You are viewing a plain text version of this content. The canonical link for it is here.
Posted to yarn-issues@hadoop.apache.org by "aimahou (Jira)" <ji...@apache.org> on 2020/06/01 09:46:00 UTC

[jira] [Updated] (YARN-10298) TimeLine entity information only stored in one region when use apache HBase as backend storage

     [ https://issues.apache.org/jira/browse/YARN-10298?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

aimahou updated YARN-10298:
---------------------------
    Description: 
h2. Issue

TimeLine entity information only stored in one region when use apache HBase as backend storage
h2. Probable cause

We found in the source code that the rowKey is composed of clusterId、userId、flowName、flowRunId and appId when hbase timeline writer stores timeline entity info,which probably cause the rowKey is sorted by dictionary order. Thus timeline entity may only store in one region or few adjacent regions.
h2. Related code snippet

HBaseTimelineWriterImpl.java
{quote}public TimelineWriteResponse write(TimelineCollectorContext context,
 TimelineEntities data, UserGroupInformation callerUgi)
 throws IOException {

...

boolean isApplication = ApplicationEntity.isApplicationEntity(te);
 byte[] rowKey;
 if (isApplication)

{

ApplicationRowKey applicationRowKey = new ApplicationRowKey(clusterId, userId, flowName, flowRunId, appId);

rowKey = applicationRowKey.getRowKey();

store(rowKey, te, flowVersion, Tables.APPLICATION_TABLE);

}

else

{

EntityRowKey entityRowKey = new EntityRowKey(clusterId, userId, flowName, flowRunId, appId, te.getType(), te.getIdPrefix(), te.getId());

rowKey = entityRowKey.getRowKey();

store(rowKey, te, flowVersion, Tables.ENTITY_TABLE); }

if (!isApplication && SubApplicationEntity.isSubApplicationEntity(te))

{

SubApplicationRowKey subApplicationRowKey = new SubApplicationRowKey(subApplicationUser, clusterId, te.getType(), te.getIdPrefix(), te.getId(), userId);

rowKey = subApplicationRowKey.getRowKey();

store(rowKey, te, flowVersion, Tables.SUBAPPLICATION_TABLE);

}

...

}
{quote}
h2. Suggestion

We can use the hash code of original rowKey as the rowKey to store and read timeline entity data.

  was:
h2. Issue

TimeLine entity information only stored in one region when use apache HBase as backend storage
h2. Probable cause

We found in the source code that the rowKey is composed of clusterId、userId、flowName、flowRunId and appId when hbase timeline writer stores timeline entity info,which probably cause the rowKey is sorted by dictionary order. Thus timeline entity may only store in one region or few adjacent regions.
h2. Related code snippet

HBaseTimelineWriterImpl.java

public TimelineWriteResponse write(TimelineCollectorContext context,
 TimelineEntities data, UserGroupInformation callerUgi)
 throws IOException {

...

boolean isApplication = ApplicationEntity.isApplicationEntity(te);
byte[] rowKey;
if (isApplication) {
 ApplicationRowKey applicationRowKey =
 new ApplicationRowKey(clusterId, userId, flowName, flowRunId,
 appId);
 rowKey = applicationRowKey.getRowKey();
 store(rowKey, te, flowVersion, Tables.APPLICATION_TABLE);
} else {
 EntityRowKey entityRowKey =
 new EntityRowKey(clusterId, userId, flowName, flowRunId, appId,
 te.getType(), te.getIdPrefix(), te.getId());
 rowKey = entityRowKey.getRowKey();
 store(rowKey, te, flowVersion, Tables.ENTITY_TABLE);
}

if (!isApplication && SubApplicationEntity.isSubApplicationEntity(te)) {
 SubApplicationRowKey subApplicationRowKey =
 new SubApplicationRowKey(subApplicationUser, clusterId,
 te.getType(), te.getIdPrefix(), te.getId(), userId);
 rowKey = subApplicationRowKey.getRowKey();
 store(rowKey, te, flowVersion, Tables.SUBAPPLICATION_TABLE);
}

...

}
h2. Suggestion

We can use the hash code of original rowKey as the rowKey to store and read timeline entity data.


> TimeLine entity information only stored in one region when use apache HBase as backend storage
> ----------------------------------------------------------------------------------------------
>
>                 Key: YARN-10298
>                 URL: https://issues.apache.org/jira/browse/YARN-10298
>             Project: Hadoop YARN
>          Issue Type: Improvement
>          Components: ATSv2, timelineservice
>    Affects Versions: 3.1.1
>            Reporter: aimahou
>            Priority: Major
>
> h2. Issue
> TimeLine entity information only stored in one region when use apache HBase as backend storage
> h2. Probable cause
> We found in the source code that the rowKey is composed of clusterId、userId、flowName、flowRunId and appId when hbase timeline writer stores timeline entity info,which probably cause the rowKey is sorted by dictionary order. Thus timeline entity may only store in one region or few adjacent regions.
> h2. Related code snippet
> HBaseTimelineWriterImpl.java
> {quote}public TimelineWriteResponse write(TimelineCollectorContext context,
>  TimelineEntities data, UserGroupInformation callerUgi)
>  throws IOException {
> ...
> boolean isApplication = ApplicationEntity.isApplicationEntity(te);
>  byte[] rowKey;
>  if (isApplication)
> {
> ApplicationRowKey applicationRowKey = new ApplicationRowKey(clusterId, userId, flowName, flowRunId, appId);
> rowKey = applicationRowKey.getRowKey();
> store(rowKey, te, flowVersion, Tables.APPLICATION_TABLE);
> }
> else
> {
> EntityRowKey entityRowKey = new EntityRowKey(clusterId, userId, flowName, flowRunId, appId, te.getType(), te.getIdPrefix(), te.getId());
> rowKey = entityRowKey.getRowKey();
> store(rowKey, te, flowVersion, Tables.ENTITY_TABLE); }
> if (!isApplication && SubApplicationEntity.isSubApplicationEntity(te))
> {
> SubApplicationRowKey subApplicationRowKey = new SubApplicationRowKey(subApplicationUser, clusterId, te.getType(), te.getIdPrefix(), te.getId(), userId);
> rowKey = subApplicationRowKey.getRowKey();
> store(rowKey, te, flowVersion, Tables.SUBAPPLICATION_TABLE);
> }
> ...
> }
> {quote}
> h2. Suggestion
> We can use the hash code of original rowKey as the rowKey to store and read timeline entity data.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: yarn-issues-help@hadoop.apache.org