[
https://issues.apache.org/jira/browse/TEZ-3347?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15668030#comment-15668030
]
Hitesh Shah commented on TEZ-3347:
----------------------------------
I think this is probably a manisfestation of another bug where the UI does not
use the RM as the source of truth on whether the app has completed or not.
Additionally, the Tez AM probably needs a webservice to indicate the status of
a previously run DAG so that the UI can make more intelligent decisions in
situations where the AM has run multiple dags but there is a lag with ATS data.
> Vertex UI throws an error while getting vertexProgress for a killed Vertex
> --------------------------------------------------------------------------
>
> Key: TEZ-3347
> URL: https://issues.apache.org/jira/browse/TEZ-3347
> Project: Apache Tez
> Issue Type: Bug
> Components: UI
> Reporter: Kuhu Shukla
> Attachments: ErrorCodeFailedVertex.png
>
>
> Given an AM that fails all its attempts, the application fails and the very
> first click on the killed/failed vertex throws the following error:
> {code}
> error code: Unknown, message: expected expression, got '<'
> {code}
> It self corrects if tried again immediately after the failure.
> This is because the RM proxy redirects the call to the AHS server and the
> REST call is malformed for that server. Upon inspection of the responses, it
> was seen that the URL looked something like this:
> {code}
> http://<hostname>:<ahsport>/applicationhistory/app/application_123_456/ws/v1/tez/vertexProgress?dagID=1&vertexID=01&_=123
> {code}
> which is not a proper Rest call on the AHS.
> I think the following code can cause this issue:
> {code}
> // Load progress in parallel for v1 version of the api
> _loadProgress: function (vertices) {
> var that = this,
> runningVerticesIdx = vertices
> .filterBy('status', 'RUNNING')
> .map(function(item) {
> return item.get('id').split('_').splice(-1).pop();
> });
> if (runningVerticesIdx.length > 0) {
> this.store.unloadAll('vertexProgress');
> this.store.findQuery('vertexProgress', {
> metadata: {
> appId: that.get('applicationId'),
> dagIdx: that.get('idx'),
> vertexIds: runningVerticesIdx.join(',')
> }
> }).then(function(vertexProgressInfo) {
> App.Helpers.emData.mergeRecords(
> that.get('rowsDisplayed'),
> vertexProgressInfo,
> ['progress']
> );
> }).catch(function(error) {
> error.message = "Failed to fetch vertexProgress. Application Master
> (AM) is out of reach. Either it's down, or CORS is not enabled for YARN
> ResourceManager.";
> Em.Logger.error(error);
> var err = App.Helpers.misc.formatError(error);
> var msg = 'Error code: %@, message: %@'.fmt(err.errCode, err.msg);
> App.Helpers.ErrorBar.getInstance().show(msg, err.details);
> });
> {code}
> which uses AMInfo that gets the response based on what loadApp method finds:
> {code}
> loadApp: function (store, appId, useCache) {
> if(!useCache) {
> App.Helpers.misc.removeRecord(store, 'appDetail', appId);
> App.Helpers.misc.removeRecord(store, 'clusterApp', appId);
> }
> return store.find('clusterApp', appId).catch(function () {
> return store.find('appDetail', appId);
> }).catch(function (error) {
> error.message = "Couldn't get details of application %@. RM is not
> reachable, and history service is not enabled.".fmt(appId);
> throw error;
> });
> }
> {code}
> We can check here in the catch block if the response type is not JSON or not
> try and get vertexProgress since it knows that the application/AM has failed.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
