[HADOOP-4614] "Too many open files" error while processing a large gzip file

• Type: Bug
• Status: Closed
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: 0.18.2
• Fix Version/s: 0.18.3
• Hadoop Flags: Reviewed
• Assignee: Yuri Pradkin
• Reporter: Abdul Qadeer
• Created: 07/Nov/08 08:56
• Updated: 08/Jul/09 16:53
• Resolved: 26/Nov/08 01:42

Description

I am running a simple word count program on a gzip compressed data of size 4 GB (Uncompressed size is about 7 GB). I have setup of 17 nodes in my Hadoop cluster. After some time, I get the following exception:

java.io.FileNotFoundException: /usr/local/hadoop/hadoop-hadoop/mapred/local/taskTracker/jobcache/job_200811041109_0003/attempt_200811041109_0003_m_000000_0/output/spill4055.out.index
(Too many open files)
at java.io.FileInputStream.open(Native Method)
at java.io.FileInputStream.(FileInputStream.java:137)
at org.apache.hadoop.fs.RawLocalFileSystem$TrackingFileInputStream.(RawLocalFileSystem.java:62)
at org.apache.hadoop.fs.RawLocalFileSystem$LocalFSFileInputStream.(RawLocalFileSystem.java:98)
at org.apache.hadoop.fs.RawLocalFileSystem.open(RawLocalFileSystem.java:168)
at org.apache.hadoop.fs.FileSystem.open(FileSystem.java:359)
at org.apache.hadoop.mapred.IndexRecord.readIndexFile(IndexRecord.java:47)
at org.apache.hadoop.mapred.MapTask$MapOutputBuffer.getIndexInformation(MapTask.java:1339)
at org.apache.hadoop.mapred.MapTask$MapOutputBuffer.mergeParts(MapTask.java:1237)
at org.apache.hadoop.mapred.MapTask$MapOutputBuffer.flush(MapTask.java:857)
at org.apache.hadoop.mapred.MapTask.run(MapTask.java:333)
at org.apache.hadoop.mapred.Child.main(Child.java:155)

From a user's perspective I know that Hadoop will use only one mapper for a gzipped file. The above exception suggests that probably Hadoop puts the intermediate data into many files. But the question is that "exactly how many open files are required in the worst case for any data size and cluster size?" Currently it looks as if Hadoop needs more number of open files as the size of input or the cluster size (in terms of nodes, mapper, reducers) increases. This is not plausible as far as scalability is concerned. A user needs to write some number in the /etc/security/limits.conf file that how many open files are allowed by hadoop node. The question is what that "magical number" should be?

So probably the best solution to this problem is to change Hadoop such a way that it can work with some moderate number of allowed open files (e.g. 4 K) or any other number should be suggested as an upper limit such that a user is sure that for any data size and cluster size, hadoop will not run into this "too many open files" issue.