[TEZ-3478] Cleanup fetcher data for failing task attempts (Unordered fetcher)

• Type: Improvement
• Status: Patch Available
• Priority: Minor
• Resolution: Unresolved
• Assignee: Rajesh Balamohan
• Reporter: Rajesh Balamohan
• Created: 18/Oct/16 22:54
• Updated: 28/Dec/16 23:45

Description

Env: 3 node AWS cluster with entire dataset in S3. Since data is in S3, it does have not additional storage for HDFS (uses existing space available in VMs). tez version is 0.7.

With some workloads (e.g q29 in tpcds), unordered fetchers download data in parallel for different vertices and runs out of disk space. However, downloaded
data related to these failed task attempts are not cleared. So subsequent task attempts also encounter similar situation and fails with "No space" exception. e.g stack trace

, errorMessage=Fetch failed:org.apache.hadoop.fs.FSError: java.io.IOException: No space left on device
        at org.apache.hadoop.fs.RawLocalFileSystem$LocalFSFileOutputStream.write(RawLocalFileSystem.java:261)
        at java.io.BufferedOutputStream.flushBuffer(BufferedOutputStream.java:82)
        at java.io.BufferedOutputStream.write(BufferedOutputStream.java:126)
        at org.apache.hadoop.fs.FSDataOutputStream$PositionCache.write(FSDataOutputStream.java:58)
        at java.io.DataOutputStream.write(DataOutputStream.java:107)
        at org.apache.hadoop.fs.ChecksumFileSystem$ChecksumFSOutputSummer.writeChunk(ChecksumFileSystem.java:426)
        at org.apache.hadoop.fs.FSOutputSummer.writeChecksumChunks(FSOutputSummer.java:206)
        at org.apache.hadoop.fs.FSOutputSummer.write1(FSOutputSummer.java:124)
        at org.apache.hadoop.fs.FSOutputSummer.write(FSOutputSummer.java:110)
        at org.apache.hadoop.fs.FSDataOutputStream$PositionCache.write(FSDataOutputStream.java:58)
        at java.io.DataOutputStream.write(DataOutputStream.java:107)
        at org.apache.tez.runtime.library.common.shuffle.ShuffleUtils.shuffleToDisk(ShuffleUtils.java:146)
        at org.apache.tez.runtime.library.common.shuffle.Fetcher.fetchInputs(Fetcher.java:771)
        at org.apache.tez.runtime.library.common.shuffle.Fetcher.doHttpFetch(Fetcher.java:497)
        at org.apache.tez.runtime.library.common.shuffle.Fetcher.doHttpFetch(Fetcher.java:396)
        at org.apache.tez.runtime.library.common.shuffle.Fetcher.callInternal(Fetcher.java:195)
        at org.apache.tez.runtime.library.common.shuffle.Fetcher.callInternal(Fetcher.java:70)
        at org.apache.tez.common.CallableWithNdc.call(CallableWithNdc.java:36)
        at java.util.concurrent.FutureTask.run(FutureTask.java:262)
        at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1145)
        at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:615)
        at java.lang.Thread.run(Thread.java:745)
Caused by: java.io.IOException: No space left on device
        at java.io.FileOutputStream.writeBytes(Native Method)
        at java.io.FileOutputStream.write(FileOutputStream.java:345)
        at org.apache.hadoop.fs.RawLocalFileSystem$LocalFSFileOutputStream.write(RawLocalFileSy

This would also affect any other job running in the cluster at the same time. It would be helpful to clean up the data downloaded for the failed task attempts.
Creating this ticket mainly for unordered fetcher case, though it could be similar case for ordered shuffle case as well.

e.g files

17M	/hadoopfs/fs1/yarn/nodemanager/usercache/cloudbreak/appcache/application_1476667862449_0043/attempt_1476667862449_0043_1_07_000028_0_10023_src_62_spill_-1.out
18M	/hadoopfs/fs1/yarn/nodemanager/usercache/cloudbreak/appcache/application_1476667862449_0043/attempt_1476667862449_0043_1_07_000028_0_10023_src_63_spill_-1.out
16M	/hadoopfs/fs1/yarn/nodemanager/usercache/cloudbreak/appcache/application_1476667862449_0043/attempt_1476667862449_0043_1_07_000028_0_10023_src_64_spill_-1.out
..
..

18M	/hadoopfs/fs1/yarn/nodemanager/usercache/cloudbreak/appcache/application_1476667862449_0043/attempt_1476667862449_0043_1_07_000028_2_10003_src_0_spill_-1.out
17M	/hadoopfs/fs1/yarn/nodemanager/usercache/cloudbreak/appcache/application_1476667862449_0043/attempt_1476667862449_0043_1_07_000028_2_10003_src_13_spill_-1.out
16M	/hadoopfs/fs1/yarn/nodemanager/usercache/cloudbreak/appcache/application_1476667862449_0043/attempt_1476667862449_0043_1_07_000028_2_10003_src_15_spill_-1.out
16M	/hadoopfs/fs1/yarn/nodemanager/usercache/cloudbreak/appcache/application_1476667862449_0043/attempt_1476667862449_0043_1_07_000028_2_10003_src_17_spill_-1.ou