> On 13 Jan 2016, at 03:23, Maciej Bryński <mac...@brynski.pl> wrote: > > Thanks. > I successfully compiled Spark 1.6.0 with Jackson 2.2.3 from source. > > I'll try to using it. >
This is the eternal classpath version problem, with Jackson turning out to be incredibly brittle. After one point update of the 1.x JAR broke things (it removed a method), there's ~0 enthusiasm for incrementing the version counters again. 1. Hadoop, even Hadoop trunk, is built with : <jackson2.version>2.2.3</jackson2.version> this means that right now, the Spark 1.6 shuffle isn't going to work in *any* Hadoop cluster that hasn't been built with a compatible Jackson version: either rebuild spark 1.6 or hadoop-core itself. 2. There's a YARN JIRA for classpath isolation in aux services: https://issues.apache.org/jira/browse/YARN-4577 , with a longer term one for forking off the services entirely: https://issues.apache.org/jira/browse/YARN-1593 . there's a patch for the first one, which, if someone want's to apply and test locally, would be valued. It wouldn't ship in Hadoop until 2.9. . Patch #2 got assigned to someone last week, so maybe it'll surface in Hadoop 2.9 instead/as well. 3. I'm going to open a SPARK JIRA here, cross link it to the YARN ones -so at least there'll be a central record. (Done: SPARK-12807) 4. I'll also add an "upgrade jackson" issue under HADOOP-9991, though like I said: enthusiasm will be low. 5. You can D/L a version of spark 1.6 built against HDP 2.3: http://hortonworks.com/hadoop-tutorial/apache-spark-1-6-technical-preview-with-hdp-2-3/ This isn't likely to work against HDP 2.2 BTW; later Hadoop JAR versions. -I suspect for things to work on CDH there'll be something similar. For ASF Hadoop, rebuilding spark is all you have. 6. Looking @spark/master, it's been on jackson 2.5.3 since last month, from SPARK-12269. Which is just going to make versioning even more traumatic. And we know that amazon-aws has a back compatibility track record, so swapping things around there is going to be fun. You'd probably need to rebuild Hadoop-2.7.2+ with the later aws/s3 JARs to keep everything aligned on #4: has anyone found any compatibility problems if they swap out Jackson 2.2.3 for Jackson 2.4.4 or 2.5.3 *without recompiling anything*? That's what we need to know for Hadoop JAR updates. -Steve > 2016-01-13 11:25 GMT+01:00 Ted Yu <yuzhih...@gmail.com>: >> I would suggest trying option #1 first. >> >> Thanks >> >>> On Jan 13, 2016, at 2:12 AM, Maciej Bryński <mac...@brynski.pl> wrote: >>> >>> Hi, >>> I/m trying to run Spark 1.6.0 on HDP 2.2 >>> Everything was fine until I tried to turn on dynamic allocation. >>> According to instruction I need to add shuffle service to yarn classpath. >>> The problem is that HDP 2.2 has jackson 2.2.3 and Spark is using 2.4.4. >>> So connecting it gives error: >>> >>> 2016-01-11 16:56:51,222 INFO containermanager.AuxServices >>> (AuxServices.java:addService(72)) - Adding auxiliary service >>> spark_shuffle, "spark_shuffle" >>> 2016-01-11 16:56:51,439 FATAL nodemanager.NodeManager >>> (NodeManager.java:initAndStartNodeManager(465)) - Error starting >>> NodeManager >>> java.lang.NoSuchMethodError: >>> com.fasterxml.jackson.core.JsonFactory.requiresPropertyOrdering()Z >>> at >>> com.fasterxml.jackson.databind.ObjectMapper.<init>(ObjectMapper.java:457) >>> at >>> com.fasterxml.jackson.databind.ObjectMapper.<init>(ObjectMapper.java:379) >>> at >>> org.apache.spark.network.shuffle.ExternalShuffleBlockResolver.<clinit>(ExternalShuffleBlockResolver.java:57) >>> at >>> org.apache.spark.network.shuffle.ExternalShuffleBlockHandler.<init>(ExternalShuffleBlockHandler.java:56) >>> at >>> org.apache.spark.network.yarn.YarnShuffleService.serviceInit(YarnShuffleService.java:128) >>> at >>> org.apache.hadoop.service.AbstractService.init(AbstractService.java:163) >>> at >>> org.apache.hadoop.yarn.server.nodemanager.containermanager.AuxServices.serviceInit(AuxServices.java:143) >>> at >>> org.apache.hadoop.service.AbstractService.init(AbstractService.java:163) >>> at >>> org.apache.hadoop.service.CompositeService.serviceInit(CompositeService.java:107) >>> at >>> org.apache.hadoop.yarn.server.nodemanager.containermanager.ContainerManagerImpl.serviceInit(ContainerManagerImpl.java:237) >>> at >>> org.apache.hadoop.service.AbstractService.init(AbstractService.java:163) >>> at >>> org.apache.hadoop.service.CompositeService.serviceInit(CompositeService.java:107) >>> at >>> org.apache.hadoop.yarn.server.nodemanager.NodeManager.serviceInit(NodeManager.java:253) >>> at >>> org.apache.hadoop.service.AbstractService.init(AbstractService.java:163) >>> at >>> org.apache.hadoop.yarn.server.nodemanager.NodeManager.initAndStartNodeManager(NodeManager.java:462) >>> at >>> org.apache.hadoop.yarn.server.nodemanager.NodeManager.main(NodeManager.java:509) >>> >>> >>> What can I do ? >>> I have following ideas: >>> 1) Compile Spark 1.6.0 with modified pom.xml (change jackson version >>> to 2.2.3). I'm not sure if this will be working >>> 2) I tried to put shuffle service from different version of Spark. >>> 1.4.1 works on HDP 2.2. >>> Is it possible to run shuffle service from 1.4.1 with Spark 1.6.0 ? >>> 3) Other ideas ? >>> >>> Regards, >>> -- >>> Maciek Bryński >>> >>> --------------------------------------------------------------------- >>> To unsubscribe, e-mail: dev-unsubscr...@spark.apache.org >>> For additional commands, e-mail: dev-h...@spark.apache.org >>> > > > > -- > Maciek Bryński > > --------------------------------------------------------------------- > To unsubscribe, e-mail: dev-unsubscr...@spark.apache.org > For additional commands, e-mail: dev-h...@spark.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@spark.apache.org For additional commands, e-mail: dev-h...@spark.apache.org
