hadoop-mapreduce-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jason Lowe (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (MAPREDUCE-7133) History Server task attempts REST API returns invalid data
Date Wed, 12 Sep 2018 22:26:00 GMT

    [ https://issues.apache.org/jira/browse/MAPREDUCE-7133?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16612812#comment-16612812

Jason Lowe commented on MAPREDUCE-7133:

The problem with doing too much code cleanup in a fix that may need to be backported to many
older releases is that it unnecessarily complicates and adds risk to the backports.  If this
applies cleanly all the way back to 2.7.8 then it may not be a big deal in this particular
case, but in general it's not a good idea to do a lot of code refactoring that's not required
to fix the problem.  Because of the unnecessary refactoring, this patch changes 13 files when
the fix only requires modification to 3.  If the patch applies cleanly to all places it should
go then I'm OK with leaving the refactoring in, otherwise IMHO the cleanup should be filed
as a separate improvement JIRA that goes into trunk while the targeted fix goes into all the
various active branches that need it.

I'm fine with leaving TaskAttemptInfo an abstract class.

bq. I agree that using XMLElementRef annotations can be unobvious.

To me that statement warrants the comment, and adding a comment also aligns with the HowToContribute
wiki which mentions, "comment code whose function or rationale is not obvious".  Note that
the comment doesn't have to be a treatise, rather just a one-liner why it's used instead of
the typical FIELD annotations used elsewhere.  If it's too hard to concisely explain then
we can reference this analysis here, e.g.: "Use of XMLElementRef instead of XmlAccessorType
is critical, see https://issues.apache.org/jira/browse/MAPREDUCE-7133"  A small comment next
to the code in question makes it much easier for someone new to the code to come up to speed
than requiring them to dig through the project's version control.

> History Server task attempts REST API returns invalid data
> ----------------------------------------------------------
>                 Key: MAPREDUCE-7133
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-7133
>             Project: Hadoop Map/Reduce
>          Issue Type: Bug
>          Components: jobhistoryserver
>            Reporter: Oleksandr Shevchenko
>            Assignee: Oleksandr Shevchenko
>            Priority: Major
>         Attachments: MAPREDUCE-7133.001.patch, MAPREDUCE-7133.002.patch, MAPREDUCE-7133.003.patch,
> When we send a request to History Server with headers : Accept: application/json [https://nodename:19888/ws/v1/history/mapreduce/jobs/job_1535363926925_0040/tasks/task_1535363926925_0040_r_000003/attempts|] 
> we get the following JSON:
> {code:java}
> {
> "taskAttempts": {
> "taskAttempt": [{
> "type": "reduceTaskAttemptInfo",
> "startTime": 1535372984638,
> "finishTime": 1535372986149,
> "elapsedTime": 1511,
> "progress": 100.0,
> "id": "attempt_1535363926925_0040_r_000003_0",
> "rack": "/default-rack",
> "state": "SUCCEEDED",
> "status": "reduce > reduce",
> "nodeHttpAddress": "node2.cluster.com:8044",
> "diagnostics": "",
> "type": "REDUCE",
> "assignedContainerId": "container_e01_1535363926925_0040_01_000006",
> "shuffleFinishTime": 1535372986056,
> "mergeFinishTime": 1535372986075,
> "elapsedShuffleTime": 1418,
> "elapsedMergeTime": 19,
> "elapsedReduceTime": 74
> }]
> }
> }
> {code}
> As you can see "type" property has duplicates:
> "type": "reduceTaskAttemptInfo"
> "type": "REDUCE"
> It's lead to an error during parsing response body as JSON is not valid.
> When we use application/xml we get the following response:
> {code:java}
> <taskAttempts>
> <taskAttempt xmlns:xsi="[http://www.w3.org/2001/XMLSchema-instance]" xsi:type="reduceTaskAttemptInfo"><startTime>1535372984638</startTime><finishTime>1535372986149</finishTime><elapsedTime>1511</elapsedTime><progress>100.0</progress><id>attempt_1535363926925_0040_r_000003_0</id><rack>/default-rack</rack><state>SUCCEEDED</state><status>reduce
> reduce</status><nodeHttpAddress>[node2.cluster.com:8044|http://node2.cluster.com:8044]</nodeHttpAddress><diagnostics/><type>REDUCE</type><assignedContainerId>container_e01_1535363926925_0040_01_000006</assignedContainerId><shuffleFinishTime>1535372986056</shuffleFinishTime><mergeFinishTime>1535372986075</mergeFinishTime><elapsedShuffleTime>1418</elapsedShuffleTime><elapsedMergeTime>19</elapsedMergeTime><elapsedReduceTime>74</elapsedReduceTime></taskAttempt>
> </taskAttempts>
> {code}
> Take a look at the following string:
> {code:java}
> <taskAttempt xmlns:xsi="[http://www.w3.org/2001/XMLSchema-instance]" xsi:type="reduceTaskAttemptInfo">
> {code}
> We got "xsi:type" attribute which incorectly marshall later to duplicated field if we
use JSON format.
> It acceptable only to REDUCE task. For MAP task we get xml without "xsi:type" attribute.
> {code:java}
> <taskAttempts>
> <taskAttempt>
> <startTime>1535370756528</startTime>
> <finishTime>1535370760318</finishTime>
> <elapsedTime>3790</elapsedTime>
> <progress>100.0</progress>
> <id>attempt_1535363926925_0029_m_000001_0</id>
> <rack>/default-rack</rack>
> <state>SUCCEEDED</state>
> <status>map > sort</status>
> <nodeHttpAddress>[node2.cluster.com:8044|http://node2.cluster.com:8044]</nodeHttpAddress>
> <diagnostics/>
> <type>MAP</type>
> <assignedContainerId>container_e01_1535363926925_0029_01_000003</assignedContainerId>
> </taskAttempt>
> </taskAttempts>
> {code}
> This happens since we have two different hierarchical classes for MAP ->TaskAttemptInfo
and REDUCE- > ReduceTaskAttemptInfo tasks.
> ReduceTaskAttemptInfo extends TaskAttemptInfo, later we marshal all tasks (map and reduce)
by TaskAttemptsInfo.getTaskAttempt(). In this place, we do not have any information about ReduceTaskAttemptInfo
type as we store all tasks in ArrayList<TaskAttemptInfo>. 
> During marshaling we see that actual type of task ReduceTaskAttemptInfo instead of TaskAttemptsInfo
and add meta information for this. That's why we get duplicated fields.
> Unfortunately we do not catch it before in TestHsWebServicesAttempts since we use 
> org.codehaus.jettison.json.JSONObject library which overrides duplicated fields. Even
when we use Postman to do request we get valid JSON. Only when we change represent type to
Raw we can notice this issue. Also, we able to reproduce this bug by using "org.json:json"
> Something like this:
> {code:java}
> BufferedReader inReader = new BufferedReader( new InputStreamReader(connection.getInputStream()
) );
>  String inputLine;
>  StringBuilder response = new StringBuilder();
> while ( (inputLine = inReader.readLine()) != null ) {
>  response.append(inputLine);
>  }
> inReader.close();
> JSONObject o = new JSONObject(response.toString());
> {code}

This message was sent by Atlassian JIRA

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

View raw message