[jira] [Commented] (SPARK-19624) --conf spark.app.name=test is not working with spark-shell/pyspark
[ https://issues.apache.org/jira/browse/SPARK-19624?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15874027#comment-15874027 ] Sachin Aggarwal commented on SPARK-19624: - [~srowen] I agree that --name is working but we should not ignore --conf spark.app.name="test" > --conf spark.app.name=test is not working with spark-shell/pyspark > -- > > Key: SPARK-19624 > URL: https://issues.apache.org/jira/browse/SPARK-19624 > Project: Spark > Issue Type: Bug > Components: Spark Core, Spark Shell, Spark Submit >Affects Versions: 1.6.0, 2.0.0, 2.1.0 >Reporter: Sachin Aggarwal >Priority: Minor > > On starting a spark-shell or pyshark shell if we pass --conf > spark.app.name=test it is not working as --name "Spark shell" takes > precedence over --conf > line refrence for spark-shell > https://github.com/apache/spark/blob/master/bin/spark-shell#L53 > similarly line refrence for pyspark > https://github.com/apache/spark/blob/master/bin/pyspark#L77 -- This message was sent by Atlassian JIRA (v6.3.15#6346) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Created] (SPARK-19624) --conf spark.app.name=test is not working with spark-shell/pyspark
Sachin Aggarwal created SPARK-19624: --- Summary: --conf spark.app.name=test is not working with spark-shell/pyspark Key: SPARK-19624 URL: https://issues.apache.org/jira/browse/SPARK-19624 Project: Spark Issue Type: Bug Components: Spark Core, Spark Shell, Spark Submit Affects Versions: 2.1.0, 2.0.0, 1.6.0 Reporter: Sachin Aggarwal Priority: Minor On starting a spark-shell or pyshark shell if we pass --conf spark.app.name=test it is not working as --name "Spark shell" takes precedence over --conf line refrence for spark-shell https://github.com/apache/spark/blob/master/bin/spark-shell#L53 similarly line refrence for pyspark https://github.com/apache/spark/blob/master/bin/pyspark#L77 -- This message was sent by Atlassian JIRA (v6.3.15#6346) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Closed] (SPARK-15183) Adding outputMode to structure Streaming Experimental Api
[ https://issues.apache.org/jira/browse/SPARK-15183?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Sachin Aggarwal closed SPARK-15183. --- Resolution: Duplicate > Adding outputMode to structure Streaming Experimental Api > - > > Key: SPARK-15183 > URL: https://issues.apache.org/jira/browse/SPARK-15183 > Project: Spark > Issue Type: Improvement > Components: SQL, Streaming >Reporter: Sachin Aggarwal >Priority: Trivial > > while experimenting with structure streaming. I found that mode() is used for > non-continuous queries while outputMode() is used for continuous queries. > ouputMode is not defined, so I have written the some raw implementation and > test cases just to make sure the streaming app works > Note:- > /** Start a query */ > private[sql] def startQuery( > name: String, > checkpointLocation: String, > df: DataFrame, > sink: Sink, > trigger: Trigger = ProcessingTime(0), > triggerClock: Clock = new SystemClock(), > outputMode: OutputMode = Append): ContinuousQuery = { > As per me outputMode should be defined before triggerClock, the constructor > with outputMode defined will be used more often then triggerClock. > I have added triggerClock() method also -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Updated] (SPARK-15183) Adding outputMode to structure Streaming Experimental Api
[ https://issues.apache.org/jira/browse/SPARK-15183?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Sachin Aggarwal updated SPARK-15183: Summary: Adding outputMode to structure Streaming Experimental Api (was: adding outputMode to structure Streaming Experimental Api) > Adding outputMode to structure Streaming Experimental Api > - > > Key: SPARK-15183 > URL: https://issues.apache.org/jira/browse/SPARK-15183 > Project: Spark > Issue Type: Improvement > Components: SQL, Streaming >Reporter: Sachin Aggarwal >Priority: Trivial > > while experimenting with structure streaming. I found that mode() is used for > non-continuous queries while outputMode() is used for continuous queries. > ouputMode is not defined, so I have written the some raw implementation and > test cases just to make sure the streaming app works > Note:- > /** Start a query */ > private[sql] def startQuery( > name: String, > checkpointLocation: String, > df: DataFrame, > sink: Sink, > trigger: Trigger = ProcessingTime(0), > triggerClock: Clock = new SystemClock(), > outputMode: OutputMode = Append): ContinuousQuery = { > As per me outputMode should be defined before triggerClock, the constructor > with outputMode defined will be used more often then triggerClock. > I have added triggerClock() method also -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Created] (SPARK-15183) adding outputMode to structure Streaming Experimental Api
Sachin Aggarwal created SPARK-15183: --- Summary: adding outputMode to structure Streaming Experimental Api Key: SPARK-15183 URL: https://issues.apache.org/jira/browse/SPARK-15183 Project: Spark Issue Type: Improvement Components: SQL, Streaming Reporter: Sachin Aggarwal Priority: Trivial while experimenting with structure streaming. I found that mode() is used for non-continuous queries while outputMode() is used for continuous queries. ouputMode is not defined, so I have written the some raw implementation and test cases just to make sure the streaming app works Note:- /** Start a query */ private[sql] def startQuery( name: String, checkpointLocation: String, df: DataFrame, sink: Sink, trigger: Trigger = ProcessingTime(0), triggerClock: Clock = new SystemClock(), outputMode: OutputMode = Append): ContinuousQuery = { As per me outputMode should be defined before triggerClock, the constructor with outputMode defined will be used more often then triggerClock. I have added triggerClock() method also -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-15183) adding outputMode to structure Streaming Experimental Api
[ https://issues.apache.org/jira/browse/SPARK-15183?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15274241#comment-15274241 ] Sachin Aggarwal commented on SPARK-15183: - please mark closed if not qualifies for jira > adding outputMode to structure Streaming Experimental Api > - > > Key: SPARK-15183 > URL: https://issues.apache.org/jira/browse/SPARK-15183 > Project: Spark > Issue Type: Improvement > Components: SQL, Streaming >Reporter: Sachin Aggarwal >Priority: Trivial > > while experimenting with structure streaming. I found that mode() is used for > non-continuous queries while outputMode() is used for continuous queries. > ouputMode is not defined, so I have written the some raw implementation and > test cases just to make sure the streaming app works > Note:- > /** Start a query */ > private[sql] def startQuery( > name: String, > checkpointLocation: String, > df: DataFrame, > sink: Sink, > trigger: Trigger = ProcessingTime(0), > triggerClock: Clock = new SystemClock(), > outputMode: OutputMode = Append): ContinuousQuery = { > As per me outputMode should be defined before triggerClock, the constructor > with outputMode defined will be used more often then triggerClock. > I have added triggerClock() method also -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Comment Edited] (SPARK-15183) adding outputMode to structure Streaming Experimental Api
[ https://issues.apache.org/jira/browse/SPARK-15183?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15274241#comment-15274241 ] Sachin Aggarwal edited comment on SPARK-15183 at 5/6/16 3:45 PM: - please mark closed if not qualifies as jira was (Author: sachin aggarwal): please mark closed if not qualifies for jira > adding outputMode to structure Streaming Experimental Api > - > > Key: SPARK-15183 > URL: https://issues.apache.org/jira/browse/SPARK-15183 > Project: Spark > Issue Type: Improvement > Components: SQL, Streaming >Reporter: Sachin Aggarwal >Priority: Trivial > > while experimenting with structure streaming. I found that mode() is used for > non-continuous queries while outputMode() is used for continuous queries. > ouputMode is not defined, so I have written the some raw implementation and > test cases just to make sure the streaming app works > Note:- > /** Start a query */ > private[sql] def startQuery( > name: String, > checkpointLocation: String, > df: DataFrame, > sink: Sink, > trigger: Trigger = ProcessingTime(0), > triggerClock: Clock = new SystemClock(), > outputMode: OutputMode = Append): ContinuousQuery = { > As per me outputMode should be defined before triggerClock, the constructor > with outputMode defined will be used more often then triggerClock. > I have added triggerClock() method also -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Comment Edited] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15261802#comment-15261802 ] Sachin Aggarwal edited comment on SPARK-14597 at 5/2/16 11:50 AM: -- Hi [~prashant_] I have generated few metric to support the usefulness, here are the graphs for the same. !withSortByKey.png|width=300,height=400! this graph explains that when there is sortByKey operation the jobSetCreationDelay is nearly equal to processingDelay !WithOutSortByKey.png|width=300,height=400! this graph explains that when the sortByKey operation is removed the jobSetCreationDelay is nearly equal to 10ms and there is a little increase in processingDelay. with these two graphs we can see the usefulness of JobSetCreationDelay metric, for the end user to analyze where the time is consumed. was (Author: sachin aggarwal): Hi [~prashant_] I have generated few metric to support the usefulness, here are the graphs for the same. !withSortByKey.png|width=300,height=400! this graph explains that when there is sortByKey operation the jobSetCreationDelay is nearly equal to processingDelay !WithOutSortByKey.png|width=300,height=400! this graph explains that when there is sortByKey operation is removed the jobSetCreationDelay is nearly equal to 10ms and there is a little increase in processingDelay. with these two graphs we can see the usefulness of JobSetCreationDelay metric, for the end user to analyze where the time is consumed. > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > Attachments: WithOutSortByKey.png, withSortByKey.png > > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Updated] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Sachin Aggarwal updated SPARK-14597: Attachment: withSortByKey.png > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > Attachments: WithOutSortByKey.png, withSortByKey.png > > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Comment Edited] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15261802#comment-15261802 ] Sachin Aggarwal edited comment on SPARK-14597 at 5/2/16 11:47 AM: -- Hi [~prashant_] I have generated few metric to support the usefulness, here are the graphs for the same. !withSortByKey.png|width=300,height=400! this graph explains that when there is sortByKey operation the jobSetCreationDelay is nearly equal to processingDelay !WithOutSortByKey.png|width=300,height=400! this graph explains that when there is sortByKey operation is removed the jobSetCreationDelay is nearly equal to 10ms and there is a little increase in processingDelay. with these two graphs we can see the usefulness of JobSetCreationDelay metric, for the end user to analyze where the time is consumed. was (Author: sachin aggarwal): Hi [~prashant_] I have generated few metric to support the usefulness, here are the graphs for the same. !with_sortbyKey.png|width=300,height=400! this graph explains that when there is sortByKey operation the jobSetCreationDelay is nearly equal to processingDelay !WithOutSortByKey.png|width=300,height=400! this graph explains that when there is sortByKey