[
https://issues.apache.org/jira/browse/HDFS-15025?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17203240#comment-17203240
]
YaYun Wang edited comment on HDFS-15025 at 9/29/20, 1:08 AM:
-------------------------------------------------------------
[~liuml07]，[~ayushtkn]， we have tried to make sure that {{NVDIMM}} and
{{isRAM}} will not affect the function and existing data on hadoop by
flollowing ways:
# Running test case: run test cases on
{{FsEditLogOp#SetQuotaByStorageTypeOp}}, such as {{TestWebHDFS}}, and that can
pass without failures or errors.
# Combing the code about {{FsEditLogOp＃SetQuotaByStorageTypeOp}}, including
classes of {{DistributedFileSystem}}, {{DFSClient}}, {{NamenodeRPCServer}},
{{FSNamesystem}}, {{FSDirAttrOp}}, {{FSEditLog}} and {{FSEditLogOp,}} etc.. We
can draw a conclusion that {{ordinal()}} of StorageType which is enum is
flexible and compatible, that is, {{ordinal()}} can vary with the value of
{{StorageType}} and that's different with
{{TestRouterQuota.testStorageTypeQuota}} where {{ordinal()}} is used as the
index of array.
{code:java}
public void testStorageTypeQuota() throws Exception {
...
//the first parameters is a array which length is 5 for 5 storage types.
verifyTypeQuotaAndConsume(new long[] {-1, -1, ssQuota * 2, -1, -1}, null,
usage);
...
}
private void verifyTypeQuotaAndConsume(long[] quota, long[] consume,
QuotaUsage usage) {
for (StorageType t : StorageType.values()) {
if (quota != null) {
assertEquals(quota[t.ordinal()], usage.getTypeQuota(t)); //ordinal() is
the index of quota
}
if (consume != null) {
assertEquals(consume[t.ordinal()], usage.getTypeConsumed(t));
}
}
}{code}
# Verifing the function of hadoop with {{NVDIMM}} and {{isRAM}} : the
functions of DFSIO, wordcount, put/get built-in hadoop can used normally. The
old storage types and storage policies are normal too after upgrade.
# Verifing the function after upgrade by existing data: first, we write data
to hadoop without the patch using DFSIO, wordcount and put/get. Then update the
version of hadoop with the patch. After that, the existing data of old version
can be accessed and used normally.
was (Author: wangyayun):
[~liuml07]，[~ayushtkn]， we have tried to make sure that {{NVDIMM}} and
{{isRAM}} will not affect the function and existing data on hadoop by
flollowing ways:
# Running test case: run test cases on
{{FsEditLogOp#SetQuotaByStorageTypeOp}}, such as {{TestWebHDFS}}, and that can
pass without failures or errors.
# Combing the code about {{FsEditLogOp＃SetQuotaByStorageTypeOp}}, including
{{DistributedFileSystem}}, {{DFSClient}}, {{NamenodeRPCServer}},
{{FSNamesystem}}, {{FSDirAttrOp}}, {{FSEditLog}} and {{FSEditLogOp,}} etc.. We
can draw a conclusion that {{ordinal()}} of StorageType which is enum is
flexible and compatible, that is, {{ordinal()}} can vary with the value of
{{StorageType}} and that's different with
{{TestRouterQuota.testStorageTypeQuota}} where {{ordinal()}} is used as the
index of array.
{code:java}
public void testStorageTypeQuota() throws Exception {
...
//the first parameters is a array which length is 5 for 5 storage types.
verifyTypeQuotaAndConsume(new long[] {-1, -1, ssQuota * 2, -1, -1}, null,
usage);
...
}
private void verifyTypeQuotaAndConsume(long[] quota, long[] consume,
QuotaUsage usage) {
for (StorageType t : StorageType.values()) {
if (quota != null) {
assertEquals(quota[t.ordinal()], usage.getTypeQuota(t)); //ordinal() is
the index of quota
}
if (consume != null) {
assertEquals(consume[t.ordinal()], usage.getTypeConsumed(t));
}
}
}{code}
# Verifing the function of hadoop with {{NVDIMM}} and {{isRAM}} : the
functions of DFSIO, wordcount, put/get built-in hadoop can used normally. The
old storage types and storage policies are normal too after upgrade.
# Verifing the function after upgrade by existing data: first, we write data
to hadoop without the patch using DFSIO, wordcount and put/get. Then updata the
version of hadoop with the patch. After that, the existing data of old version
can be accessed and used normally.
> Applying NVDIMM storage media to HDFS
> -------------------------------------
>
> Key: HDFS-15025
> URL: https://issues.apache.org/jira/browse/HDFS-15025
> Project: Hadoop HDFS
> Issue Type: New Feature
> Components: datanode, hdfs
> Reporter: YaYun Wang
> Assignee: YaYun Wang
> Priority: Major
> Labels: pull-request-available
> Fix For: 3.4.0
>
> Attachments: Applying NVDIMM to HDFS.pdf, HDFS-15025.001.patch,
> HDFS-15025.002.patch, HDFS-15025.003.patch, HDFS-15025.004.patch,
> HDFS-15025.005.patch, HDFS-15025.006.patch, NVDIMM_patch(WIP).patch
>
> Time Spent: 12h 10m
> Remaining Estimate: 0h
>
> The non-volatile memory NVDIMM is faster than SSD, it can be used
> simultaneously with RAM, DISK, SSD. The data of HDFS stored directly on
> NVDIMM can not only improves the response rate of HDFS, but also ensure the
> reliability of the data.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
