[
https://issues.apache.org/jira/browse/HDFS-14355?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16792605#comment-16792605
]
Rakesh R commented on HDFS-14355:
-
Thanks [~PhiloHe] for the incremental patch. Following are few quick comments,
I will continue reviewing the patch.
# Please rename configs: {{dfs.datanode.cache.loader.impl.classname}} to =>
{{dfs.datanode.cache.loader.class}}
{{DFS_DATANODE_CACHE_LOADER_IMPL_CLASSNAME}} to =>
{{DFS_DATANODE_CACHE_LOADER_CLASS}}
{{DFS_DATANODE_CACHE_LOADER_IMPL_CLASSNAME_DEFAULT}} to =>
{{DFS_DATANODE_CACHE_LOADER_CLASS_DEFAULT}}
# Replace the config reading logic like below. Also, this would help avoiding
if-else checks : {{if
(cacheLoader.equals(MemoryMappableBlockLoader.class.getSimpleName()))}} to
determine which class is configured by the user.
{code:java}
DFSConfigKeys.java
public static final String DFS_DATANODE_CACHE_LOADER_CLASS =
"dfs.datanode.cache.loader.class";
public static final Class
DFS_DATANODE_CACHE_LOADER_CLASS_DEFAULT = MemoryMappableBlockLoader.class;
You can use the following way to instantiate the cache loader.
..
..
this.cacheLoader = getConf().getClass(
DFSConfigKeys.DFS_DATANODE_CACHE_LOADER_CLASS,
DFSConfigKeys.DFS_DATANODE_CACHE_LOADER_CLASS_DEFAULT,
MappableBlockLoader.class);
{code}
# Add config name into the message {{"The persistent memory volumes are not
configured!"}} to => {{"The persistent memory volume, " +
DFSConfigKeys.DFS_DATANODE_CACHE_PMEM_DIR_KEY + " is not configured!"}}
# Good to Unmaps the block {{mappedoutbuffer}} before deleting the file like
below,
{code:java}
FileMappableBlockLoader.java
#verifyIfValidPmemVolume(){
...
...
if (file != null) {
IOUtils.closeQuietly(file);
NativeIO.POSIX.munmap(out);
try {
FsDatasetUtil.deleteMappedFile(testFilePath);
} catch (IOException e) {
LOG.warn("Failed to delete test file " + testFilePath +
" from persistent memory", e);
}
{code}
# FileMappableBlockLoader - Please remove the {{assert
NativeIO.isAvailable();}} check, its not needed right?
# Describe briefly about the filepath string formation pattern
'{{PmemDir/BlockPoolId-BlockId'}} either at class or function level javadocs.
{code:java}
FileMappableBlockLoader#load()
filePath = getOneLocation() + "/" + key.getBlockPoolId() +
"-" + key.getBlockId();
{code}
# Add @VisibleForTesting to {{public static void verifyIfValidPmemVolume(File
pmemDir)}} function
# Add annotation to the new classes FileMappedBlock, FileMappableBlockLoader.
{code:java}
@InterfaceAudience.Private
@InterfaceStability.Unstable
{code}
# Comments on TestCacheWithFileMappableBlockLoader:
## Remove MLOCK config, which is not required.
{code:java}
myConf.setLong(DFSConfigKeys.DFS_DATANODE_MAX_LOCKED_MEMORY_KEY,
CACHE_CAPACITY);
{code}
## Move the test TestCacheWithFileMappableBlockLoader.java class to {{package
org.apache.hadoop.hdfs.server.datanode.fsdataset.impl}}. This will avoid making
the class FsDatasetImpl public and infact no changes to FsDatasetImpl class
required.
> Implement SCM cache using pure java mapped byte buffer
> --
>
> Key: HDFS-14355
> URL: https://issues.apache.org/jira/browse/HDFS-14355
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Components: caching, datanode
>Reporter: Feilong He
>Assignee: Feilong He
>Priority: Major
> Attachments: HDFS-14355.000.patch, HDFS-14355.001.patch
>
>
> This task is to implement the caching to persistent memory using pure
> {{java.nio.MappedByteBuffer}}, which could be useful in case native support
> isn't available or convenient in some environments or platforms.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
-
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