operation is removed the jobSetCreationDelay is nearly equal to 10ms and there is a little increase in processingDelay. with these two graphs we can see the usefulness of JobSetCreationDelay metric, for the end user to analyze where the time is consumed. > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > Attachments: WithOutSortByKey.png, withSortByKey.png > > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Updated] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Sachin Aggarwal updated SPARK-14597: Attachment: (was: with_sortbyKey.png) > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > Attachments: WithOutSortByKey.png > > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Updated] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Sachin Aggarwal updated SPARK-14597: Attachment: (was: 2.png) > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > Attachments: WithOutSortByKey.png, with_sortbyKey.png > > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Updated] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Sachin Aggarwal updated SPARK-14597: Attachment: (was: 1.png) > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > Attachments: WithOutSortByKey.png, with_sortbyKey.png > > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Issue Comment Deleted] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Sachin Aggarwal updated SPARK-14597: Comment: was deleted (was: !with_sortbyKey.png|width=300,height=400! this graph explains that when there is sortByKey operation the jobSetCreationDelay is nearly equal to processingDelay !WithOutSortByKey.png|width=300,height=400! this graph explains that when there is sortByKey operation is removed the jobSetCreationDelay is nearly equal to 10ms and there is a little increase in processingDelay.) > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > Attachments: 1.png, 2.png, WithOutSortByKey.png, with_sortbyKey.png > > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Comment Edited] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15261802#comment-15261802 ] Sachin Aggarwal edited comment on SPARK-14597 at 5/2/16 11:40 AM: -- Hi [~prashant_] I have generated few metric to support the usefulness, here are the graphs for the same. !with_sortbyKey.png|width=300,height=400! this graph explains that when there is sortByKey operation the jobSetCreationDelay is nearly equal to processingDelay !WithOutSortByKey.png|width=300,height=400! this graph explains that when there is sortByKey operation is removed the jobSetCreationDelay is nearly equal to 10ms and there is a little increase in processingDelay. with these two graphs we can see the usefulness of JobSetCreationDelay metric, for the end user to analyze where the time is consumed. was (Author: sachin aggarwal): Hi [~prashant_] I have generated few metric to support the usefulness, here are the graphs for the same. Image 1 shows the increase in jobSetGenerateTimeDelay with decrease in batchInterval Image 2 shows for a batch Interval of 100ms, how jobSetGenerateTimeDelay keeps on increasing with time. !1.png|width=300,height=400! !2.png|width=300,height=400! > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > Attachments: 1.png, 2.png, WithOutSortByKey.png, with_sortbyKey.png > > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Comment Edited] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15266297#comment-15266297 ] Sachin Aggarwal edited comment on SPARK-14597 at 5/2/16 11:33 AM: -- !with_sortbyKey.png|width=300,height=400! this graph explains that when there is sortByKey operation the jobSetCreationDelay is nearly equal to processingDelay !WithOutSortByKey.png|width=300,height=400! this graph explains that when there is sortByKey operation is removed the jobSetCreationDelay is nearly equal to 10ms and there is a little increase in processingDelay. was (Author: sachin aggarwal): !WithOutSortByKey.png|width=300,height=400! this graph explains that when there is sortByKey operation the jobSetCreationDelay is nearly equal to processingDelay !with_sortbyKey.png|width=300,height=400! this graph explains that when there is sortByKey operation is removed the jobSetCreationDelay is nearly equal to 10ms and there is a little increase in processingDelay. > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > Attachments: 1.png, 2.png, WithOutSortByKey.png, with_sortbyKey.png > > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Comment Edited] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15266297#comment-15266297 ] Sachin Aggarwal edited comment on SPARK-14597 at 5/2/16 9:50 AM: - !WithOutSortByKey.png|width=300,height=400! this graph explains that when there is sortByKey operation the jobSetCreationDelay is nearly equal to processingDelay !with_sortbyKey.png|width=300,height=400! this graph explains that when there is sortByKey operation is removed the jobSetCreationDelay is nearly equal to 10ms and there is a little increase in processingDelay. was (Author: sachin aggarwal): !WithOutSortByKey.png|! !with_sortbyKey.png|! > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > Attachments: 1.png, 2.png, WithOutSortByKey.png, with_sortbyKey.png > > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Updated] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Sachin Aggarwal updated SPARK-14597: Attachment: with_sortbyKey.png WithOutSortByKey.png !WithOutSortByKey.png|! !with_sortbyKey.png|! > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > Attachments: 1.png, 2.png, WithOutSortByKey.png, with_sortbyKey.png > > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Comment Edited] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15261802#comment-15261802 ] Sachin Aggarwal edited comment on SPARK-14597 at 4/28/16 9:37 AM: -- Hi [~prashant_] I have generated few metric to support the usefulness, here are the graphs for the same. Image 1 shows the increase in jobSetGenerateTimeDelay with decrease in batchInterval Image 2 shows for a batch Interval of 100ms, how jobSetGenerateTimeDelay keeps on increasing with time. !1.png|width=300,height=400! !2.png|width=300,height=400! was (Author: sachin aggarwal): Hi [~prashant_] I have generated few metric to support the usefulness, here are the graphs for the same. Image 1 shows the increase in jobSetGenerateTimeDelay with decrease in batchInterval Image 2 shows for a batch Interval of 100ms, how jobSetGenerateTimeDelay keeps on increasing with time. !1.png! !2.png! > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > Attachments: 1.png, 2.png > > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Comment Edited] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15261802#comment-15261802 ] Sachin Aggarwal edited comment on SPARK-14597 at 4/28/16 9:35 AM: -- Hi [~prashant_] I have generated few metric to support the usefulness, here are the graphs for the same. Image 1 shows the increase in jobSetGenerateTimeDelay with decrease in batchInterval Image 2 shows for a batch Interval of 100ms, how jobSetGenerateTimeDelay keeps on increasing with time. !1.png! !2.png! was (Author: sachin aggarwal): Hi [~prashant_] I have generated few metric to support the usefulness, here are the graphs for the same. Image 1 shows the increase in jobSetGenerateTimeDelay with decrease in batchInterval Image 2 shows for a batch Interval of 100ms, how jobSetGenerateTimeDelay keeps on increasing with time. !1.png|thumbnail! !2.png|thumbnail! > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > Attachments: 1.png, 2.png > > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Comment Edited] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15261858#comment-15261858 ] Sachin Aggarwal edited comment on SPARK-14597 at 4/28/16 9:33 AM: -- based on over use case and analysis we have found that the major contributor to job generate time are : 1) sortByKey function as it submits a job to cluster for sketch function (this is a directly proportional to the data size being processed in that batch) 2) In direct kafka fetching offset takes nearly 2 ms 3) nearly for every two transformation we add 1ms to our job generate time delay , this time is consumed i executing getOrCompute method. was (Author: sachin aggarwal): based on over use case and analysis we have found that the major contributor to job generate time are : 1) sortByKey function as it submits a job to cluster for sketch function 2) In direct kafka fetching offset takes nearly 2 ms 3) nearly for every two transformation we add 1ms to our job generate time delay , this time is consumed i executing getOrCompute method. > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > Attachments: 1.png, 2.png > > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15261858#comment-15261858 ] Sachin Aggarwal commented on SPARK-14597: - based on over use case and analysis we have found that the major contributor to job generate time are : 1) sortByKey function as it submits a job to cluster for sketch function 2) In direct kafka fetching offset takes nearly 2 ms 3) nearly for every two transformation we add 1ms to our job generate time delay , this time is consumed i executing getOrCompute method. > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > Attachments: 1.png, 2.png > > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Updated] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Sachin Aggarwal updated SPARK-14597: Attachment: 2.png 1.png Hi [~prashant_] I have generated few metric to support the usefulness, here are the graphs for the same. Image 1 shows the increase in jobSetGenerateTimeDelay with decrease in batchInterval Image 2 shows for a batch Interval of 100ms, how jobSetGenerateTimeDelay keeps on increasing with time. !1.png|thumbnail! !2.png|thumbnail! > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > Attachments: 1.png, 2.png > > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Updated] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Sachin Aggarwal updated SPARK-14597: Attachment: (was: 2.png) > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Issue Comment Deleted] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Sachin Aggarwal updated SPARK-14597: Comment: was deleted (was: Hi [~prashant_] i have generated few metric to support the usefulness, here are the graphs for the same !1.jpg|thumbnail! !2.jpg|thumbnail!) > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Updated] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Sachin Aggarwal updated SPARK-14597: Attachment: (was: 1.png) > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Updated] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Sachin Aggarwal updated SPARK-14597: Attachment: 2.png 1.png adding analysis graph > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > Attachments: 1.png, 2.png > > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Comment Edited] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15261797#comment-15261797 ] Sachin Aggarwal edited comment on SPARK-14597 at 4/28/16 8:59 AM: -- Hi [~prashant_] i have generated few metric to support the usefulness, here are the graphs for the same !1.jpg|thumbnail! !2.jpg|thumbnail! was (Author: sachin aggarwal): adding analysis graph > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > Attachments: 1.png, 2.png > > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15250661#comment-15250661 ] Sachin Aggarwal commented on SPARK-14597: - hi Mario, I have extended the approach2 added a new parameter to job class to capture the job creation time delay, once the job gets created, I set time taken to create each job and user can get this information in StreamingListener methods onOutputOperationStarted and onOutputOperationCompleted corresponding to each job, for batch level data user can use batchCompleted.batchInfo.batchJobSetCreationDelay