[jira] [Commented] (SPARK-26907) Does ShuffledRDD Replication Work With External Shuffle Service
[ https://issues.apache.org/jira/browse/SPARK-26907?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16776150#comment-16776150 ] Han Altae-Tran commented on SPARK-26907: Ok, thank you I will try the mailing list. I think my main point is that Spark could be improved for use with preemptible virtual machines if shuffle files can be replicated across the cluster. In my experience, whenever there is a shuffle map task, a single node being preempted can cause the entire stage to be retried, causing a huge loss of uptime as all tasks fail until the retry is initiated. Using persist with replication doesn't seem to help this issue, so I figured there is an optimization around shuffle files that could be made for this use case. > Does ShuffledRDD Replication Work With External Shuffle Service > --- > > Key: SPARK-26907 > URL: https://issues.apache.org/jira/browse/SPARK-26907 > Project: Spark > Issue Type: Question > Components: Block Manager, YARN >Affects Versions: 2.3.2 >Reporter: Han Altae-Tran >Priority: Major > > I am interested in working with high replication environments for extreme > fault tolerance (e.g. 10x replication), but have noticed that when using > groupBy or groupWith followed by persist (with 10x replication), even if one > node fails, the entire stage can fail with FetchFailedException. > > Is this because the External Shuffle Service writes and services intermediate > shuffle data only to/from the local disk attached to the executor that > generated it, causing spark to ignore possible replicated shuffle data (from > the persist) that may be serviced elsewhere? If so, is there any way to > increase the replication factor of the External Shuffle Service to make it > fault tolerant? -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Created] (SPARK-26907) Does ShuffledRDD Replication Work With External Shuffle Service
Han Altae-Tran created SPARK-26907: -- Summary: Does ShuffledRDD Replication Work With External Shuffle Service Key: SPARK-26907 URL: https://issues.apache.org/jira/browse/SPARK-26907 Project: Spark Issue Type: Question Components: Block Manager, YARN Affects Versions: 2.3.2 Reporter: Han Altae-Tran I am interested in working with high replication environments for extreme fault tolerance (e.g. 10x replication), but have noticed that when using groupBy or groupWith followed by persist (with 10x replication), even if one node fails, the entire stage can fail with FetchFailedException. Is this because the External Shuffle Service writes and services intermediate shuffle data only to/from the local disk attached to the executor that generated it, causing spark to ignore possible replicated shuffle data (from the persist) that may be serviced elsewhere? If so, is there any way to increase the replication factor of the External Shuffle Service to make it fault tolerant? -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Updated] (SPARK-26906) Pyspark RDD Replication Potentially Not Working
[ https://issues.apache.org/jira/browse/SPARK-26906?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Han Altae-Tran updated SPARK-26906: --- Attachment: spark_ui.png > Pyspark RDD Replication Potentially Not Working > --- > > Key: SPARK-26906 > URL: https://issues.apache.org/jira/browse/SPARK-26906 > Project: Spark > Issue Type: Bug > Components: PySpark, Web UI >Affects Versions: 2.3.2 > Environment: I am using Google Cloud's Dataproc version [1.3.19-deb9 > 2018/12/14|https://cloud.google.com/dataproc/docs/release-notes#december_14_2018] > (version 2.3.2 Spark and version 2.9.0 Hadoop) with version Debian 9, with > python version 3.7. PySpark shell is activated using pyspark --num-executors > = 100 >Reporter: Han Altae-Tran >Priority: Minor > Attachments: spark_ui.png > > > Pyspark RDD replication doesn't seem to be functioning properly. Even with a > simple example, the UI reports only 1x replication, despite using the flag > for 2x replication > {code:java} > rdd = sc.range(10**9) > mapped = rdd.map(lambda x: x) > mapped.persist(pyspark.StorageLevel.DISK_ONLY_2) \\ PythonRDD[1] at RDD at > PythonRDD.scala:52 > mapped.count(){code} > > Interestingly, if you catch the UI page at just the right time, you see that > it starts off 2x replicated, but ends up 1x replicated afterward. Perhaps the > RDD is replicated, but it is just the UI that is unable to register this. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Updated] (SPARK-26906) Pyspark RDD Replication Potentially Not Working
