big-data/hadoop
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

HADOOP-18128. Fix typo issues of outputstream.md (#4025)

Browse Source
This commit is contained in:
ted12138 2022-03-02 18:25:56 +08:00 committed by GitHub
parent 9e475aede6
commit 902a7935e9
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
hadoop-common-project/hadoop-common/src/site/markdown/filesystem/outputstream.md
View File

@ -42,7 +42,7 @@ or `FSDataOutputStreamBuilder.build()`.
These all return instances of `FSDataOutputStream`, through which data These all return instances of `FSDataOutputStream`, through which data
can be written through various `write()` methods. can be written through various `write()` methods.
After a stream's `close()` method is called, all data written to the After a stream's `close()` method is called, all data written to the
stream MUST BE persisted to the fileysystem and visible to oll other stream MUST BE persisted to the filesystem and visible to oll other
clients attempting to read data from that path via `FileSystem.open()`. clients attempting to read data from that path via `FileSystem.open()`.
As well as operations to write the data, Hadoop's `OutputStream` implementations As well as operations to write the data, Hadoop's `OutputStream` implementations
@ -126,7 +126,7 @@ consistent with the data at the filesystem.
The output stream returned from a call of The output stream returned from a call of
`FileSystem.append(path, buffersize, progress)` within a filesystem `FS`, `FileSystem.append(path, buffersize, progress)` within a filesystem `FS`,
can be modelled as a stream whose `buffer` is intialized to that of can be modelled as a stream whose `buffer` is initialized to that of
the original file: the original file:
```python ```python
@ -184,7 +184,7 @@ This document covers the requirements of such implementations.
HDFS's `FileSystem` implementation, `DistributedFileSystem`, returns an instance HDFS's `FileSystem` implementation, `DistributedFileSystem`, returns an instance
of `HdfsDataOutputStream`. This implementation has at least two behaviors of `HdfsDataOutputStream`. This implementation has at least two behaviors
which are not explicitly declared by the base Java implmentation which are not explicitly declared by the base Java implementation
1. Writes are synchronized: more than one thread can write to the same 1. Writes are synchronized: more than one thread can write to the same
output stream. This is a use pattern which HBase relies on. output stream. This is a use pattern which HBase relies on.
@ -648,7 +648,7 @@ in production.
1. `OutputStream.write()` MAY persist the data, synchronously or asynchronously 1. `OutputStream.write()` MAY persist the data, synchronously or asynchronously
1. `OutputStream.flush()` flushes data to the destination. There 1. `OutputStream.flush()` flushes data to the destination. There
are no strict persistence requirements. are no strict persistence requirements.
1. `Syncable.hflush()` synchronously sends all outstaning data to the destination 1. `Syncable.hflush()` synchronously sends all outstanding data to the destination
filesystem. After returning to the caller, the data MUST be visible to other readers, filesystem. After returning to the caller, the data MUST be visible to other readers,
it MAY be durable. That is: it does not have to be persisted, merely guaranteed it MAY be durable. That is: it does not have to be persisted, merely guaranteed
to be consistently visible to all clients attempting to open a new stream reading to be consistently visible to all clients attempting to open a new stream reading
@ -678,7 +678,7 @@ public void hflush() throws IOException {
``` ```
This is perfectly acceptable as an implementation: the semantics of `hflush()` This is perfectly acceptable as an implementation: the semantics of `hflush()`
are satisifed. are satisfied.
What is not acceptable is downgrading `hsync()` to `hflush()`, as the durability guarantee is no longer met. What is not acceptable is downgrading `hsync()` to `hflush()`, as the durability guarantee is no longer met.
@ -863,7 +863,7 @@ local data as can be written to full checksummed blocks of data.
That is, the hsync/hflush operations are not guaranteed to write all the pending That is, the hsync/hflush operations are not guaranteed to write all the pending
data until the file is finally closed. data until the file is finally closed.
For this reason, the local fileystem accessed via `file://` URLs For this reason, the local filesystem accessed via `file://` URLs
does not support `Syncable` unless `setWriteChecksum(false)` was does not support `Syncable` unless `setWriteChecksum(false)` was
called on that FileSystem instance so as to disable checksum creation. called on that FileSystem instance so as to disable checksum creation.
After which, obviously, checksums are not generated for any file. After which, obviously, checksums are not generated for any file.