in onBatchCompleted method of StreamingListener > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15239122#comment-15239122 ] Sachin Aggarwal commented on SPARK-14597: - I can see 2 approaches to providing the above additional metrics. I have attached a PR for each approach. This is to review either approach and we can pick the better one 1) add new events to listener bus:- add more event case classes case class StreamingListenerBatchGenerateStarted(time: Time) extends StreamingListenerEvent case class StreamingListenerBatchGenerateCompleted(time: Time) extends StreamingListenerEvent case class StreamingListenerCheckpointingStarted(time: Time) extends StreamingListenerEvent case class StreamingListenerCheckpointingCompleted(time: Time) extends StreamingListenerEvent and new functions to the listener interface for receiving information def onBatchGenerateStarted(batchGenerateStarted: StreamingListenerBatchGenerateStarted) { } def onBatchGenerateCompleted(batchGenerateCompleted: StreamingListenerBatchGenerateCompleted) { } def onCheckpointingStarted(checkpointingStarted: StreamingListenerCheckpointingStarted) { } def onCheckpointingCompleted(checkpointingCompleted: StreamingListenerCheckpointingCompleted) { } 2) add one parameter to JobSet and pass it to BatchInfo Class to track JobSet CreationDelay for each batch. As jobSet CreationDelay is related to a batch, this can be a part of BatchInfo. For checkpointing we can use listener approach as checkpointing as checkpointing is triggered at checkpoint interval not manadaterally at batch interval. case class JobSet( time: Time, jobSetCreationDelay: Option[Long], jobs: Seq[Job], streamIdToInputInfo: Map[Int, StreamInputInfo] = Map.empty) { case class BatchInfo( batchTime: Time, creationDelay: Option[Long], streamIdToInputInfo: Map[Int, StreamInputInfo], submissionTime: Long, processingStartTime: Option[Long], processingEndTime: Option[Long], outputOperationInfos: Map[Int, OutputOperationInfo] ) { } > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) > b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Updated] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Sachin Aggarwal updated SPARK-14597: Description: While looking to tune our streaming application, the piece of info we were looking for was actual processing time per batch. The StreamingListener.onBatchCompleted event provides a BatchInfo object that provided this information. It provides the following data - processingDelay - schedulingDelay - totalDelay - Submission Time The above are essentially calculated from the streaming JobScheduler clocking the processingStartTime and processingEndTime for each JobSet. Another metric available is submissionTime which is when a Jobset was put on the Streaming Scheduler's Queue. So we took processing delay as our actual processing time per batch. However to maintain a stable streaming application, we found that the our batch interval had to be a little less than DOUBLE of the processingDelay metric reported. (We are using a DirectKafkaInputStream). On digging further, we found that processingDelay is only clocking time spent in the ForEachRDD closure of the Streaming application and that JobGenerator's graph.generateJobs (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) method takes a significant more amount of time. Thus a true reflection of processing time is a - Time spent in JobGenerator's Job Queue (JobGeneratorQueueDelay) b - Time spent in JobGenerator's graph.generateJobs (JobSetCreationDelay) c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay metric) d - Time spent in Jobset's job run (existing processingDelay metric) Additionally a JobGeneratorQueue delay (#a) could be due to either graph.generateJobs taking longer than batchInterval or other JobGenerator events like checkpointing adding up time. Thus it would be beneficial to report time taken by the checkpointing Job as well was: While looking to tune our streaming application, the piece of info we were looking for was actual processing time per batch. The StreamingListener.onBatchCompleted event provides a BatchInfo object that provided this information. It provides the following data - processingDelay - schedulingDelay - totalDelay - Submission Time The above are essentially calculated from the streaming JobScheduler clocking the processingStartTime and processingEndTime for each JobSet. Another metric available is submissionTime which is when a Jobset was put on the Streaming Scheduler's Queue. So we took processing delay as our actual processing time per batch. However to maintain a stable streaming application, we found that the our batch interval had to be a little less than DOUBLE of the processingDelay metric reported. (We are using a DirectKafkaInputStream). On digging further, we found that processingDelay is only clocking time spent in the ForEachRDD closure of the Streaming application and that JobGenerator's graph.generateJobs (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) method takes a significant more amount of time. Thus a true reflection of processing time is a - Time spent in JobGenerator's Job Queue (jobGenerator scheduling delay or JobGeneratorQueue delay) b - Time spent in JobGenerator's graph.generateJobs (JobSetCreation delay) c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay metric) d - Time spent in Jobset's job run (existing processingDelay metric) Additionally a JobGeneratorQueue delay (#a) could be due to either graph.generateJobs taking longer than batchInterval or other JobGenerator events like checkpointing adding up time. Thus it would be beneficial to report time taken by the checkpointing Job as well > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. >
[jira] [Updated] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Sachin Aggarwal updated SPARK-14597: Description: While looking to tune our streaming application, the piece of info we were looking for was actual processing time per batch. The StreamingListener.onBatchCompleted event provides a BatchInfo object that provided this information. It provides the following data - processingDelay - schedulingDelay - totalDelay - Submission Time The above are essentially calculated from the streaming JobScheduler clocking the processingStartTime and processingEndTime for each JobSet. Another metric available is submissionTime which is when a Jobset was put on the Streaming Scheduler's Queue. So we took processing delay as our actual processing time per batch. However to maintain a stable streaming application, we found that the our batch interval had to be a little less than DOUBLE of the processingDelay metric reported. (We are using a DirectKafkaInputStream). On digging further, we found that processingDelay is only clocking time spent in the ForEachRDD closure of the Streaming application and that JobGenerator's graph.generateJobs (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) method takes a significant more amount of time. Thus a true reflection of processing time is a - Time spent in JobGenerator's Job Queue (jobGenerator scheduling delay or JobGeneratorQueue delay) b - Time spent in JobGenerator's graph.generateJobs (JobSetCreation delay) c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay metric) d - Time spent in Jobset's job run (existing processingDelay metric) Additionally a JobGeneratorQueue delay (#a) could be due to either graph.generateJobs taking longer than batchInterval or other JobGenerator events like checkpointing adding up time. Thus it would be beneficial to report time taken by the checkpointing Job as well was: While looking to tune our streaming application, the piece of info we were looking for was actual processing time per batch. The StreamingListener.onBatchCompleted event provides a BatchInfo object that provided this information. It provides the following data - processingDelay - schedulingDelay - totalDelay - Submission Time The above are essentially calculated from the streaming JobScheduler clocking the processingStartTime and processingEndTime for each JobSet. Another metric available is submissionTime which is when a Jobset was put on the Streaming Scheduler's Queue. So we took processing delay as our actual processing time per batch. However to maintain a stable streaming application, we found that the our batch interval had to be a little less than DOUBLE of the processingDelay metric reported. (We are using a DirectKafkaInputStream). On digging further, we found that processingDelay is only clocking time spent in the ForEachRDD closure of the Streaming application and that JobGenerator's graph.generateJobs (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) method takes a significant more amount of time. Thus a true reflection of processing time is a - Time spent in JobGenerator's Job Queue (jobGenerator scheduling delay or JobGeneratorQueue delay) b - Time spent in JobGenerator's graph.generateJobs (generateJobProcessingDelay) c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay metric) d - Time spent in Jobset's job run (existing processingDelay metric) Additionally a JobGeneratorQueue delay (#a) could be due to either graph.generateJobs taking longer than batchInterval or other JobGenerator events like checkpointing adding up time. Thus it would be beneficial to report time taken by the checkpointing Job as well > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTim
[jira] [Updated] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
[ https://issues.apache.org/jira/browse/SPARK-14597?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Sachin Aggarwal updated SPARK-14597: Component/s: Spark Core > Streaming Listener timing metrics should include time spent in JobGenerator's > graph.generateJobs > > > Key: SPARK-14597 > URL: https://issues.apache.org/jira/browse/SPARK-14597 > Project: Spark > Issue Type: Improvement > Components: Spark Core, Streaming >Affects Versions: 1.6.1, 2.0.0 >Reporter: Sachin Aggarwal >Priority: Minor > > While looking to tune our streaming application, the piece of info we were > looking for was actual processing time per batch. The > StreamingListener.onBatchCompleted event provides a BatchInfo object that > provided this information. It provides the following data > - processingDelay > - schedulingDelay > - totalDelay > - Submission Time > The above are essentially calculated from the streaming JobScheduler > clocking the processingStartTime and processingEndTime for each JobSet. > Another metric available is submissionTime which is when a Jobset was put on > the Streaming Scheduler's Queue. > > So we took processing delay as our actual processing time per batch. However > to maintain a stable streaming application, we found that the our batch > interval had to be a little less than DOUBLE of the processingDelay metric > reported. (We are using a DirectKafkaInputStream). On digging further, we > found that processingDelay is only clocking time spent in the ForEachRDD > closure of the Streaming application and that JobGenerator's > graph.generateJobs > (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) > method takes a significant more amount of time. > Thus a true reflection of processing time is > a - Time spent in JobGenerator's Job Queue (jobGenerator scheduling delay or > JobGeneratorQueue delay) > b - Time spent in JobGenerator's graph.generateJobs > (generateJobProcessingDelay) > c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay > metric) > d - Time spent in Jobset's job run (existing processingDelay metric) > > Additionally a JobGeneratorQueue delay (#a) could be due to either > graph.generateJobs taking longer than batchInterval or other JobGenerator > events like checkpointing adding up time. Thus it would be beneficial to > report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Created] (SPARK-14597) Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs
