[jira] [Commented] (SPARK-7942) Receiver's life cycle is inconsistent with streaming job.
[ https://issues.apache.org/jira/browse/SPARK-7942?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=14741800#comment-14741800 ] Tathagata Das commented on SPARK-7942: -- This has been resolved in Spark 1.5.0 where we reimplemented the receiver scheduling to keep relaunching and not be limited by the max task retries > Receiver's life cycle is inconsistent with streaming job. > - > > Key: SPARK-7942 > URL: https://issues.apache.org/jira/browse/SPARK-7942 > Project: Spark > Issue Type: Bug > Components: Streaming >Affects Versions: 1.4.0 >Reporter: SaintBacchus > > Streaming consider the receiver as a common spark job, thus if an error > occurs in the receiver's logical(after 4 times(default) retries ), streaming > will no longer get any data but the streaming job is still running. > A general scenario is that: we config the > `spark.streaming.receiver.writeAheadLog.enable` as true to use the > `ReliableKafkaReceiver` but do not set the checkpoint dir. Then the receiver > will soon be shut down but the streaming is alive. -- 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-7942) Receiver's life cycle is inconsistent with streaming job.
[ https://issues.apache.org/jira/browse/SPARK-7942?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14575483#comment-14575483 ] Tathagata Das commented on SPARK-7942: -- At the very least, we should throw route an exception to the streamingcontext (so that awaitTermination throws that exception) when ALL of them have exited. That is strictly better than what it is right now. Receiver's life cycle is inconsistent with streaming job. - Key: SPARK-7942 URL: https://issues.apache.org/jira/browse/SPARK-7942 Project: Spark Issue Type: Bug Components: Streaming Affects Versions: 1.4.0 Reporter: SaintBacchus Streaming consider the receiver as a common spark job, thus if an error occurs in the receiver's logical(after 4 times(default) retries ), streaming will no longer get any data but the streaming job is still running. A general scenario is that: we config the `spark.streaming.receiver.writeAheadLog.enable` as true to use the `ReliableKafkaReceiver` but do not set the checkpoint dir. Then the receiver will soon be shut down but the streaming is alive. -- 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-7942) Receiver's life cycle is inconsistent with streaming job.
[ https://issues.apache.org/jira/browse/SPARK-7942?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14572212#comment-14572212 ] SaintBacchus commented on SPARK-7942: - [~tdas] Now I'm not clear about how to deal with only some of the receivers had broken down, should we shutdown the StreamingContext ? Or ignore this and leave the alive receivers still running Receiver's life cycle is inconsistent with streaming job. - Key: SPARK-7942 URL: https://issues.apache.org/jira/browse/SPARK-7942 Project: Spark Issue Type: Bug Components: Streaming Affects Versions: 1.4.0 Reporter: SaintBacchus Streaming consider the receiver as a common spark job, thus if an error occurs in the receiver's logical(after 4 times(default) retries ), streaming will no longer get any data but the streaming job is still running. A general scenario is that: we config the `spark.streaming.receiver.writeAheadLog.enable` as true to use the `ReliableKafkaReceiver` but do not set the checkpoint dir. Then the receiver will soon be shut down but the streaming is alive. -- 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-7942) Receiver's life cycle is inconsistent with streaming job.
[ https://issues.apache.org/jira/browse/SPARK-7942?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14566222#comment-14566222 ] Tathagata Das commented on SPARK-7942: -- That is a very good idea. In fact please update the JIRA title to describe that feature. If there were receivers started, and all the receivers have shutdown, then stop the StreamingContext and throw error such that ssc.awaitTermination exits. This will be a good feature to add. Receiver's life cycle is inconsistent with streaming job. - Key: SPARK-7942 URL: https://issues.apache.org/jira/browse/SPARK-7942 Project: Spark Issue Type: Bug Components: Streaming Affects Versions: 1.4.0 Reporter: SaintBacchus Streaming consider the receiver as a common spark job, thus if an error occurs in the receiver's logical(after 4 times(default) retries ), streaming will no longer get any data but the streaming job is still running. A general scenario is that: we config the `spark.streaming.receiver.writeAheadLog.enable` as true to use the `ReliableKafkaReceiver` but do not set the checkpoint dir. Then the receiver will soon be shut down but the streaming is alive. -- 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-7942) Receiver's life cycle is inconsistent with streaming job.
[ https://issues.apache.org/jira/browse/SPARK-7942?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14564411#comment-14564411 ] SaintBacchus commented on SPARK-7942: - [~tdas] Should we add a logic to shut down the StreamingContext if the Receiver had been down? Receiver's life cycle is inconsistent with streaming job. - Key: SPARK-7942 URL: https://issues.apache.org/jira/browse/SPARK-7942 Project: Spark Issue Type: Bug Components: Streaming Affects Versions: 1.4.0 Reporter: SaintBacchus Streaming consider the receiver as a common spark job, thus if an error occurs in the receiver's logical(after 4 times(default) retries ), streaming will no longer get any data but the streaming job is still running. A general scenario is that: we config the `spark.streaming.receiver.writeAheadLog.enable` as true to use the `ReliableKafkaReceiver` but do not set the checkpoint dir. Then the receiver will soon be shut down but the streaming is alive. -- 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