy reads imply the use of Buffer-returning Read() overloads.

virtual Status Write(const void *data, int64_t nbytes) override

Write data at the current position in the file. Thread-safe.

Status Resize(int64_t new_size)

Set the size of the map to new_size.

virtual Status WriteAt(int64_t position, const void *data, int64_t nbytes) override

Write data at a particular position in the file. Thread-safe.

virtual Result<int64_t> GetSize() override

Return the total file size in bytes.

This method does not read or move the current file position, so is safe to call concurrently with e.g. ReadAt().

Public Static Functions

static Result<std::shared_ptr<MemoryMappedFile>> Create(const std::string &path, int64_t size)

Create new file with indicated size, return in read/write mode.

Buffering input / output wrappers

class arrow::io::BufferedInputStream : public arrow::io::internal::InputStreamConcurrencyWrapper<BufferedInputStream>

An InputStream that performs buffered reads from an unbuffered InputStream, which can mitigate the overhead of many small reads in some cases.

Public Functions

Status SetBufferSize(int64_t new_buffer_size)

Resize internal read buffer; calls to Read(…) will read at least.

Parameters

[in] new_buffer_size – the new read buffer size

Returns

Status

int64_t bytes_buffered() const

Return the number of remaining bytes in the read buffer.

int64_t buffer_size() const

Return the current size of the internal buffer.

std::shared_ptr<InputStream> Detach()

Release the raw InputStream.

Any data buffered will be discarded. Further operations on this object are invalid

Returns

raw the underlying InputStream

std::shared_ptr<InputStream> raw() const

Return the unbuffered InputStream.

virtual bool closed() const override

Return whether the stream is closed.

virtual Result<std::shared_ptr<const KeyValueMetadata>> ReadMetadata() override

Read and return stream metadata.

If the stream implementation doesn’t support metadata, empty metadata is returned. Note that it is allowed to return a null pointer rather than an allocated empty metadata.

virtual Future<std::shared_ptr<const KeyValueMetadata>> ReadMetadataAsync(const IOContext &io_context) override

Read stream metadata asynchronously.

Public Static Functions

static Result<std::shared_ptr<BufferedInputStream>> Create(int64_t buffer_size, MemoryPool *pool, std::shared_ptr<InputStream> raw, int64_t raw_read_bound = -1)

Create a BufferedInputStream from a raw InputStream.

Parameters

  • [in] buffer_size – the size of the temporary read buffer

  • [in] pool – a MemoryPool to use for allocations

  • [in] raw – a raw InputStream

  • [in] raw_read_bound – a bound on the maximum number of bytes to read from the raw input stream. The default -1 indicates that it is unbounded

Returns

the created BufferedInputStream

class arrow::io::BufferedOutputStream : public arrow::io::OutputStream

Public Functions

Status SetBufferSize(int64_t new_buffer_size)

Resize internal buffer.

Parameters

[in] new_buffer_size – the new buffer size

Returns

Status

int64_t buffer_size() const

Return the current size of the internal buffer.

int64_t bytes_buffered() const

Return the number of remaining bytes that have not been flushed to the raw OutputStream.

Result<std::shared_ptr<OutputStream>> Detach()

Flush any buffered writes and release the raw OutputStream.

Further operations on this object are invalid

Returns

the underlying OutputStream

virtual Status Close() override

Close the buffered output stream.

This implicitly closes the underlying raw output stream.

virtual Status Abort() override

Close the stream abruptly.

This method does not guarantee that any pending data is flushed. It merely releases any underlying resource used by the stream for its operation.

After Abort() is called, closed() returns true and the stream is not available for further operations.

virtual bool closed() const override

Return whether the stream is closed.

virtual Result<int64_t> Tell() const override

Return the position in this stream.

virtual Status Write(const void *data, int64_t nbytes) override

Write the given data to the stream.

This method always processes the bytes in full. Depending on the semantics of the stream, the data may be written out immediately, held in a buffer, or written asynchronously. In the case where the stream buffers the data, it will be copied. To avoid potentially large copies, use the Write variant that takes an owned Buffer.

virtual Status Write(const std::shared_ptr<Buffer> &data) override

Write the given data to the stream.

Since the Buffer owns its memory, this method can avoid a copy if buffering is required. See Write(const void*, int64_t) for details.

virtual Status Flush() override

Flush buffered bytes, if any.

std::shared_ptr<OutputStream> raw() const

Return the underlying raw output stream.

Public Static Functions

static Result<std::shared_ptr<BufferedOutputStream>> Create(int64_t buffer_size, MemoryPool *pool, std::shared_ptr<OutputStream> raw)

Create a buffered output stream wrapping the given output stream.

Parameters

  • [in] buffer_size – the size of the temporary write buffer

  • [in] pool – a MemoryPool to use for allocations

  • [in] raw – another OutputStream

Returns

the created BufferedOutputStream

Compressed input / output wrappers

class arrow::io::CompressedInputStream : public arrow::io::internal::InputStreamConcurrencyWrapper<CompressedInputStream>

Public Functions

virtual bool closed() const override

Return whether the stream is closed.

virtual Result<std::shared_ptr<const KeyValueMetadata>> ReadMetadata() override

Read and return stream metadata.

If the stream implementation doesn’t support metadata, empty metadata is returned. Note that it is allowed to return a null pointer rather than an allocated empty metadata.

virtual Future<std::shared_ptr<const KeyValueMetadata>> ReadMetadataAsync(const IOContext &io_context) override

Read stream metadata asynchronously.

std::shared_ptr<InputStream> raw() const

Return the underlying raw input stream.

Public Static Functions

static Result<std::shared_ptr<CompressedInputStream>> Make(util::Codec *codec, const std::shared_ptr<InputStream> &raw, MemoryPool *pool = default_memory_pool())

Create a compressed input stream wrapping the given input stream.

class arrow::io::CompressedOutputStream : public arrow::io::OutputStream

Public Functions

virtual Status Close() override

Close the compressed output stream.

This implicitly closes the underlying raw output stream.

virtual Status Abort() override

Close the stream abruptly.

This method does not guarantee that any pending data is flushed. It merely releases any underlying resource used by the stream for its operation.

After Abort() is called, closed() returns true and the stream is not available for further operations.

virtual bool closed() const override

Return whether the stream is closed.

virtual Result<int64_t> Tell() const override

Return the position in this stream.

virtual Status Write(const void *data, int64_t nbytes) override

Write the given data to the stream.

This method always processes the bytes in full. Depending on the semantics of the stream, the data may be written out immediately, held in a buffer, or written asynchronously. In the case where the stream buffers the data, it will be copied. To avoid potentially large copies, use the Write variant that takes an owned Buffer.

virtual Status Flush() override

Flush buffered bytes, if any.

std::shared_ptr<OutputStream> raw() const

Return the underlying raw output stream.

Public Static Functions

static Result<std::shared_ptr<CompressedOutputStream>> Make(util::Codec *codec, const std::shared_ptr<OutputStream> &raw, MemoryPool *pool = default_memory_pool())

Create a compressed output stream wrapping the given output stream.

previous

Utilities

next

Arrow IPC