Sachin Aggarwal created SPARK-14597: --- Summary: Streaming Listener timing metrics should include time spent in JobGenerator's graph.generateJobs Key: SPARK-14597 URL: https://issues.apache.org/jira/browse/SPARK-14597 Project: Spark Issue Type: Improvement Components: Streaming Affects Versions: 1.6.1, 2.0.0 Reporter: Sachin Aggarwal Priority: Minor While looking to tune our streaming application, the piece of info we were looking for was actual processing time per batch. The StreamingListener.onBatchCompleted event provides a BatchInfo object that provided this information. It provides the following data - processingDelay - schedulingDelay - totalDelay - Submission Time The above are essentially calculated from the streaming JobScheduler clocking the processingStartTime and processingEndTime for each JobSet. Another metric available is submissionTime which is when a Jobset was put on the Streaming Scheduler's Queue. So we took processing delay as our actual processing time per batch. However to maintain a stable streaming application, we found that the our batch interval had to be a little less than DOUBLE of the processingDelay metric reported. (We are using a DirectKafkaInputStream). On digging further, we found that processingDelay is only clocking time spent in the ForEachRDD closure of the Streaming application and that JobGenerator's graph.generateJobs (https://github.com/apache/spark/blob/branch-1.6/streaming/src/main/scala/org/apache/spark/streaming/scheduler/JobGenerator.scala#L248) method takes a significant more amount of time. Thus a true reflection of processing time is a - Time spent in JobGenerator's Job Queue (jobGenerator scheduling delay or JobGeneratorQueue delay) b - Time spent in JobGenerator's graph.generateJobs (generateJobProcessingDelay) c - Time spent in JobScheduler Queue for a Jobset (existing schedulingDelay metric) d - Time spent in Jobset's job run (existing processingDelay metric) Additionally a JobGeneratorQueue delay (#a) could be due to either graph.generateJobs taking longer than batchInterval or other JobGenerator events like checkpointing adding up time. Thus it would be beneficial to report time taken by the checkpointing Job as well -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-13172) Stop using RichException.getStackTrace it is deprecated
[ https://issues.apache.org/jira/browse/SPARK-13172?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15138498#comment-15138498 ] sachin aggarwal commented on SPARK-13172: - Now the question is where should I add this function so it can be leveraged across modules? > Stop using RichException.getStackTrace it is deprecated > --- > > Key: SPARK-13172 > URL: https://issues.apache.org/jira/browse/SPARK-13172 > Project: Spark > Issue Type: Sub-task > Components: Spark Core >Reporter: holdenk >Priority: Trivial > > Throwable getStackTrace is the recommended alternative. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Comment Edited] (SPARK-13172) Stop using RichException.getStackTrace it is deprecated
[ https://issues.apache.org/jira/browse/SPARK-13172?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15138423#comment-15138423 ] sachin aggarwal edited comment on SPARK-13172 at 2/9/16 6:49 AM: - There are two ways we can proceed :- first use printStackTrace and second to use mkString() 1) can be ecapsulated in function :- {code} def getStackTraceAsString(t: Throwable) = { val sw = new StringWriter t.printStackTrace(new PrintWriter(sw)) sw.toString } {code} 2) println(t.getStackTrace.mkString("\n")) mkstring approach give extractly same string as old function getStackTraceString but the output of first approach is more readable. h3.Example {code:title=TrySuccessFailure.scala|borderStyle=solid} import scala.util.{Try, Success, Failure} import java.io._ object TrySuccessFailure extends App { badAdder(3) match { case Success(i) => println(s"success, i = $i") case Failure(t) => // this works, but it's not too useful/readable println(t.getStackTrace.mkString("\n")) println("===") println(t.getStackTraceString) // this works much better val sw = new StringWriter t.printStackTrace(new PrintWriter(sw)) println(sw.toString) } def badAdder(a: Int): Try[Int] = { Try({ val b = a + 1 if (b == 3) b else { val ioe = new IOException("Boom!") throw new AlsException("Bummer!", ioe) } }) } class AlsException(s: String, e: Exception) extends Exception(s: String, e: Exception) } {code} was (Author: sachin aggarwal): there are two ways we can proceed :- first use printStackTrace and second to use mkString() 1) can be ecapsulated in function :- {code} def getStackTraceAsString(t: Throwable) = { val sw = new StringWriter t.printStackTrace(new PrintWriter(sw)) sw.toString } {code} 2) println(t.getStackTrace.mkString("\n")) mkstring approach give extractly same string as old function getStackTraceString but the output of first approach is more readable. h3.Example {code:title=TrySuccessFailure.scala|borderStyle=solid} import scala.util.{Try, Success, Failure} import java.io._ object TrySuccessFailure extends App { badAdder(3) match { case Success(i) => println(s"success, i = $i") case Failure(t) => // this works, but it's not too useful/readable println(t.getStackTrace.mkString("\n")) println("===") println(t.getStackTraceString) // this works much better val sw = new StringWriter t.printStackTrace(new PrintWriter(sw)) println(sw.toString) } def badAdder(a: Int): Try[Int] = { Try({ val b = a + 1 if (b == 3) b else { val ioe = new IOException("Boom!") throw new AlsException("Bummer!", ioe) } }) } class AlsException(s: String, e: Exception) extends Exception(s: String, e: Exception) } {code} > Stop using RichException.getStackTrace it is deprecated > --- > > Key: SPARK-13172 > URL: https://issues.apache.org/jira/browse/SPARK-13172 > Project: Spark > Issue Type: Sub-task > Components: Spark Core >Reporter: holdenk >Priority: Trivial > > Throwable getStackTrace is the recommended alternative. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Comment Edited] (SPARK-13172) Stop using RichException.getStackTrace it is deprecated
[ https://issues.apache.org/jira/browse/SPARK-13172?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15138423#comment-15138423 ] sachin aggarwal edited comment on SPARK-13172 at 2/9/16 6:35 AM: - there are two ways we can proceed :- first use printStackTrace and second to use mkString() 1) can be ecapsulated in function :- {code} def getStackTraceAsString(t: Throwable) = { val sw = new StringWriter t.printStackTrace(new PrintWriter(sw)) sw.toString } {code} 2) println(t.getStackTrace.mkString("\n")) mkstring approach give extractly same string as old function getStackTraceString but the output of first approach is more readable. h3.Example {code:title=TrySuccessFailure.scala|borderStyle=solid} import scala.util.{Try, Success, Failure} import java.io._ object TrySuccessFailure extends App { badAdder(3) match { case Success(i) => println(s"success, i = $i") case Failure(t) => // this works, but it's not too useful/readable println(t.getStackTrace.mkString("\n")) println("===") println(t.getStackTraceString) // this works much better val sw = new StringWriter t.printStackTrace(new PrintWriter(sw)) println(sw.toString) } def badAdder(a: Int): Try[Int] = { Try({ val b = a + 1 if (b == 3) b else { val ioe = new IOException("Boom!") throw new AlsException("Bummer!", ioe) } }) } class AlsException(s: String, e: Exception) extends Exception(s: String, e: Exception) } {code} was (Author: sachin aggarwal): there are two ways we can proceed :- first use printStackTrace and second to use mkString() 1) can be ecapsulated in function :- {code} def getStackTraceAsString(t: Throwable) = { val sw = new StringWriter t.printStackTrace(new PrintWriter(sw)) sw.toString } {code} 2) println(t.getStackTrace.mkString("\n")) mkstring approach give extractly same string as old function getStackTraceString but the output of first approach is more readable. try this code to see the difference {code:title=TrySuccessFailure.scala|borderStyle=solid} import scala.util.{Try, Success, Failure} import java.io._ object TrySuccessFailure extends App { badAdder(3) match { case Success(i) => println(s"success, i = $i") case Failure(t) => // this works, but it's not too useful/readable println(t.getStackTrace.mkString("\n")) println("===") println(t.getStackTraceString) // this works much better val sw = new StringWriter t.printStackTrace(new PrintWriter(sw)) println(sw.toString) } def badAdder(a: Int): Try[Int] = { Try({ val b = a + 1 if (b == 3) b else { val ioe = new IOException("Boom!") throw new AlsException("Bummer!", ioe) } }) } class AlsException(s: String, e: Exception) extends Exception(s: String, e: Exception) } {code} > Stop using RichException.getStackTrace it is deprecated > --- > > Key: SPARK-13172 > URL: https://issues.apache.org/jira/browse/SPARK-13172 > Project: Spark > Issue Type: Sub-task > Components: Spark Core >Reporter: holdenk >Priority: Trivial > > Throwable getStackTrace is the recommended alternative. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Comment Edited] (SPARK-13172) Stop using RichException.getStackTrace it is deprecated
[ https://issues.apache.org/jira/browse/SPARK-13172?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15138423#comment-15138423 ] sachin aggarwal edited comment on SPARK-13172 at 2/9/16 6:33 AM: - there are two ways we can proceed :- first use printStackTrace and second to use mkString() 1) can be ecapsulated in function :- def getStackTraceAsString(t: Throwable) = { val sw = new StringWriter t.printStackTrace(new PrintWriter(sw)) sw.toString } 2) println(t.getStackTrace.mkString("\n")) mkstring approach give extractly same string as old function getStackTraceString but the output of first approach is more readable. try this code to see the difference {code:title=TrySuccessFailure.scala|borderStyle=solid} import scala.util.{Try, Success, Failure} import java.io._ object TrySuccessFailure extends App { badAdder(3) match { case Success(i) => println(s"success, i = $i") case Failure(t) => // this works, but it's not too useful/readable println(t.getStackTrace.mkString("\n")) println("===") println(t.getStackTraceString) // this works much better val sw = new StringWriter t.printStackTrace(new PrintWriter(sw)) println(sw.toString) } def badAdder(a: Int): Try[Int] = { Try({ val b = a + 1 if (b == 3) b else { val ioe = new IOException("Boom!") throw new AlsException("Bummer!", ioe) } }) } class AlsException(s: String, e: Exception) extends Exception(s: String, e: Exception) } {code} was (Author: sachin aggarwal): there are two ways we can proceed :- first use printStackTrace and second to use mkString() 1) can be ecapsulated in function :- def getStackTraceAsString(t: Throwable) = { val sw = new StringWriter t.printStackTrace(new PrintWriter(sw)) sw.toString } 2) println(t.getStackTrace.mkString("\n")) mkstring approach give extractly same string as old function getStackTraceString but the output of first approach is more readable. try this code to see the difference ``` import scala.util.{Try, Success, Failure} import java.io._ object TrySuccessFailure extends App { badAdder(3) match { case Success(i) => println(s"success, i = $i") case Failure(t) => // this works, but it's not too useful/readable println(t.getStackTrace.mkString("\n")) println("===") println(t.getStackTraceString) // this works much better val sw = new StringWriter t.printStackTrace(new PrintWriter(sw)) println(sw.toString) } def badAdder(a: Int): Try[Int] = { Try({ val b = a + 1 if (b == 3) b else { val ioe = new IOException("Boom!") throw new AlsException("Bummer!", ioe) } }) } class AlsException(s: String, e: Exception) extends Exception(s: String, e: Exception) } ``` > Stop using RichException.getStackTrace it is deprecated > --- > > Key: SPARK-13172 > URL: https://issues.apache.org/jira/browse/SPARK-13172 > Project: Spark > Issue Type: Sub-task > Components: Spark Core >Reporter: holdenk >Priority: Trivial > > Throwable getStackTrace is the recommended alternative. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Comment Edited] (SPARK-13172) Stop using RichException.getStackTrace it is deprecated
