public interface ReadableChannel extends ReadableByteChannel
Modifier and Type | Method and Description |
---|---|
int |
blockSize()
Get the recommended alignment for reads.
|
long |
position()
Get the current position of the channel.
|
void |
position(long newPosition)
Seek the current position of the channel to a new offset.
|
void |
setReadAheadBytes(int bufferSize)
Recommend the channel maintain a read-ahead buffer.
|
long |
size()
Get the total size of the channel.
|
read
long position() throws IOException
IOException
- the channel's current position cannot be obtained.void position(long newPosition) throws IOException
newPosition
- position to move the channel to. The next read will start from
here. This should be a multiple of the blockSize()
.IOException
- the position cannot be updated. This may be because the
channel only supports block aligned IO and the current
position is not block aligned.long size() throws IOException
Prior to reading from a channel the size might not yet be known. Implementors may return -1 until after the first read method call. Once a read has been completed, the underlying file size should be available.
IOException
- the size cannot be determined.int blockSize()
Starting a read at multiples of the blockSize is more efficient than starting a read at any other position. If 0 or -1 the channel does not have any specific block size recommendation.
Channels should not recommend large block sizes. Sizes up to 1-4 MiB may
be reasonable, but sizes above that may be horribly inefficient. The
DfsBlockCache
favors the alignment suggested by the channel
rather than the configured size under the assumption that reads are very
expensive and the channel knows what size is best to access it with.
void setReadAheadBytes(int bufferSize) throws IOException
A read-ahead buffer of approximately bufferSize
in bytes may be
allocated and used by the channel to smooth out latency for read.
Callers can continue to read in smaller than bufferSize
chunks.
With read-ahead buffering enabled read latency may fluctuate in a pattern
of one slower read followed by (bufferSize / readSize) - 1
fast
reads satisfied by the read-ahead buffer. When summed up overall time to
read the same contiguous range should be lower than if read-ahead was not
enabled, as the implementation can combine reads to increase throughput.
To avoid unnecessary IO callers should only enable read-ahead if the majority of the channel will be accessed in order.
Implementations may chose to read-ahead using asynchronous APIs or background threads, or may simply aggregate reads using a buffer.
This read ahead stays in effect until the channel is closed or the buffer size is set to 0.
bufferSize
- requested size of the read ahead buffer, in bytes.IOException
- if the read ahead cannot be adjusted.Copyright © 2015 Eclipse JGit Project. All rights reserved.