[
https://issues.apache.org/jira/browse/SPARK-2445?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14129226#comment-14129226
]
Tomas Barton commented on SPARK-2445:
-------------------------------------
well, the workaround doesn't seem to work, here's output with TRACE log level:
(produced by running same job:
{code}
MASTER=mesos://`cat /etc/mesos/zk` ./bin/run-example SparkLR
{code}
from different node)
{code}
...
On iteration 5
14/09/11 00:02:45 INFO SparkContext: Starting job: reduce at SparkLR.scala:64
14/09/11 00:02:45 TRACE DAGScheduler: Checking for newly runnable parent stages
14/09/11 00:02:45 TRACE DAGScheduler: running: Set()
14/09/11 00:02:45 TRACE DAGScheduler: waiting: Set()
14/09/11 00:02:45 TRACE DAGScheduler: failed: Set()
14/09/11 00:02:45 INFO DAGScheduler: Got job 4 (reduce at SparkLR.scala:64)
with 2 output partitions (allowLocal=false)
14/09/11 00:02:45 INFO DAGScheduler: Final stage: Stage 4(reduce at
SparkLR.scala:64)
14/09/11 00:02:45 INFO DAGScheduler: Parents of final stage: List()
14/09/11 00:02:45 INFO DAGScheduler: Missing parents: List()
14/09/11 00:02:45 DEBUG DAGScheduler: submitStage(Stage 4)
14/09/11 00:02:45 DEBUG DAGScheduler: missing: List()
14/09/11 00:02:45 INFO DAGScheduler: Submitting Stage 4 (MappedRDD[5] at map at
SparkLR.scala:62), which has no missing parents
14/09/11 00:02:45 DEBUG DAGScheduler: submitMissingTasks(Stage 4)
14/09/11 00:02:45 INFO DAGScheduler: Submitting 2 missing tasks from Stage 4
(MappedRDD[5] at map at SparkLR.scala:62)
14/09/11 00:02:45 DEBUG DAGScheduler: New pending tasks: Set(ResultTask(4, 0),
ResultTask(4, 1))
14/09/11 00:02:45 INFO TaskSchedulerImpl: Adding task set 4.0 with 2 tasks
14/09/11 00:02:45 DEBUG TaskSetManager: Epoch for TaskSet 4.0: 5
14/09/11 00:02:45 DEBUG TaskSetManager: Valid locality levels for TaskSet 4.0:
PROCESS_LOCAL, NODE_LOCAL, ANY
14/09/11 00:02:45 TRACE DAGScheduler: Checking for newly runnable parent stages
14/09/11 00:02:45 TRACE DAGScheduler: running: Set(Stage 4)
14/09/11 00:02:45 TRACE DAGScheduler: waiting: Set()
14/09/11 00:02:45 TRACE DAGScheduler: failed: Set()
14/09/11 00:02:45 DEBUG TaskSchedulerImpl: parentName: , name: TaskSet_4,
runningTasks: 0
14/09/11 00:02:45 INFO TaskSetManager: Starting task 4.0:1 as TID 15 on
executor 20140910-231511-185277356-5050-425-101: 172.27.11.11 (PROCESS_LOCAL)
14/09/11 00:02:45 INFO TaskSetManager: Serialized task 4.0:1 as 667083 bytes in
18 ms
14/09/11 00:02:45 TRACE DAGScheduler: Checking for newly runnable parent stages
14/09/11 00:02:45 INFO TaskSetManager: Starting task 4.0:0 as TID 16 on
executor 20140910-231511-185277356-5050-425-102: 172.27.11.13 (PROCESS_LOCAL)
14/09/11 00:02:45 TRACE DAGScheduler: running: Set(Stage 4)
14/09/11 00:02:45 TRACE DAGScheduler: waiting: Set()
14/09/11 00:02:45 TRACE DAGScheduler: failed: Set()
14/09/11 00:02:45 INFO TaskSetManager: Serialized task 4.0:0 as 667083 bytes in
17 ms
14/09/11 00:02:45 TRACE DAGScheduler: Checking for newly runnable parent stages
14/09/11 00:02:45 TRACE DAGScheduler: running: Set(Stage 4)
14/09/11 00:02:45 TRACE DAGScheduler: waiting: Set()
14/09/11 00:02:45 TRACE DAGScheduler: failed: Set()
14/09/11 00:02:46 INFO TaskSetManager: Re-queueing tasks for
20140910-231511-185277356-5050-425-101 from TaskSet 4.0
14/09/11 00:02:46 WARN TaskSetManager: Lost TID 15 (task 4.0:1)
14/09/11 00:02:46 TRACE DAGScheduler: Checking for newly runnable parent stages
14/09/11 00:02:46 TRACE DAGScheduler: running: Set(Stage 4)
14/09/11 00:02:46 TRACE DAGScheduler: waiting: Set()
14/09/11 00:02:46 TRACE DAGScheduler: failed: Set()
14/09/11 00:02:46 INFO DAGScheduler: Executor lost:
20140910-231511-185277356-5050-425-101 (epoch 5)
14/09/11 00:02:46 INFO BlockManagerMasterActor: Trying to remove executor
20140910-231511-185277356-5050-425-101 from BlockManagerMaster.