[ https://issues.apache.org/jira/browse/SPARK-13172?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15138423#comment-15138423 ] sachin aggarwal edited comment on SPARK-13172 at 2/9/16 6:33 AM: - there are two ways we can proceed :- first use printStackTrace and second to use mkString() 1) can be ecapsulated in function :- {code} def getStackTraceAsString(t: Throwable) = { val sw = new StringWriter t.printStackTrace(new PrintWriter(sw)) sw.toString } {code} 2) println(t.getStackTrace.mkString("\n")) mkstring approach give extractly same string as old function getStackTraceString but the output of first approach is more readable. try this code to see the difference {code:title=TrySuccessFailure.scala|borderStyle=solid} import scala.util.{Try, Success, Failure} import java.io._ object TrySuccessFailure extends App { badAdder(3) match { case Success(i) => println(s"success, i = $i") case Failure(t) => // this works, but it's not too useful/readable println(t.getStackTrace.mkString("\n")) println("===") println(t.getStackTraceString) // this works much better val sw = new StringWriter t.printStackTrace(new PrintWriter(sw)) println(sw.toString) } def badAdder(a: Int): Try[Int] = { Try({ val b = a + 1 if (b == 3) b else { val ioe = new IOException("Boom!") throw new AlsException("Bummer!", ioe) } }) } class AlsException(s: String, e: Exception) extends Exception(s: String, e: Exception) } {code} was (Author: sachin aggarwal): there are two ways we can proceed :- first use printStackTrace and second to use mkString() 1) can be ecapsulated in function :- def getStackTraceAsString(t: Throwable) = { val sw = new StringWriter t.printStackTrace(new PrintWriter(sw)) sw.toString } 2) println(t.getStackTrace.mkString("\n")) mkstring approach give extractly same string as old function getStackTraceString but the output of first approach is more readable. try this code to see the difference {code:title=TrySuccessFailure.scala|borderStyle=solid} import scala.util.{Try, Success, Failure} import java.io._ object TrySuccessFailure extends App { badAdder(3) match { case Success(i) => println(s"success, i = $i") case Failure(t) => // this works, but it's not too useful/readable println(t.getStackTrace.mkString("\n")) println("===") println(t.getStackTraceString) // this works much better val sw = new StringWriter t.printStackTrace(new PrintWriter(sw)) println(sw.toString) } def badAdder(a: Int): Try[Int] = { Try({ val b = a + 1 if (b == 3) b else { val ioe = new IOException("Boom!") throw new AlsException("Bummer!", ioe) } }) } class AlsException(s: String, e: Exception) extends Exception(s: String, e: Exception) } {code} > Stop using RichException.getStackTrace it is deprecated > --- > > Key: SPARK-13172 > URL: https://issues.apache.org/jira/browse/SPARK-13172 > Project: Spark > Issue Type: Sub-task > Components: Spark Core >Reporter: holdenk >Priority: Trivial > > Throwable getStackTrace is the recommended alternative. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-13172) Stop using RichException.getStackTrace it is deprecated
[ https://issues.apache.org/jira/browse/SPARK-13172?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15138423#comment-15138423 ] sachin aggarwal commented on SPARK-13172: - there are two ways we can proceed :- first use printStackTrace and second to use mkString() 1) can be ecapsulated in function :- def getStackTraceAsString(t: Throwable) = { val sw = new StringWriter t.printStackTrace(new PrintWriter(sw)) sw.toString } 2) println(t.getStackTrace.mkString("\n")) mkstring approach give extractly same string as old function getStackTraceString but the output of first approach is more readable. try this code to see the difference ``` import scala.util.{Try, Success, Failure} import java.io._ object TrySuccessFailure extends App { badAdder(3) match { case Success(i) => println(s"success, i = $i") case Failure(t) => // this works, but it's not too useful/readable println(t.getStackTrace.mkString("\n")) println("===") println(t.getStackTraceString) // this works much better val sw = new StringWriter t.printStackTrace(new PrintWriter(sw)) println(sw.toString) } def badAdder(a: Int): Try[Int] = { Try({ val b = a + 1 if (b == 3) b else { val ioe = new IOException("Boom!") throw new AlsException("Bummer!", ioe) } }) } class AlsException(s: String, e: Exception) extends Exception(s: String, e: Exception) } ``` > Stop using RichException.getStackTrace it is deprecated > --- > > Key: SPARK-13172 > URL: https://issues.apache.org/jira/browse/SPARK-13172 > Project: Spark > Issue Type: Sub-task > Components: Spark Core >Reporter: holdenk >Priority: Trivial > > Throwable getStackTrace is the recommended alternative. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-13172) Stop using RichException.getStackTrace it is deprecated
[ https://issues.apache.org/jira/browse/SPARK-13172?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15137136#comment-15137136 ] sachin aggarwal commented on SPARK-13172: - instead of getStackTraceString should I use e.getStackTrace or e.printStackTrace > Stop using RichException.getStackTrace it is deprecated > --- > > Key: SPARK-13172 > URL: https://issues.apache.org/jira/browse/SPARK-13172 > Project: Spark > Issue Type: Sub-task > Components: Spark Core >Reporter: holdenk >Priority: Trivial > > Throwable getStackTrace is the recommended alternative. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-13172) Stop using RichException.getStackTrace it is deprecated
[ https://issues.apache.org/jira/browse/SPARK-13172?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15133963#comment-15133963 ] sachin aggarwal commented on SPARK-13172: - as scala also recommends the same http://www.scala-lang.org/api/2.11.1/index.html#scala.runtime.RichException it should be change , I will go ahead and make the change > Stop using RichException.getStackTrace it is deprecated > --- > > Key: SPARK-13172 > URL: https://issues.apache.org/jira/browse/SPARK-13172 > Project: Spark > Issue Type: Sub-task > Components: Spark Core >Reporter: holdenk >Priority: Trivial > > Throwable getStackTrace is the recommended alternative. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Comment Edited] (SPARK-13177) Update ActorWordCount example to not directly use low level linked list as it is deprecated.
[ https://issues.apache.org/jira/browse/SPARK-13177?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15133954#comment-15133954 ] sachin aggarwal edited comment on SPARK-13177 at 2/5/16 10:23 AM: -- Hi [~holdenk] I will like to work on this, thanks Sachin was (Author: sachin aggarwal): Hi [~AlHolden], I will like to work on this, thanks Sachin > Update ActorWordCount example to not directly use low level linked list as it > is deprecated. > > > Key: SPARK-13177 > URL: https://issues.apache.org/jira/browse/SPARK-13177 > Project: Spark > Issue Type: Sub-task > Components: Examples >Reporter: holdenk >Priority: Minor > -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-13177) Update ActorWordCount example to not directly use low level linked list as it is deprecated.
[ https://issues.apache.org/jira/browse/SPARK-13177?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15133954#comment-15133954 ] sachin aggarwal commented on SPARK-13177: - Hi [~AlHolden], I will like to work on this, thanks Sachin > Update ActorWordCount example to not directly use low level linked list as it > is deprecated. > > > Key: SPARK-13177 > URL: https://issues.apache.org/jira/browse/SPARK-13177 > Project: Spark > Issue Type: Sub-task > Components: Examples >Reporter: holdenk >Priority: Minor > -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-13065) streaming-twitter pass twitter4j.FilterQuery argument to TwitterUtils.createStream()
[ https://issues.apache.org/jira/browse/SPARK-13065?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15129802#comment-15129802 ] sachin aggarwal commented on SPARK-13065: - happy to see, thats exactly what I have added have a look at this file to see how to use new API for java use case :- https://github.com/agsachin/spark/blob/SPARK-13065/external/twitter/src/test/java/org/apache/spark/streaming/twitter/JavaTwitterStreamSuite.java and for scala check this out https://github.com/agsachin/spark/blob/SPARK-13065/external/twitter/src/test/scala/org/apache/spark/streaming/twitter/TwitterStreamSuite.scala > streaming-twitter pass twitter4j.FilterQuery argument to > TwitterUtils.createStream() > > > Key: SPARK-13065 > URL: https://issues.apache.org/jira/browse/SPARK-13065 > Project: Spark > Issue Type: Improvement > Components: Streaming >Affects Versions: 1.6.0 > Environment: all >Reporter: Andrew Davidson >Priority: Minor > Labels: twitter > Attachments: twitterFilterQueryPatch.tar.gz > > Original Estimate: 2h > Remaining Estimate: 2h > > The twitter stream api is very powerful provides a lot of support for > twitter.com side filtering of status objects. When ever possible we want to > let twitter do as much work as possible for us. > currently the spark twitter api only allows you to configure a small sub set > of possible filters > String{} filters = {"tag1", tag2"} > JavaDStream tweets =TwitterUtils.createStream(ssc, twitterAuth, > filters); > The current implemenation does > private[streaming] > class TwitterReceiver( > twitterAuth: Authorization, > filters: Seq[String], > storageLevel: StorageLevel > ) extends Receiver[Status](storageLevel) with Logging { > . . . > val query = new FilterQuery > if (filters.size > 0) { > query.track(filters.mkString(",")) > newTwitterStream.filter(query) > } else { > newTwitterStream.sample() > } > ... > rather than construct the FilterQuery object in TwitterReceiver.onStart(). we > should be able to pass a FilterQueryObject > looks like an easy fix. See source code links bellow > kind regards > Andy > https://github.com/apache/spark/blob/master/external/twitter/src/main/scala/org/apache/spark/streaming/twitter/TwitterInputDStream.scala#L60 > https://github.com/apache/spark/blob/master/external/twitter/src/main/scala/org/apache/spark/streaming/twitter/TwitterInputDStream.scala#L89 > $ 2/2/16 > attached is my java implementation for this problem. Feel free to reuse it > how ever you like. In my streaming spark app main() I have the following code >FilterQuery query = config.getFilterQuery().fetch(); > if (query != null) { > // TODO https://issues.apache.org/jira/browse/SPARK-13065 > tweets = TwitterFilterQueryUtils.createStream(ssc, twitterAuth, > query); > } /*else > spark native api > String[] filters = {"tag1", tag2"} > tweets = TwitterUtils.createStream(ssc, twitterAuth, filters); > > see > https://github.com/apache/spark/blob/master/external/twitter/src/main/scala/org/apache/spark/streaming/twitter/TwitterInputDStream.scala#L89 > > causes > val query = new FilterQuery > if (filters.size > 0) { > query.track(filters.mkString(",")) > newTwitterStream.filter(query) > } */ -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-13069) ActorHelper is not throttled by rate limiter
