Also changed the semantics of the index parameter in mapPartitionsWithIndex from the partition index of the output partition to the partition index in the current RDD.
AppendOnlyMap fixes
- Chose a more random reshuffling step for values returned by Object.hashCode to avoid some long chaining that was happening for consecutive integers (e.g. `sc.makeRDD(1 to 100000000, 100).map(t => (t, t)).reduceByKey(_ + _).count`)
- Some other small optimizations throughout (see commit comments)
- Don't check keys for equality when re-inserting due to growing the
table; the keys will already be unique
- Remember the grow threshold instead of recomputing it on each insert
- Use Murmur Hash 3 finalization step to scramble the bits of HashCode
instead of the simpler version in java.util.HashMap; the latter one
had trouble with ranges of consecutive integers. Murmur Hash 3 is used
by fastutil.
- Use Object.equals() instead of Scala's == to compare keys, because the
latter does extra casts for numeric types (see the equals method in
https://github.com/scala/scala/blob/master/src/library/scala/runtime/BoxesRunTime.java)
Support preservesPartitioning in RDD.zipPartitions
In `RDD.zipPartitions`, add support for a `preservesPartitioning` option (similar to `RDD.mapPartitions`) that reuses the first RDD's partitioner.
PartitionPruningRDD is using index from parent
I was getting a ArrayIndexOutOfBoundsException exception after doing union on pruned RDD. The index it was using on the partition was the index in the original RDD not the new pruned RDD.
Cleanup to remove semicolons (;) from Scala code
-) The main reason for this PR is to remove semicolons from single statements of Scala code.
-) Remove unused imports as I see them
-) Fix ASF comment header from some of files (bad copy paste I suppose)
Impove Spark on Yarn Error handling
Improve cli error handling and only allow a certain number of worker failures before failing the application. This will help prevent users from doing foolish things and their jobs running forever. For instance using 32 bit java but trying to allocate 8G containers. This loops forever without this change, now it errors out after a certain number of retries. The number of tries is configurable. Also increase the frequency we ping the RM to increase speed at which we get containers if they die. The Yarn MR app defaults to pinging the RM every 1 seconds, so the default of 5 seconds here is fine. But that is configurable as well in case people want to change it.
I do want to make sure there aren't any cases that calling stopExecutors in CoarseGrainedSchedulerBackend would cause problems? I couldn't think of any and testing on standalone cluster as well as yarn.
Simple cleanup on Spark's Scala code
Simple cleanup on Spark's Scala code while testing some modules:
-) Remove some of unused imports as I found them
-) Remove ";" in the imports statements
-) Remove () at the end of method calls like size that does not have size effect.
-) Remove some of unused imports as I found them
-) Remove ";" in the imports statements
-) Remove () at the end of method call like size that does not have size effect.
Fix bug where scheduler could hang after task failure.
When a task fails, we need to call reviveOffers() so that the
task can be rescheduled on a different machine. In the current code,
the state in ClusterTaskSetManager indicating which tasks are
pending may be updated after revive offers is called (there's a
race condition here), so when revive offers is called, the task set
manager does not yet realize that there are failed tasks that need
to be relaunched.
This isn't currently unit tested but will be once my pull request for
merging the cluster and local schedulers goes in -- at which point
many more of the unit tests will exercise the code paths through
the cluster scheduler (currently the failure test suite uses the local
scheduler, which is why we didn't see this bug before).
Don't retry tasks when they fail due to a NotSerializableException
As with my previous pull request, this will be unit tested once the Cluster and Local schedulers get merged.
When a task fails, we need to call reviveOffers() so that the
task can be rescheduled on a different machine. In the current code,
the state in ClusterTaskSetManager indicating which tasks are
pending may be updated after revive offers is called (there's a
race condition here), so when revive offers is called, the task set
manager does not yet realize that there are failed tasks that need
to be relaunched.
Don't ignore spark.cores.max when using Mesos Coarse mode
totalCoresAcquired is decremented but never incremented, causing Spark to effectively ignore spark.cores.max in coarse grained Mesos mode.
Migrate the daemon thread started by DAGScheduler to Akka actor
`DAGScheduler` adopts an event queue and a daemon thread polling the it to process events sent to a `DAGScheduler`. This is a classical actor use case. By migrating this thread to Akka actor, we may benefit from both cleaner code and better performance (context switching cost of Akka actor is much less than that of a native thread).
But things become a little complicated when taking existing test code into consideration.
Code in `DAGSchedulerSuite` is somewhat tightly coupled with `DAGScheduler`, and directly calls `DAGScheduler.processEvent` instead of posting event messages to `DAGScheduler`. To minimize code change, I chose to let the actor to delegate messages to `processEvent`. Maybe this doesn't follow conventional actor usage, but I tried to make it apparently correct.
Another tricky part is that, since `DAGScheduler` depends on the `ActorSystem` provided by its field `env`, `env` cannot be null. But the `dagScheduler` field created in `DAGSchedulerSuite.before` was given a null `env`. What's more, `BlockManager.blockIdsToBlockManagers` checks whether `env` is null to determine whether to run the production code or the test code (bad smell here, huh?). I went through all callers of `BlockManager.blockIdsToBlockManagers`, and made sure that if `env != null` holds, then `blockManagerMaster == null` must also hold. That's the logic behind `BlockManager.scala` [line 896](https://github.com/liancheng/incubator-spark/compare/dagscheduler-actor-refine?expand=1#diff-2b643ea78c1add0381754b1f47eec132L896).
At last, since `DAGScheduler` instances are always `start()`ed after creation, I removed the `start()` method, and starts the `eventProcessActor` within the constructor.
spark-assembly.jar fails to authenticate with YARN ResourceManager
The META-INF/services/ sbt MergeStrategy was discarding support for Kerberos, among others. This pull request changes to a merge strategy similar to sbt-assembly's default. I've also included an update to sbt-assembly 0.9.2, a minor fix to it's zip file handling.
Allow spark on yarn to be run from HDFS.
Allows the spark.jar, app.jar, and log4j.properties to be put into hdfs. Allows you to specify the files on a different hdfs cluster and it will copy them over. It makes sure permissions are correct and makes sure to put things into public distributed cache so they can be reused amongst users if their permissions are appropriate. Also add a bit of error handling for missing arguments.
Enable stopping and starting a spot cluster
Clusters launched using `--spot-price` contain an on-demand master and spot slaves. Because EC2 does not support stopping spot instances, the spark-ec2 script previously could only destroy such clusters.
This pull request makes it possible to stop and restart a spot cluster.
* The `stop` command works as expected for a spot cluster: the master is stopped and the slaves are terminated.
* To start a stopped spot cluster, the user must invoke `launch --use-existing-master`. This launches fresh spot slaves but resumes the existing master.
