[MAPREDUCE-4088] Task stuck in JobLocalizer prevented other tasks on the same node from committing

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 0.20.205.0
• Fix Version/s: 1.1.0
• Component/s: mrv1
• Target Version/s: 1.1.0
• Assignee: Ravi Prakash
• Reporter: Ravi Prakash
• Created: 30/Mar/12 16:15
• Updated: 30/Oct/12 20:55
• Resolved: 01/May/12 16:19

Description

We saw that as a result of HADOOP-6963, one task was stuck in this

Thread 23668: (state = IN_NATIVE)

• java.io.UnixFileSystem.getBooleanAttributes0(java.io.File) @bci=0 (Compiled frame; information may be imprecise)
• java.io.UnixFileSystem.getBooleanAttributes(java.io.File) @bci=2, line=228 (Compiled frame)
• java.io.File.exists() @bci=20, line=733 (Compiled frame)
• org.apache.hadoop.fs.FileUtil.getDU(java.io.File) @bci=3, line=446 (Compiled frame)
• org.apache.hadoop.fs.FileUtil.getDU(java.io.File) @bci=52, line=455 (Compiled frame)
• org.apache.hadoop.fs.FileUtil.getDU(java.io.File) @bci=52, line=455 (Compiled frame)
  ....
  .... TONS MORE OF THIS SAME LINE
• org.apache.hadoop.fs.FileUtil.getDU(java.io.File) @bci=52, line=455 (Compiled frame)
  .....
  .....
• org.apache.hadoop.fs.FileUtil.getDU(java.io.File) @bci=52, line=455 (Compiled frame)
• org.apache.hadoop.fs.FileUtil.getDU(java.io.File) @bci=52, line=455 (Interpreted frame)
  ne=451 (Interpreted frame)
• org.apache.hadoop.mapred.JobLocalizer.downloadPrivateCacheObjects(org.apache.hadoop.conf.Configuration, java.net.URI[], org.apache.hadoop.fs.Path[], long[], boolean[], boolean) @bci=150, line=324 (Interpreted frame)
• org.apache.hadoop.mapred.JobLocalizer.downloadPrivateCache(org.apache.hadoop.conf.Configuration) @bci=40, line=349 (Interpreted frame) 51, line=383 (Interpreted frame)
• org.apache.hadoop.mapred.JobLocalizer.runSetup(java.lang.String, java.lang.String, org.apache.hadoop.fs.Path, org.apache.hadoop.mapred.TaskUmbilicalProtocol) @bci=46, line=477 (Interpreted frame)
• org.apache.hadoop.mapred.JobLocalizer$3.run() @bci=20, line=534 (Interpreted frame)
• org.apache.hadoop.mapred.JobLocalizer$3.run() @bci=1, line=531 (Interpreted frame)
• java.security.AccessController.doPrivileged(java.security.PrivilegedExceptionAction, java.security.AccessControlContext) @bci=0 (Interpreted frame)
• javax.security.auth.Subject.doAs(javax.security.auth.Subject, java.security.PrivilegedExceptionAction) @bci=42, line=396 (Interpreted frame)
• org.apache.hadoop.security.UserGroupInformation.doAs(java.security.PrivilegedExceptionAction) @bci=14, line=1082 (Interpreted frame)
• org.apache.hadoop.mapred.JobLocalizer.main(java.lang.String[]) @bci=266, line=530 (Interpreted frame)

While all other tasks on the same node were stuck in
Thread 32141: (state = BLOCKED)

• java.lang.Thread.sleep(long) @bci=0 (Interpreted frame)
• org.apache.hadoop.mapred.Task.commit(org.apache.hadoop.mapred.TaskUmbilicalProtocol, org.apache.hadoop.mapred.Task$TaskReporter, org.apache.hadoop.mapreduce.OutputCommitter) @bci=24, line=980 (Compiled frame)
• org.apache.hadoop.mapred.Task.done(org.apache.hadoop.mapred.TaskUmbilicalProtocol, org.apache.hadoop.mapred.Task$TaskReporter) @bci=146, line=871 (Interpreted frame)
• org.apache.hadoop.mapred.ReduceTask.run(org.apache.hadoop.mapred.JobConf, org.apache.hadoop.mapred.TaskUmbilicalProtocol) @bci=470, line=423 (Interpreted frame)
• org.apache.hadoop.mapred.Child$4.run() @bci=29, line=255 (Interpreted frame)
• java.security.AccessController.doPrivileged(java.security.PrivilegedExceptionAction, java.security.AccessControlContext) @bci=0 (Interpreted frame)
• javax.security.auth.Subject.doAs(javax.security.auth.Subject, java.security.PrivilegedExceptionAction) @bci=42, line=396 (Interpreted frame)
• org.apache.hadoop.security.UserGroupInformation.doAs(java.security.PrivilegedExceptionAction) @bci=14, line=1082 (Interpreted frame)
• org.apache.hadoop.mapred.Child.main(java.lang.String[]) @bci=738, line=249 (Interpreted frame)

This should never happen. A stuck task should never prevent other tasks from different jobs on the same node from committing.

Issue Links

• breaks: MAPREDUCE-4749 Killing multiple attempts of a task taker longer as more attempts are killed
• relates to: MAPREDUCE-2364 Shouldn't hold lock on rjob while localizing resources.