[ https://issues.apache.org/jira/browse/SPARK-13069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15129787#comment-15129787 ] sachin aggarwal commented on SPARK-13069: - As of Spark 2.0 (not yet released), Spark does not use Akka any more. See https://issues.apache.org/jira/browse/SPARK-5293 can you check with latest 2.0 build, to see if similar problem exists. > ActorHelper is not throttled by rate limiter > > > Key: SPARK-13069 > URL: https://issues.apache.org/jira/browse/SPARK-13069 > Project: Spark > Issue Type: Bug > Components: Streaming >Affects Versions: 1.6.0 >Reporter: Lin Zhao > > The rate an actor receiver sends data to spark is not limited by maxRate or > back pressure. Spark would control how fast it writes the data to block > manager, but the receiver actor sends events asynchronously and would fill > out akka mailbox with millions of events until memory runs out. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-13065) streaming-twitter pass twitter4j.FilterQuery argument to TwitterUtils.createStream()
[ https://issues.apache.org/jira/browse/SPARK-13065?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15127935#comment-15127935 ] sachin aggarwal commented on SPARK-13065: - [~aedwip] I got a doubt after reading ur last comment you mentioned FilterQuery in description and here you are addressing twitter4j.query, please clarify. > streaming-twitter pass twitter4j.FilterQuery argument to > TwitterUtils.createStream() > > > Key: SPARK-13065 > URL: https://issues.apache.org/jira/browse/SPARK-13065 > Project: Spark > Issue Type: Improvement > Components: Streaming >Affects Versions: 1.6.0 > Environment: all >Reporter: Andrew Davidson >Priority: Minor > Labels: twitter > Original Estimate: 2h > Remaining Estimate: 2h > > The twitter stream api is very powerful provides a lot of support for > twitter.com side filtering of status objects. When ever possible we want to > let twitter do as much work as possible for us. > currently the spark twitter api only allows you to configure a small sub set > of possible filters > String{} filters = {"tag1", tag2"} > JavaDStream tweets =TwitterUtils.createStream(ssc, twitterAuth, > filters); > The current implemenation does > private[streaming] > class TwitterReceiver( > twitterAuth: Authorization, > filters: Seq[String], > storageLevel: StorageLevel > ) extends Receiver[Status](storageLevel) with Logging { > . . . > val query = new FilterQuery > if (filters.size > 0) { > query.track(filters.mkString(",")) > newTwitterStream.filter(query) > } else { > newTwitterStream.sample() > } > ... > rather than construct the FilterQuery object in TwitterReceiver.onStart(). we > should be able to pass a FilterQueryObject > looks like an easy fix. See source code links bellow > kind regards > Andy > https://github.com/apache/spark/blob/master/external/twitter/src/main/scala/org/apache/spark/streaming/twitter/TwitterInputDStream.scala#L60 > https://github.com/apache/spark/blob/master/external/twitter/src/main/scala/org/apache/spark/streaming/twitter/TwitterInputDStream.scala#L89 -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-13065) streaming-twitter pass twitter4j.FilterQuery argument to TwitterUtils.createStream()
[ https://issues.apache.org/jira/browse/SPARK-13065?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15127925#comment-15127925 ] sachin aggarwal commented on SPARK-13065: - List of changes: 1) Added support for passing FilterQuery object instead of just Seq of keywords 2) Java had more flexible Api syntax than that of Scala so added similar Api syntax for Scala also 3) added test cases for the all the new Api's > streaming-twitter pass twitter4j.FilterQuery argument to > TwitterUtils.createStream() > > > Key: SPARK-13065 > URL: https://issues.apache.org/jira/browse/SPARK-13065 > Project: Spark > Issue Type: Improvement > Components: Streaming >Affects Versions: 1.6.0 > Environment: all >Reporter: Andrew Davidson >Priority: Minor > Labels: twitter > Original Estimate: 2h > Remaining Estimate: 2h > > The twitter stream api is very powerful provides a lot of support for > twitter.com side filtering of status objects. When ever possible we want to > let twitter do as much work as possible for us. > currently the spark twitter api only allows you to configure a small sub set > of possible filters > String{} filters = {"tag1", tag2"} > JavaDStream tweets =TwitterUtils.createStream(ssc, twitterAuth, > filters); > The current implemenation does > private[streaming] > class TwitterReceiver( > twitterAuth: Authorization, > filters: Seq[String], > storageLevel: StorageLevel > ) extends Receiver[Status](storageLevel) with Logging { > . . . > val query = new FilterQuery > if (filters.size > 0) { > query.track(filters.mkString(",")) > newTwitterStream.filter(query) > } else { > newTwitterStream.sample() > } > ... > rather than construct the FilterQuery object in TwitterReceiver.onStart(). we > should be able to pass a FilterQueryObject > looks like an easy fix. See source code links bellow > kind regards > Andy > https://github.com/apache/spark/blob/master/external/twitter/src/main/scala/org/apache/spark/streaming/twitter/TwitterInputDStream.scala#L60 > https://github.com/apache/spark/blob/master/external/twitter/src/main/scala/org/apache/spark/streaming/twitter/TwitterInputDStream.scala#L89 -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-13065) streaming-twitter pass twitter4j.FilterQuery argument to TwitterUtils.createStream()
[ https://issues.apache.org/jira/browse/SPARK-13065?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15125834#comment-15125834 ] sachin aggarwal commented on SPARK-13065: - I would like to work on this , will issue a pull request soon.. > streaming-twitter pass twitter4j.FilterQuery argument to > TwitterUtils.createStream() > > > Key: SPARK-13065 > URL: https://issues.apache.org/jira/browse/SPARK-13065 > Project: Spark > Issue Type: Improvement > Components: Streaming >Affects Versions: 1.6.0 > Environment: all >Reporter: Andrew Davidson >Priority: Minor > Labels: twitter > Original Estimate: 2h > Remaining Estimate: 2h > > The twitter stream api is very powerful provides a lot of support for > twitter.com side filtering of status objects. When ever possible we want to > let twitter do as much work as possible for us. > currently the spark twitter api only allows you to configure a small sub set > of possible filters > String{} filters = {"tag1", tag2"} > JavaDStream tweets =TwitterUtils.createStream(ssc, twitterAuth, > filters); > The current implemenation does > private[streaming] > class TwitterReceiver( > twitterAuth: Authorization, > filters: Seq[String], > storageLevel: StorageLevel > ) extends Receiver[Status](storageLevel) with Logging { > . . . > val query = new FilterQuery > if (filters.size > 0) { > query.track(filters.mkString(",")) > newTwitterStream.filter(query) > } else { > newTwitterStream.sample() > } > ... > rather than construct the FilterQuery object in TwitterReceiver.onStart(). we > should be able to pass a FilterQueryObject > looks like an easy fix. See source code links bellow > kind regards > Andy > https://github.com/apache/spark/blob/master/external/twitter/src/main/scala/org/apache/spark/streaming/twitter/TwitterInputDStream.scala#L60 > https://github.com/apache/spark/blob/master/external/twitter/src/main/scala/org/apache/spark/streaming/twitter/TwitterInputDStream.scala#L89 -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-12117) Column Aliases are Ignored in callUDF while using struct()
[ https://issues.apache.org/jira/browse/SPARK-12117?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15039767#comment-15039767 ] sachin aggarwal commented on SPARK-12117: - Hi, can u suggest me some work around to make this use case work in 1.5.1 ? thanks > Column Aliases are Ignored in callUDF while using struct() > -- > > Key: SPARK-12117 > URL: https://issues.apache.org/jira/browse/SPARK-12117 > Project: Spark > Issue Type: Bug > Components: SQL >Affects Versions: 1.5.1 >Reporter: sachin aggarwal > > case where this works: > val TestDoc1 = sqlContext.createDataFrame(Seq(("sachin aggarwal", "1"), > ("Rishabh", "2"))).toDF("myText", "id") > > TestDoc1.select(callUDF("mydef",struct($"myText".as("Text"),$"id".as("label"))).as("col1")).show > steps to reproduce error case: > 1)create a file copy following text--filename(a.json) > { "myText": "Sachin Aggarwal","id": "1"} > { "myText": "Rishabh","id": "2"} > 2)define a simple UDF > def mydef(r:Row)={println(r.schema); r.getAs("Text").asInstanceOf[String]} > 3)register the udf > sqlContext.udf.register("mydef" ,mydef _) > 4)read the input file > val TestDoc2=sqlContext.read.json("/tmp/a.json") > 5)make a call to UDF > TestDoc2.select(callUDF("mydef",struct($"myText".as("Text"),$"id".as("label"))).as("col1")).show > ERROR received: > java.lang.IllegalArgumentException: Field "Text" does not exist. > at > org.apache.spark.sql.types.StructType$$anonfun$fieldIndex$1.apply(StructType.scala:234) > at > org.apache.spark.sql.types.StructType$$anonfun$fieldIndex$1.apply(StructType.scala:234) > at scala.collection.MapLike$class.getOrElse(MapLike.scala:128) > at scala.collection.AbstractMap.getOrElse(Map.scala:58) > at org.apache.spark.sql.types.StructType.fieldIndex(StructType.scala:233) > at > org.apache.spark.sql.catalyst.expressions.GenericRowWithSchema.fieldIndex(rows.scala:212) > at org.apache.spark.sql.Row$class.getAs(Row.scala:325) > at org.apache.spark.sql.catalyst.expressions.GenericRow.getAs(rows.scala:191) > at > $line414.$read$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$c57ec8bf9b0d5f6161b97741d596ff0wC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC.mydef(:107) > at > $line419.$read$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$c57ec8bf9b0d5f6161b97741d596ff0wC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$anonfun$1.apply(:110) > at > $line419.$read$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$c57ec8bf9b0d5f6161b97741d596ff0wC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$anonfun$1.apply(:110) > at > org.apache.spark.sql.catalyst.expressions.ScalaUDF$$anonfun$2.apply(ScalaUDF.scala:75) > at > org.apache.spark.sql.catalyst.expressions.ScalaUDF$$anonfun$2.apply(ScalaUDF.scala:74) > at > org.apache.spark.sql.catalyst.expressions.ScalaUDF.eval(ScalaUDF.scala:964) > at > org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificMutableProjection.apply(Unknown > Source) > at > org.apache.spark.sql.execution.Project$$anonfun$1$$anonfun$apply$2.apply(basicOperators.scala:55) > at > org.apache.spark.sql.execution.Project$$anonfun$1$$anonfun$apply$2.apply(basicOperators.scala:53) > at scala.collection.Iterator$$anon$11.next(Iterator.scala:328) > at scala.collection.Iterator$$anon$11.next(Iterator.scala:328) > at scala.collection.Iterator$$anon$10.next(Iterator.scala:312) > at scala.collection.Iterator$class.foreach(Iterator.scala:727) > at scala.collection.AbstractIterator.foreach(Iterator.scala:1157) > at scala.collection.generic.Growable$class.$plus$plus$eq(Growable.scala:48) > at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:103) > at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:47) > at scala.collection.TraversableOnce$class.to(TraversableOnce.scala:273) > at scala.collection.AbstractIterator.to(Iterator.scala:1157) > at scala.collection.TraversableOnce$class.toBuffer(TraversableOnce.scala:265) > at scala.collection.AbstractIterator.toBuffer(Iterator.scala:1157) > at scala.collection.TraversableOnce$class.toArray(TraversableOnce.scala:252) > at scala.collection.AbstractIterator.toArray(Iterator.scala:1157) > at > org.apache.spark.sql.execution.SparkPlan$$anonfun$5.apply(SparkPlan.scala:215) > at > org.apache.spark.sql.execution.SparkPlan$$anonfun$5.apply(SparkPlan.scala:215) > at > org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1848) > at > org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1848) > at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:66) > at org.apache.spark.scheduler.Task.run(Task.scala:88)
[jira] [Updated] (SPARK-12117) Column Aliases are Ignored in callUDF while using struct()
