[CASSANDRA-6169] Too many splits causes a "OutOfMemoryError: unable to create new native thread" in AbstractColumnFamilyInputFormat

• Type: Bug
• Status: Resolved
• Priority: Low
• Resolution: Fixed
• Fix Version/s: 1.2.11, 2.0.2
• Labels: hadoop

Environment

1.2.10
vnodes (server side)
Mac OS x (client)

• Severity: Low
• Since Version: 1.2.10
• Assignee: Patricio Echague
• Reporter: Patricio Echague

Authors

Patricio Echague

Reviewers

Jonathan Ellis

• Created: 09/Oct/13 00:28
• Updated: 16/Apr/19 09:32
• Resolved: 09/Oct/13 18:36

Description

The problem is caused by having 2300+ tokens due to vnodes.

In the client side I get this exception

Exception in thread "main" java.lang.OutOfMemoryError: unable to create new native thread
	at java.lang.Thread.start0(Native Method)
	at java.lang.Thread.start(Thread.java:691)
	at java.util.concurrent.ThreadPoolExecutor.addWorker(ThreadPoolExecutor.java:943)
	at java.util.concurrent.ThreadPoolExecutor.execute(ThreadPoolExecutor.java:1336)
	at java.util.concurrent.AbstractExecutorService.submit(AbstractExecutorService.java:132)
	at org.apache.cassandra.hadoop.AbstractColumnFamilyInputFormat.getSplits(AbstractColumnFamilyInputFormat.java:187)
	at org.apache.hadoop.mapred.JobClient.writeNewSplits(JobClient.java:1054)
	at org.apache.hadoop.mapred.JobClient.writeSplits(JobClient.java:1071)
	at org.apache.hadoop.mapred.JobClient.access$700(JobClient.java:179)
	at org.apache.hadoop.mapred.JobClient$2.run(JobClient.java:983)
	at org.apache.hadoop.mapred.JobClient$2.run(JobClient.java:936)
	at java.security.AccessController.doPrivileged(Native Method)
	at javax.security.auth.Subject.doAs(Subject.java:415)
	at org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1190)
	at org.apache.hadoop.mapred.JobClient.submitJobInternal(JobClient.java:936)
	at org.apache.hadoop.mapreduce.Job.submit(Job.java:550)
	at org.apache.hadoop.mapreduce.Job.waitForCompletion(Job.java:580)
	at com.relateiq.hadoop.cassandra.etl.CassandraETLJob.run(CassandraETLJob.java:58)
	at org.apache.hadoop.util.ToolRunner.run(ToolRunner.java:65)
	at com.relateiq.hadoop.cassandra.etl.CassandraETLJob.main(CassandraETLJob.java:149)

The problem seem to be in AbstractColumnFamilyInputFormat line ~180 which has an unbounded upper limit (actually it is Integer.MAX_INT)

ExecutorService executor = Executors.newCachedThreadPool();

Followed by:

            for (TokenRange range : masterRangeNodes)
            {
                if (jobRange == null)
                {
                    // for each range, pick a live owner and ask it to compute bite-sized splits
                    splitfutures.add(executor.submit(new SplitCallable(range, conf)));
                }
                else
                .....

which gets called one time per token and creates one thread just as many times.

The easy fix unless there is a longer term fix I'm unaware of would be to set an upper limit to the thread pool.

Something like this:

ExecutorService executor = new ThreadPoolExecutor(0, ConfigHelper.getMaxConcurrentSplitsResolution(), 60L, TimeUnit.SECONDS, new LinkedBlockingQueue<Runnable>());

Shall I proceed with a patch ?