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

HDFS-14111. hdfsOpenFile on HDFS causes unnecessary IO from file offset 0. Contributed by Sahil Takiar.

Browse Source 
Signed-off-by: Wei-Chiu Chuang <weichiu@apache.org>
This commit is contained in:
Sahil Takiar 2019-03-06 15:04:06 -08:00 committed by Wei-Chiu Chuang
parent 6192c1fe3b
commit 618e009ac0
5 changed files with 30 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/crypto/CryptoInputStream.java
View File

@ -740,6 +740,7 @@ public boolean hasCapability(String capability) {
case StreamCapabilities.READAHEAD:
case StreamCapabilities.DROPBEHIND:
case StreamCapabilities.UNBUFFER:
case StreamCapabilities.READBYTEBUFFER:
return true;
default:
return false;

6
hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/StreamCapabilities.java
View File

@ -59,6 +59,12 @@ public interface StreamCapabilities {
*/
String UNBUFFER = "in:unbuffer";
/**
* Stream read(ByteBuffer) capability implemented by
* {@link ByteBufferReadable#read(java.nio.ByteBuffer)}.
*/
String READBYTEBUFFER = "in:readbytebuffer";
/**
* Capabilities that a stream can support and be queried for.
*/

1
hadoop-hdfs-project/hadoop-hdfs-client/src/main/java/org/apache/hadoop/hdfs/DFSInputStream.java
View File

@ -1779,6 +1779,7 @@ public boolean hasCapability(String capability) {
case StreamCapabilities.READAHEAD:
case StreamCapabilities.DROPBEHIND:
case StreamCapabilities.UNBUFFER:
case StreamCapabilities.READBYTEBUFFER:
return true;
default:
return false;

29
hadoop-hdfs-project/hadoop-hdfs-native-client/src/main/native/libhdfs/hdfs.c
View File

@ -1013,7 +1013,7 @@ static hdfsFile hdfsOpenFileImpl(hdfsFS fs, const char *path, int flags,
return f{is|os};
*/
int accmode = flags & O_ACCMODE;
jstring jStrBufferSize = NULL, jStrReplication = NULL;
jstring jStrBufferSize = NULL, jStrReplication = NULL, jCapabilityString = NULL;
jobject jConfiguration = NULL, jPath = NULL, jFile = NULL;
jobject jFS = (jobject)fs;
jthrowable jthr;
@ -1171,16 +1171,22 @@ static hdfsFile hdfsOpenFileImpl(hdfsFS fs, const char *path, int flags,
file->flags = 0;
if ((flags & O_WRONLY) == 0) {
// Try a test read to see if we can do direct reads
char buf;
if (readDirect(fs, file, &buf, 0) == 0) {
// Success - 0-byte read should return 0
// Check the StreamCapabilities of jFile to see if we can do direct reads
jthr = newJavaStr(env, "in:readbytebuffer", &jCapabilityString);
if (jthr) {
ret = printExceptionAndFree(env, jthr, PRINT_EXC_ALL,
"hdfsOpenFile(%s): newJavaStr", path);
goto done;
}
jthr = invokeMethod(env, &jVal, INSTANCE, jFile, HADOOP_ISTRM,
"hasCapability", "(Ljava/lang/String;)Z", jCapabilityString);
if (jthr) {
ret = printExceptionAndFree(env, jthr, PRINT_EXC_ALL,
"hdfsOpenFile(%s): FSDataInputStream#hasCapability", path);
goto done;
}
if (jVal.z) {
file->flags |= HDFS_FILE_SUPPORTS_DIRECT_READ;
} else if (errno != ENOTSUP) {
// Unexpected error. Clear it, don't set the direct flag.
fprintf(stderr,
"hdfsOpenFile(%s): WARN: Unexpected error %d when testing "
"for direct read compatibility\n", path, errno);
}
}
ret = 0;
@ -1190,7 +1196,8 @@ done:
destroyLocalReference(env, jStrReplication);
destroyLocalReference(env, jConfiguration);
destroyLocalReference(env, jPath);
destroyLocalReference(env, jFile);
destroyLocalReference(env, jFile);
destroyLocalReference(env, jCapabilityString);
if (ret) {
if (file) {
if (file->file) {

5
hadoop-hdfs-project/hadoop-hdfs-native-client/src/main/native/libhdfspp/tests/hdfs_ext_test.cc
View File

@ -503,7 +503,10 @@ TEST_F(HdfsExtTest, TestReadStats) {
hdfsFileFreeReadStatistics(stats);
EXPECT_EQ(0, hdfsCloseFile(fs, file));
EXPECT_EQ(0, errno);
// Since libhdfs is not guaranteed to set errno to 0 on successful
// operations, we disable this check for now, see HDFS-14325 for a
// long term solution to this problem
// EXPECT_EQ(0, errno);
}
//Testing working directory