[ https://issues.apache.org/jira/browse/SPARK-12117?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] sachin aggarwal updated SPARK-12117: Description: case where this works: val TestDoc1 = sqlContext.createDataFrame(Seq(("sachin aggarwal", "1"), ("Rishabh", "2"))).toDF("myText", "id") TestDoc1.select(callUDF("mydef",struct($"myText".as("Text"),$"id".as("label"))).as("col1")).show steps to reproduce error case: 1)create a file copy following text--filename(a.json) { "myText": "Sachin Aggarwal","id": "1"} { "myText": "Rishabh","id": "2"} 2)define a simple UDF def mydef(r:Row)={println(r.schema); r.getAs("Text").asInstanceOf[String]} 3)register the udf sqlContext.udf.register("mydef" ,mydef _) 4)read the input file val TestDoc2=sqlContext.read.json("/tmp/a.json") 5)make a call to UDF TestDoc2.select(callUDF("mydef",struct($"myText".as("Text"),$"id".as("label"))).as("col1")).show ERROR received: java.lang.IllegalArgumentException: Field "Text" does not exist. at org.apache.spark.sql.types.StructType$$anonfun$fieldIndex$1.apply(StructType.scala:234) at org.apache.spark.sql.types.StructType$$anonfun$fieldIndex$1.apply(StructType.scala:234) at scala.collection.MapLike$class.getOrElse(MapLike.scala:128) at scala.collection.AbstractMap.getOrElse(Map.scala:58) at org.apache.spark.sql.types.StructType.fieldIndex(StructType.scala:233) at org.apache.spark.sql.catalyst.expressions.GenericRowWithSchema.fieldIndex(rows.scala:212) at org.apache.spark.sql.Row$class.getAs(Row.scala:325) at org.apache.spark.sql.catalyst.expressions.GenericRow.getAs(rows.scala:191) at $line414.$read$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$c57ec8bf9b0d5f6161b97741d596ff0wC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC.mydef(:107) at $line419.$read$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$c57ec8bf9b0d5f6161b97741d596ff0wC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$anonfun$1.apply(:110) at $line419.$read$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$c57ec8bf9b0d5f6161b97741d596ff0wC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$anonfun$1.apply(:110) at org.apache.spark.sql.catalyst.expressions.ScalaUDF$$anonfun$2.apply(ScalaUDF.scala:75) at org.apache.spark.sql.catalyst.expressions.ScalaUDF$$anonfun$2.apply(ScalaUDF.scala:74) at org.apache.spark.sql.catalyst.expressions.ScalaUDF.eval(ScalaUDF.scala:964) at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificMutableProjection.apply(Unknown Source) at org.apache.spark.sql.execution.Project$$anonfun$1$$anonfun$apply$2.apply(basicOperators.scala:55) at org.apache.spark.sql.execution.Project$$anonfun$1$$anonfun$apply$2.apply(basicOperators.scala:53) at scala.collection.Iterator$$anon$11.next(Iterator.scala:328) at scala.collection.Iterator$$anon$11.next(Iterator.scala:328) at scala.collection.Iterator$$anon$10.next(Iterator.scala:312) at scala.collection.Iterator$class.foreach(Iterator.scala:727) at scala.collection.AbstractIterator.foreach(Iterator.scala:1157) at scala.collection.generic.Growable$class.$plus$plus$eq(Growable.scala:48) at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:103) at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:47) at scala.collection.TraversableOnce$class.to(TraversableOnce.scala:273) at scala.collection.AbstractIterator.to(Iterator.scala:1157) at scala.collection.TraversableOnce$class.toBuffer(TraversableOnce.scala:265) at scala.collection.AbstractIterator.toBuffer(Iterator.scala:1157) at scala.collection.TraversableOnce$class.toArray(TraversableOnce.scala:252) at scala.collection.AbstractIterator.toArray(Iterator.scala:1157) at org.apache.spark.sql.execution.SparkPlan$$anonfun$5.apply(SparkPlan.scala:215) at org.apache.spark.sql.execution.SparkPlan$$anonfun$5.apply(SparkPlan.scala:215) at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1848) at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1848) at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:66) at org.apache.spark.scheduler.Task.run(Task.scala:88) at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:214) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1177) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:642) at java.lang.Thread.run(Thread.java:857) was: case where this works: val TestDoc1 = sqlContext.createDataFrame(Seq(("sachin aggarwal", "1"), ("Rishabh", "2"))).toDF("myText", "id") TestDoc1.select(callUDF("mydef",struct($"myText".as("Text"),$"id".as("label"))).as("col1")).show steps to reproduce: 1)create a file copy following text--filename(a.json) { "myText": "Sachin Aggarwal",
[jira] [Updated] (SPARK-12117) Column Aliases are Ignored in callUDF while using struct()
[ https://issues.apache.org/jira/browse/SPARK-12117?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] sachin aggarwal updated SPARK-12117: Description: case where this works: val TestDoc1 = sqlContext.createDataFrame(Seq(("sachin aggarwal", "1"), ("Rishabh", "2"))).toDF("myText", "id") TestDoc1.select(callUDF("mydef",struct($"myText".as("Text"),$"id".as("label"))).as("col1")).show steps to reproduce: 1)create a file copy following text--filename(a.json) { "myText": "Sachin Aggarwal","id": "1"} { "myText": "Rishabh","id": "2"} 2)define a simple UDF def mydef(r:Row)={println(r.schema); r.getAs("Text").asInstanceOf[String]} 3)register the udf sqlContext.udf.register("mydef" ,mydef _) 4)read the input file val TestDoc2=sqlContext.read.json("/tmp/a.json") 5)make a call to UDF TestDoc2.select(callUDF("mydef",struct($"myText".as("Text"),$"id".as("label"))).as("col1")).explain(true) ERROR received: java.lang.IllegalArgumentException: Field "Text" does not exist. at org.apache.spark.sql.types.StructType$$anonfun$fieldIndex$1.apply(StructType.scala:234) at org.apache.spark.sql.types.StructType$$anonfun$fieldIndex$1.apply(StructType.scala:234) at scala.collection.MapLike$class.getOrElse(MapLike.scala:128) at scala.collection.AbstractMap.getOrElse(Map.scala:58) at org.apache.spark.sql.types.StructType.fieldIndex(StructType.scala:233) at org.apache.spark.sql.catalyst.expressions.GenericRowWithSchema.fieldIndex(rows.scala:212) at org.apache.spark.sql.Row$class.getAs(Row.scala:325) at org.apache.spark.sql.catalyst.expressions.GenericRow.getAs(rows.scala:191) at $line414.$read$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$c57ec8bf9b0d5f6161b97741d596ff0wC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC.mydef(:107) at $line419.$read$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$c57ec8bf9b0d5f6161b97741d596ff0wC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$anonfun$1.apply(:110) at $line419.$read$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$c57ec8bf9b0d5f6161b97741d596ff0wC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$anonfun$1.apply(:110) at org.apache.spark.sql.catalyst.expressions.ScalaUDF$$anonfun$2.apply(ScalaUDF.scala:75) at org.apache.spark.sql.catalyst.expressions.ScalaUDF$$anonfun$2.apply(ScalaUDF.scala:74) at org.apache.spark.sql.catalyst.expressions.ScalaUDF.eval(ScalaUDF.scala:964) at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificMutableProjection.apply(Unknown Source) at org.apache.spark.sql.execution.Project$$anonfun$1$$anonfun$apply$2.apply(basicOperators.scala:55) at org.apache.spark.sql.execution.Project$$anonfun$1$$anonfun$apply$2.apply(basicOperators.scala:53) at scala.collection.Iterator$$anon$11.next(Iterator.scala:328) at scala.collection.Iterator$$anon$11.next(Iterator.scala:328) at scala.collection.Iterator$$anon$10.next(Iterator.scala:312) at scala.collection.Iterator$class.foreach(Iterator.scala:727) at scala.collection.AbstractIterator.foreach(Iterator.scala:1157) at scala.collection.generic.Growable$class.$plus$plus$eq(Growable.scala:48) at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:103) at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:47) at scala.collection.TraversableOnce$class.to(TraversableOnce.scala:273) at scala.collection.AbstractIterator.to(Iterator.scala:1157) at scala.collection.TraversableOnce$class.toBuffer(TraversableOnce.scala:265) at scala.collection.AbstractIterator.toBuffer(Iterator.scala:1157) at scala.collection.TraversableOnce$class.toArray(TraversableOnce.scala:252) at scala.collection.AbstractIterator.toArray(Iterator.scala:1157) at org.apache.spark.sql.execution.SparkPlan$$anonfun$5.apply(SparkPlan.scala:215) at org.apache.spark.sql.execution.SparkPlan$$anonfun$5.apply(SparkPlan.scala:215) at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1848) at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1848) at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:66) at org.apache.spark.scheduler.Task.run(Task.scala:88) at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:214) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1177) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:642) at java.lang.Thread.run(Thread.java:857) was: case where this works: val TestDoc1 = sqlContext.createDataFrame(Seq(("sachin aggarwal", "1"), ("Rishabh", "2"))).toDF("myText", "id") TestDoc1.select(callUDF("mydef",struct($"myText".as("Text"),$"id".as("label"))).as("col1")).show steps to reproduce: 1)create a file copy following text--filename(a.json) { "myText": "Sachin Aggarwal",
[jira] [Updated] (SPARK-12117) Column Aliases are Ignored in callUDF while using struct()
[ https://issues.apache.org/jira/browse/SPARK-12117?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] sachin aggarwal updated SPARK-12117: Description: case where this works: val TestDoc1 = sqlContext.createDataFrame(Seq(("sachin aggarwal", "1"), ("Rishabh", "2"))).toDF("myText", "id") TestDoc1.select(callUDF("mydef",struct($"myText".as("Text"),$"id".as("label"))).as("col1")).show steps to reproduce: 1)create a file copy following text--filename(a.json) { "myText": "Sachin Aggarwal","id": "1"} { "myText": "Rishabh","id": "2"} 2)define a simple UDF def mydef(r:Row)={println(r.schema); r.getAs("Text").asInstanceOf[String]} 3)register the udf sqlContext.udf.register("mydef" ,mydef _) 4)read the input file val TestDoc2=sqlContext.read.json("/tmp/a.json") 5)make a call to UDF TestDoc2.select(callUDF("mydef",struct($"myText".as("Text"),$"id".as("label"))).as("col1")).explain(true) ERROR received: java.lang.IllegalArgumentException: Field "Text" does not exist. at org.apache.spark.sql.types.StructType$$anonfun$fieldIndex$1.apply(StructType.scala:234) at org.apache.spark.sql.types.StructType$$anonfun$fieldIndex$1.apply(StructType.scala:234) at scala.collection.MapLike$class.getOrElse(MapLike.scala:128) at scala.collection.AbstractMap.getOrElse(Map.scala:58) at org.apache.spark.sql.types.StructType.fieldIndex(StructType.scala:233) at org.apache.spark.sql.catalyst.expressions.GenericRowWithSchema.fieldIndex(rows.scala:212) at org.apache.spark.sql.Row$class.getAs(Row.scala:325) at org.apache.spark.sql.catalyst.expressions.GenericRow.getAs(rows.scala:191) at $line414.$read$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$c57ec8bf9b0d5f6161b97741d596ff0wC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC.mydef(:107) at $line419.$read$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$c57ec8bf9b0d5f6161b97741d596ff0wC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$anonfun$1.apply(:110) at $line419.$read$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$c57ec8bf9b0d5f6161b97741d596ff0wC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$anonfun$1.apply(:110) at org.apache.spark.sql.catalyst.expressions.ScalaUDF$$anonfun$2.apply(ScalaUDF.scala:75) at org.apache.spark.sql.catalyst.expressions.ScalaUDF$$anonfun$2.apply(ScalaUDF.scala:74) at org.apache.spark.sql.catalyst.expressions.ScalaUDF.eval(ScalaUDF.scala:964) at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificMutableProjection.apply(Unknown Source) at org.apache.spark.sql.execution.Project$$anonfun$1$$anonfun$apply$2.apply(basicOperators.scala:55) at org.apache.spark.sql.execution.Project$$anonfun$1$$anonfun$apply$2.apply(basicOperators.scala:53) at scala.collection.Iterator$$anon$11.next(Iterator.scala:328) at scala.collection.Iterator$$anon$11.next(Iterator.scala:328) at scala.collection.Iterator$$anon$10.next(Iterator.scala:312) at scala.collection.Iterator$class.foreach(Iterator.scala:727) at scala.collection.AbstractIterator.foreach(Iterator.scala:1157) at scala.collection.generic.Growable$class.