[ https://issues.apache.org/jira/browse/SPARK-26906?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Han Altae-Tran updated SPARK-26906: --- Priority: Minor (was: Major) Description: Pyspark RDD replication doesn't seem to be functioning properly. Even with a simple example, the UI reports only 1x replication, despite using the flag for 2x replication {code:java} rdd = sc.range(10**9) mapped = rdd.map(lambda x: x) mapped.persist(pyspark.StorageLevel.DISK_ONLY_2) \\ PythonRDD[1] at RDD at PythonRDD.scala:52 mapped.count(){code} Interestingly, if you catch the UI page at just the right time, you see that it starts off 2x replicated, but ends up 1x replicated afterward. Perhaps the RDD is replicated, but it is just the UI that is unable to register this. was: Pyspark RDD replication doesn't seem to be functioning properly. Even with a simple example, the UI reports only 1x replication, despite using the flag for 2x replication {code:java} rdd = sc.range(10**9) mapped = rdd.map(lambda x: x) mapped.persist(pyspark.StorageLevel.DISK_ONLY_2) \\ PythonRDD[1] at RDD at PythonRDD.scala:52 mapped.count(){code} resulting in the following: !image-2019-02-17-01-33-08-551.png! Interestingly, if you catch the UI page at just the right time, you see that it starts off 2x replicated: !image-2019-02-17-01-35-37-034.png! but ends up going back to 1x replicated once the RDD is fully materialized. This is likely not a UI bug because the cached partitions page also shows only 1x replication: !image-2019-02-17-01-36-55-418.png! This could result from some type of optimization for replication, but is undesirable for users that want a specific level of replication for fault tolerance. Summary: Pyspark RDD Replication Potentially Not Working (was: Pyspark RDD Replication Not Working) > Pyspark RDD Replication Potentially Not Working > --- > > Key: SPARK-26906 > URL: https://issues.apache.org/jira/browse/SPARK-26906 > Project: Spark > Issue Type: Bug > Components: PySpark, Web UI >Affects Versions: 2.3.2 > Environment: I am using Google Cloud's Dataproc version [1.3.19-deb9 > 2018/12/14|https://cloud.google.com/dataproc/docs/release-notes#december_14_2018] > (version 2.3.2 Spark and version 2.9.0 Hadoop) with version Debian 9, with > python version 3.7. PySpark shell is activated using pyspark --num-executors > = 100 >Reporter: Han Altae-Tran >Priority: Minor > > Pyspark RDD replication doesn't seem to be functioning properly. Even with a > simple example, the UI reports only 1x replication, despite using the flag > for 2x replication > {code:java} > rdd = sc.range(10**9) > mapped = rdd.map(lambda x: x) > mapped.persist(pyspark.StorageLevel.DISK_ONLY_2) \\ PythonRDD[1] at RDD at > PythonRDD.scala:52 > mapped.count(){code} > > Interestingly, if you catch the UI page at just the right time, you see that > it starts off 2x replicated, but ends up 1x replicated afterward. Perhaps the > RDD is replicated, but it is just the UI that is unable to register this. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Created] (SPARK-26906) Pyspark RDD Replication Not Working
Han Altae-Tran created SPARK-26906: -- Summary: Pyspark RDD Replication Not Working Key: SPARK-26906 URL: https://issues.apache.org/jira/browse/SPARK-26906 Project: Spark Issue Type: Bug Components: PySpark, Web UI Affects Versions: 2.3.2 Environment: I am using Google Cloud's Dataproc version [1.3.19-deb9 2018/12/14|https://cloud.google.com/dataproc/docs/release-notes#december_14_2018] (version 2.3.2 Spark and version 2.9.0 Hadoop) with version Debian 9, with python version 3.7. PySpark shell is activated using pyspark --num-executors = 100 Reporter: Han Altae-Tran Pyspark RDD replication doesn't seem to be functioning properly. Even with a simple example, the UI reports only 1x replication, despite using the flag for 2x replication {code:java} rdd = sc.range(10**9) mapped = rdd.map(lambda x: x) mapped.persist(pyspark.StorageLevel.DISK_ONLY_2) \\ PythonRDD[1] at RDD at PythonRDD.scala:52 mapped.count(){code} resulting in the following: !image-2019-02-17-01-33-08-551.png! Interestingly, if you catch the UI page at just the right time, you see that it starts off 2x replicated: !image-2019-02-17-01-35-37-034.png! but ends up going back to 1x replicated once the RDD is fully materialized. This is likely not a UI bug because the cached partitions page also shows only 1x replication: !image-2019-02-17-01-36-55-418.png! This could result from some type of optimization for replication, but is undesirable for users that want a specific level of replication for fault tolerance. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org