14/09/11 00:02:46 INFO BlockManagerMaster: Removed
20140910-231511-185277356-5050-425-101 successfully in removeExecutor
14/09/11 00:02:46 DEBUG MapOutputTrackerMaster: Increasing epoch to 6
14/09/11 00:02:46 TRACE DAGScheduler: Checking for newly runnable parent stages
14/09/11 00:02:46 TRACE DAGScheduler: running: Set(Stage 4)
14/09/11 00:02:46 TRACE DAGScheduler: waiting: Set()
14/09/11 00:02:46 TRACE DAGScheduler: failed: Set()
14/09/11 00:02:46 INFO DAGScheduler: Host added was in lost list earlier:
172.27.11.11
14/09/11 00:02:46 TRACE DAGScheduler: Checking for newly runnable parent stages
14/09/11 00:02:46 TRACE DAGScheduler: running: Set(Stage 4)
14/09/11 00:02:46 TRACE DAGScheduler: waiting: Set()
14/09/11 00:02:46 DEBUG TaskSchedulerImpl: parentName: , name: TaskSet_4,
runningTasks: 1
14/09/11 00:02:46 TRACE DAGScheduler: failed: Set()
14/09/11 00:02:46 INFO TaskSetManager: Starting task 4.0:1 as TID 17 on
executor 20140910-231511-185277356-5050-425-101: 172.27.11.11 (PROCESS_LOCAL)
14/09/11 00:02:46 INFO TaskSetManager: Serialized task 4.0:1 as 667083 bytes in
15 ms
14/09/11 00:02:46 TRACE DAGScheduler: Checking for newly runnable parent stages
14/09/11 00:02:46 TRACE DAGScheduler: running: Set(Stage 4)
14/09/11 00:02:46 TRACE DAGScheduler: waiting: Set()
14/09/11 00:02:46 TRACE DAGScheduler: failed: Set()
14/09/11 00:02:47 DEBUG TaskSchedulerImpl: parentName: , name: TaskSet_4,
runningTasks: 2
14/09/11 00:02:48 ERROR BlockManagerMasterActor: Got two different block
manager registrations on 20140910-231511-185277356-5050-425-102
14/09/11 00:02:48 DEBUG DiskBlockManager: Shutdown hook called
{code}
> MesosExecutorBackend crashes in fine grained mode
> -------------------------------------------------
>
> Key: SPARK-2445
> URL: https://issues.apache.org/jira/browse/SPARK-2445
> Project: Spark
> Issue Type: Bug
> Components: Mesos
> Affects Versions: 1.0.0
> Reporter: Dario Rexin
>
> When multiple instances of the MesosExecutorBackend are running on the same
> slave, they will have the same executorId assigned (equal to the mesos
> slaveId), but will have a different port (which is randomly assigned).
> Because of this, it can not register a new BlockManager, because one is
> already registered with the same executorId, but a different BlockManagerId.
> More description and a fix can be found in this PR on GitHub:
> https://github.com/apache/spark/pull/1358
--
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