$plus$plus$eq(Growable.scala:48) at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:103) at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:47) at scala.collection.TraversableOnce$class.to(TraversableOnce.scala:273) at scala.collection.AbstractIterator.to(Iterator.scala:1157) at scala.collection.TraversableOnce$class.toBuffer(TraversableOnce.scala:265) at scala.collection.AbstractIterator.toBuffer(Iterator.scala:1157) at scala.collection.TraversableOnce$class.toArray(TraversableOnce.scala:252) at scala.collection.AbstractIterator.toArray(Iterator.scala:1157) at org.apache.spark.sql.execution.SparkPlan$$anonfun$5.apply(SparkPlan.scala:215) at org.apache.spark.sql.execution.SparkPlan$$anonfun$5.apply(SparkPlan.scala:215) at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1848) at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1848) at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:66) at org.apache.spark.scheduler.Task.run(Task.scala:88) at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:214) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1177) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:642) at java.lang.Thread.run(Thread.java:857) was: case where this works: val TestDoc1 = sqlContext.createDataFrame(Seq(("sachin aggarwal", "1"), ("Rishabh", "2"))).toDF("myText", "id") TestDoc1.select(callUDF("mydef",struct($"myText".as("Text"),$"id".as("label"))).as("col1")).show steps to reproduce: 1)create a file copy following text--filename(a.json) { "myText": "Sachin Aggarwal","i
[jira] [Updated] (SPARK-12117) Column Aliases are Ignored in callUDF while using struct()
[ https://issues.apache.org/jira/browse/SPARK-12117?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] sachin aggarwal updated SPARK-12117: Description: case where this works: val TestDoc1 = sqlContext.createDataFrame(Seq(("sachin aggarwal", "1"), ("Rishabh", "2"))).toDF("myText", "id") TestDoc1.select(callUDF("mydef",struct($"myText".as("Text"),$"id".as("label"))).as("col1")).show steps to reproduce: 1)create a file copy following text--filename(a.json) { "myText": "Sachin Aggarwal","id": "1"} { "myText": "Rishabh","id": "2"} 2)define a simple UDF def mydef(r:Row)={println(r.schema); r.getAs("Text").asInstanceOf[String]} 3)register the udf sqlContext.udf.register("mydef" ,mydef _) 4)read the input file val TestDoc2=sqlContext.read.json("/tmp/a.json") 5)make a call to UDF TestDoc2.select(callUDF("mydef",struct($"myText".as("Text"),$"id".as("label"))).as("col1")).explain(true) ERROR received: java.lang.IllegalArgumentException: Field "Text" does not exist. at org.apache.spark.sql.types.StructType$$anonfun$fieldIndex$1.apply(StructType.scala:234) at org.apache.spark.sql.types.StructType$$anonfun$fieldIndex$1.apply(StructType.scala:234) at scala.collection.MapLike$class.getOrElse(MapLike.scala:128) at scala.collection.AbstractMap.getOrElse(Map.scala:58) at org.apache.spark.sql.types.StructType.fieldIndex(StructType.scala:233) at org.apache.spark.sql.catalyst.expressions.GenericRowWithSchema.fieldIndex(rows.scala:212) at org.apache.spark.sql.Row$class.getAs(Row.scala:325) at org.apache.spark.sql.catalyst.expressions.GenericRow.getAs(rows.scala:191) at $line414.$read$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$c57ec8bf9b0d5f6161b97741d596ff0wC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC.mydef(:107) at $line419.$read$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$c57ec8bf9b0d5f6161b97741d596ff0wC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$anonfun$1.apply(:110) at $line419.$read$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$c57ec8bf9b0d5f6161b97741d596ff0wC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$anonfun$1.apply(:110) at org.apache.spark.sql.catalyst.expressions.ScalaUDF$$anonfun$2.apply(ScalaUDF.scala:75) at org.apache.spark.sql.catalyst.expressions.ScalaUDF$$anonfun$2.apply(ScalaUDF.scala:74) at org.apache.spark.sql.catalyst.expressions.ScalaUDF.eval(ScalaUDF.scala:964) at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificMutableProjection.apply(Unknown Source) at org.apache.spark.sql.execution.Project$$anonfun$1$$anonfun$apply$2.apply(basicOperators.scala:55) at org.apache.spark.sql.execution.Project$$anonfun$1$$anonfun$apply$2.apply(basicOperators.scala:53) at scala.collection.Iterator$$anon$11.next(Iterator.scala:328) at scala.collection.Iterator$$anon$11.next(Iterator.scala:328) at scala.collection.Iterator$$anon$10.next(Iterator.scala:312) at scala.collection.Iterator$class.foreach(Iterator.scala:727) at scala.collection.AbstractIterator.foreach(Iterator.scala:1157) at scala.collection.generic.Growable$class.$plus$plus$eq(Growable.scala:48) at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:103) at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:47) at scala.collection.TraversableOnce$class.to(TraversableOnce.scala:273) at scala.collection.AbstractIterator.to(Iterator.scala:1157) at scala.collection.TraversableOnce$class.toBuffer(TraversableOnce.scala:265) at scala.collection.AbstractIterator.toBuffer(Iterator.scala:1157) at scala.collection.TraversableOnce$class.toArray(TraversableOnce.scala:252) at scala.collection.AbstractIterator.toArray(Iterator.scala:1157) at org.apache.spark.sql.execution.SparkPlan$$anonfun$5.apply(SparkPlan.scala:215) at org.apache.spark.sql.execution.SparkPlan$$anonfun$5.apply(SparkPlan.scala:215) at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1848) at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1848) at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:66) at org.apache.spark.scheduler.Task.run(Task.scala:88) at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:214) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1177) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:642) at java.lang.Thread.run(Thread.java:857) was: case where this works: val TestDoc1 = sqlContext.createDataFrame(Seq(("sachin aggarwal", "1"), ("Rishabh", "2"))).toDF("myText", "id") TestDoc1.select(callUDF("mydef",struct($"myText".as("Text"),$"id".as("label"))).as("col1")).show steps to reproduce: 1)create a file copy following text--filename(a.json) { "myText": "Mauricio A. Hernandez", "id
[jira] [Created] (SPARK-12117) Column Aliases are Ignored in callUDF while using struct()
sachin aggarwal created SPARK-12117: --- Summary: Column Aliases are Ignored in callUDF while using struct() Key: SPARK-12117 URL: https://issues.apache.org/jira/browse/SPARK-12117 Project: Spark Issue Type: Bug Components: SQL Affects Versions: 1.5.1 Reporter: sachin aggarwal case where this works: val TestDoc1 = sqlContext.createDataFrame(Seq(("sachin aggarwal", "1"), ("Rishabh", "2"))).toDF("myText", "id") TestDoc1.select(callUDF("mydef",struct($"myText".as("Text"),$"id".as("label"))).as("col1")).show steps to reproduce: 1)create a file copy following text--filename(a.json) { "myText": "Mauricio A. Hernandez", "id": "1"} { "myText": "Popa, Lucian", "id": "2"} 2)define a simple UDF def mydef(r:Row)={println(r.schema); r.getAs("Text").asInstanceOf[String]} 3)register the udf sqlContext.udf.register("mydef" ,mydef _) 4)read the input file val TestDoc2=sqlContext.read.json("/tmp/a.json") 5)make a call to UDF TestDoc2.select(callUDF("mydef",struct($"myText".as("Text"),$"id".as("label"))).as("col1")).explain(true) ERROR received: java.lang.IllegalArgumentException: Field "Text" does not exist. at org.apache.spark.sql.types.StructType$$anonfun$fieldIndex$1.apply(StructType.scala:234) at org.apache.spark.sql.types.StructType$$anonfun$fieldIndex$1.apply(StructType.scala:234) at scala.collection.MapLike$class.getOrElse(MapLike.scala:128) at scala.collection.AbstractMap.getOrElse(Map.scala:58) at org.apache.spark.sql.types.StructType.fieldIndex(StructType.scala:233) at org.apache.spark.sql.catalyst.expressions.GenericRowWithSchema.fieldIndex(rows.scala:212) at org.apache.spark.sql.Row$class.getAs(Row.scala:325) at org.apache.spark.sql.catalyst.expressions.GenericRow.getAs(rows.scala:191) at $line414.$read$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$c57ec8bf9b0d5f6161b97741d596ff0wC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC.mydef(:107) at $line419.$read$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$c57ec8bf9b0d5f6161b97741d596ff0wC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$anonfun$1.apply(:110) at $line419.$read$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$c57ec8bf9b0d5f6161b97741d596ff0wC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$iwC$$anonfun$1.apply(:110) at org.apache.spark.sql.catalyst.expressions.ScalaUDF$$anonfun$2.apply(ScalaUDF.scala:75) at org.apache.spark.sql.catalyst.expressions.ScalaUDF$$anonfun$2.apply(ScalaUDF.scala:74) at org.apache.spark.sql.catalyst.expressions.ScalaUDF.eval(ScalaUDF.scala:964) at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificMutableProjection.apply(Unknown Source) at org.apache.spark.sql.execution.Project$$anonfun$1$$anonfun$apply$2.apply(basicOperators.scala:55) at org.apache.spark.sql.execution.Project$$anonfun$1$$anonfun$apply$2.apply(basicOperators.scala:53) at scala.collection.Iterator$$anon$11.next(Iterator.scala:328) at scala.collection.Iterator$$anon$11.next(Iterator.scala:328) at scala.collection.Iterator$$anon$10.next(Iterator.scala:312) at scala.collection.Iterator$class.foreach(Iterator.scala:727) at scala.collection.AbstractIterator.foreach(Iterator.scala:1157) at scala.collection.generic.Growable$class.$plus$plus$eq(Growable.scala:48) at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:103) at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:47) at scala.collection.TraversableOnce$class.to(TraversableOnce.scala:273) at scala.collection.AbstractIterator.to(Iterator.scala:1157) at scala.collection.TraversableOnce$class.toBuffer(TraversableOnce.scala:265) at scala.collection.AbstractIterator.toBuffer(Iterator.scala:1157) at scala.collection.TraversableOnce$class.toArray(TraversableOnce.scala:252) at scala.collection.AbstractIterator.toArray(Iterator.scala:1157) at org.apache.spark.sql.execution.SparkPlan$$anonfun$5.apply(SparkPlan.scala:215) at org.apache.spark.sql.execution.SparkPlan$$anonfun$5.apply(SparkPlan.scala:215) at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1848) at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1848) at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:66) at org.apache.spark.scheduler.Task.run(Task.scala:88) at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:214) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1177) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:642) at java.lang.Thread.run(Thread.java:857) -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues
[jira] [Commented] (SPARK-11552) Replace example code in ml-decision-tree.md using include_example
[ https://issues.apache.org/jira/browse/SPARK-11552?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14993256#comment-14993256 ] sachin aggarwal commented on SPARK-11552: - I will start with this > Replace example code in ml-decision-tree.md using include_example > - > > Key: SPARK-11552 > URL: https://issues.apache.org/jira/browse/SPARK-11552 > Project: Spark > Issue Type: Sub-task > Components: Documentation >Reporter: Xusen Yin > Labels: starter > -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org